Showing 9 of 1610 Publications in Antitrust

ICLE Comments to NTIA on Dual-Use Foundation AI Models with Widely Available Model Weights

Regulatory Comments I. Introduction We thank the National Telecommunications and Information Administration (NTIA) for the opportunity to contribute to this request for comments (RFC) in the “Dual . . .

I. Introduction

We thank the National Telecommunications and Information Administration (NTIA) for the opportunity to contribute to this request for comments (RFC) in the “Dual Use Foundation Artificial Intelligence Models with Widely Available Model Weights” proceeding. In these comments, we endeavor to offer recommendations to foster the innovative and responsible production of artificial intelligence (AI), encompassing both open-source and proprietary models. Our comments are guided by a belief in the transformative potential of AI, while recognizing NTIA’s critical role in guiding the development of regulations that not only protect consumers but also enable this dynamic field to flourish. The agency should seek to champion a balanced and forward-looking approach toward AI technologies that allows them to evolve in ways that maximize their social benefits, while navigating the complexities and challenges inherent in their deployment.

NTIA’s question “How should [the] potentially competing interests of innovation, competition, and security be addressed or balanced?”[1] gets to the heart of ongoing debates about AI regulation. There is no panacea to be discovered, as all regulatory choices require balancing tradeoffs. It is crucial to bear this in mind when evaluating, e.g., regulatory proposals that implicitly treat AI as inherently dangerous and regard as obvious that stringent regulation is the only effective strategy to mitigate such risks.[2] Such presumptions discount AI’s unknown but potentially enormous capacity to produce innovation, and inadequately account for other tradeoffs inherent to imposing a risk-based framework (e.g., requiring disclosure of trade secrets or particular kinds of transparency that could yield new cybersecurity attack vectors). Adopting an overly cautious stance risks not only stifling AI’s evolution, but may also preclude a fulsome exploration of its potential to foster social, economic, and technological advancement. A more restrictive regulatory environment may also render AI technologies more homogenous and smother development of the kinds of diverse AI applications needed to foster robust competition and innovation.

We observe this problematic framing in the executive order (EO) that serves as the provenance of this RFC.[3] The EO repeatedly proclaims the importance of “[t]he responsible development and use of AI” in order to “mitigate[e] its substantial risks.”[4] Specifically, the order highlights concerns over “dual-use foundation models”—i.e., AI systems that, while beneficial, could pose serious risks to national security, national economic security, national public health, or public safety.[5] Concerningly, one of the categories the EO flags as illicit “dual use” are systems “permitting the evasion of human control or oversight through means of deception or obfuscation.”[6] This open-ended category could be interpreted so broadly that essentially any general-purpose generative-AI system would classify.

The EO also repeatedly distinguishes “open” versus “closed” approaches to AI development, while calling for “responsible” innovation and competition.[7] On our reading, the emphasis the EO places on this distinction raises alarm bells about the administration’s inclination to stifle innovation through overly prescriptive regulatory frameworks, diminishment of the intellectual property rights that offer incentives for innovation, and regulatory capture that favors incumbents over new entrants. In favoring one model of AI development over another, the EO’s prescriptions could inadvertently hamper the dynamic competitive processes that are crucial both for technological progress and for the discovery of solutions to the challenges that AI technology poses.

Given the inchoate nature of AI technology—much less the uncertain markets in which that technology will ultimately be deployed and commercialized—NTIA has an important role to play in elucidating for policymakers the nuances that might lead innovators to choose an open or closed development model, without presuming that one model is inherently better than the other—or that either is necessarily “dangerous.” Ultimately, the preponderance of AI risks will almost certainly emerge idiosyncratically. It will be incumbent on policymakers to address such risks in an iterative fashion as they become apparent. For now, it is critical to resist the urge to enshrine crude and blunt categories for the heterogeneous suite of technologies currently gathered under the broad banner of  “AI.”

Section II of these comments highlights the importance of grounding AI regulation in actual harms, rather than speculative risks, while outlining the diversity of existing AI technologies and the need for tailored approaches. Section III starts with discussion of some of the benefits and challenges posed by both open and closed approaches to AI development, while cautioning against overly prescriptive definitions of “openness” and advocating flexibility in regulatory frameworks. It proceeds to examine the EO’s prescription to regulate so-called “dual-use” foundation models, underscoring some potential unintended consequences for open-source AI development and international collaboration. Section IV offers some principles to craft an effective regulatory model for AI, including distinguishing between low-risk and high-risk applications, avoiding static regulatory approaches, and adopting adaptive mechanisms like regulatory sandboxes and iterative rulemaking. Section V concludes.

II. Risk Versus Harm in AI Regulation

In many of the debates surrounding AI regulation, disproportionate focus is placed on the need to mitigate risks, without sufficient consideration of the immense benefits that AI technologies could yield. Moreover, because these putative risks remain largely hypothetical, proposals to regulate AI descend quickly into an exercise in shadowboxing.

Indeed, there is no single coherent definition of what even constitutes “AI.” The term encompasses a wide array of technologies, methodologies, and applications, each with distinct characteristics, capabilities, and implications for society. From foundational models that can generate human-like text, to algorithms capable of diagnosing diseases with greater accuracy than human doctors, to “simple” algorithms that facilitate a more tailored online experience, AI applications and their underlying technologies are as varied as they are transformative.

This diversity has profound implications for the regulation and development of AI. Very different regulatory considerations are relevant to AI systems designed for autonomous vehicles than for those used in financial algorithms or creative-content generation. Each application domain comes with its own set of risks, benefits, ethical dilemmas, and potential social impacts, necessitating tailored approaches to each use case. And none of these properties of AI map clearly onto the “open” and “closed” designations highlighted by the EO and this RFC. This counsels for focus on specific domains and specific harms, rather than how such technologies are developed.[8]

As in prior episodes of fast-evolving technologies, what is considered cutting-edge AI today may be obsolete tomorrow. This rapid pace of innovation further complicates the task of crafting policies and regulations that will be both effective and enduring. Policymakers and regulators must navigate this terrain with a nuanced understanding of AI’s multifaceted nature, including by embracing flexible and adaptive regulatory frameworks that can accommodate AI’s continuing evolution.[9] A one-size-fits-all approach could inadvertently stifle innovation or entrench the dominance of a few large players by imposing barriers that disproportionately affect smaller entities or emerging technologies.

Experts in law and economics have long scrutinized both market conduct and regulatory rent seeking that serve to enhance or consolidate market power by disadvantaging competitors, particularly through increasing the costs incurred by rivals.[10] Various tactics may be employed to undermine competitors or exclude them from the market that do not involve direct price competition. It is widely recognized that “engaging with legislative bodies or regulatory authorities to enact regulations that negatively impact competitors” produces analogous outcomes.[11] It is therefore critical that the emerging markets for AI technologies not engender opportunities for firms to acquire regulatory leverage over rivals. Instead, recognizing the plurality of AI technologies and encouraging a multitude of approaches to AI development could help to cultivate a more vibrant and competitive ecosystem, driving technological progress forward and maximizing AI’s potential social benefits.

This overarching approach counsels skepticism about risk-based regulatory frameworks that fail to acknowledge how the theoretical harms of one type of AI system may be entirely different from those of another. Obviously, the regulation of autonomous drones is a very different sort of problem than the regulation of predictive policing or automated homework tutors. Even within a single circumscribed domain of generative AI—such as “smart chatbots” like ChatGPT or Claude—different applications may present entirely different kinds of challenges. A highly purpose-built version of such a system might be employed by government researchers to develop new materiel for the U.S. Armed Forces, while a general-purpose commercial chatbot would employ layers of protection to ensure that ordinary users couldn’t learn how to make advanced weaponry. Rather treating “chatbots” as possible vectors for weapons development, a more appropriate focus would target high-capability systems designed to assist in developing such systems. Were it the case that a general-purpose chatbot inadvertently revealed some information on building weapons, all incentives would direct that AI’s creators to treat that as a bug to fix, not a feature to expand.

Take, for example, the recent public response to the much less problematic AI-system malfunctions that accompanied Google’s release of its Gemini program.[12] Gemini was found to generate historically inaccurate images, such as ethnically diverse U.S. senators from the 1800s, including women.[13] Google quickly acknowledged that it did not intend for Gemini to create inaccurate historical images and turned off the image-generation feature to allow time for the company to work on significant improvements before re-enabling it.[14] While Google blundered in its initial release, it had every incentive to discover and remedy the problem. The market response provided further incentive for Google to get it right in the future.[15] Placing the development of such systems under regulatory scrutiny because some users might be able to jailbreak a model and generate some undesirable material would create disincentives to the production of AI systems more generally, with little gained in terms of public safety.

Rather than focus on the speculative risks of AI, it is essential to ground regulation in the need to address tangible harms that stem from the observed impacts of AI technologies on society. Moreover, focusing on realistic harms would facilitate a more dynamic and responsive regulatory approach. As AI technologies evolve and new applications emerge, so too will the  potential harms. A regulatory framework that prioritizes actual harms can adapt more readily to these changes, enabling regulators to update or modify policies in response to new evidence or social impacts. This flexibility is particularly important for a field like AI, where technological advancements could quickly outpace regulation, creating gaps in oversight that may leave individuals and communities vulnerable to harm.

Furthermore, like any other body of regulatory law, AI regulation must be grounded in empirical evidence and data-driven decision making. Demanding a solid evidentiary basis as a threshold for intervention would help policymakers to avoid the pitfalls of reacting to sensationalized or unfounded AI fears. This would not only enhance regulators’ credibility with stakeholders, but would also ensure that resources are dedicated to addressing the most pressing and substantial issues arising from the development of AI.

III. The Regulation of Foundation Models

NTIA is right to highlight the tremendous promise that attends the open development of AI technologies:

Dual use foundation models with widely available weights (referred to here as open foundation models) could play a key role in fostering growth among less resourced actors, helping to widely share access to AI’s benefits…. Open foundation models can be readily adapted and fine-tuned to specific tasks and possibly make it easier for system developers to scrutinize the role foundation models play in larger AI systems, which is important for rights- and safety-impacting AI systems (e.g. healthcare, education, housing, criminal justice, online platforms etc.)

…Historically, widely available programming libraries have given researchers the ability to simultaneously run and understand algorithms created by other programmers. Researchers and journals have supported the movement towards open science, which includes sharing research artifacts like the data and code required to reproduce results.[16]

The RFC proceeds to seek input on how to define “open” and “widely available.”[17] These, however, are the wrong questions. NTIA should instead proceed from the assumption that there are no harms inherent to either “open” or “closed” development models; it should be seeking input on anything that might give rise to discrete harms in either open or closed systems.

NTIA can play a valuable role by recommending useful alterations to existing law where gaps currently exist, regardless of the business or distribution model employed by the AI developer. In short, there is nothing necessarily more or less harmful about adopting an “open” or a “closed” approach to software systems. The decision to pursue one path over the other will be made based on the relevant tradeoffs that particular firms face. Embedding such distinctions in regulation is arbitrary, at best, and counterproductive to the fruitful development of AI, at worst.

A. ‘Open’ or ‘Widely Available’ Model Weights

To the extent that NTIA is committed to drawing distinctions between “open” and “closed” approaches to developing foundation models, it should avoid overly prescriptive definitions of what constitutes “open” or “widely available” model weights that could significantly hamper the progress and utility of AI technologies.

Imposing narrow definitions risks creating artificial boundaries that fail to accurately reflect AI’s technical and operational realities. They could also inadvertently exclude or marginalize innovative AI models that fall outside those rigid parameters, despite their potential to contribute positively to technological advancement and social well-being. For instance, a definition of “open” that requires complete public accessibility without any form of control or restriction might discourage organizations from sharing their models, fearing misuse or loss of intellectual property.

Moreover, prescriptive definitions could stifle the organic growth and evolution of AI technologies. The AI field is characterized by its rapid pace of change, where today’s cutting-edge models may become tomorrow’s basic tools. Prescribing fixed criteria for what constitutes “openness” or “widely available” risks anchoring the regulatory landscape to this specific moment in time, leaving the regulatory framework less able to adapt to future developments and innovations.

Given AI developers’ vast array of applications, methodologies, and goals, it is imperative that any definitions of “open” or “widely available” model weights embrace flexibility. A flexible approach would acknowledge how the various stakeholders within the AI ecosystem have differing needs, resources, and objectives, from individual developers and academic researchers to startups and large enterprises. A one-size-fits-all definition of “openness” would fail to accommodate this diversity, potentially privileging certain forms of innovation over others and skewing the development of AI technologies in ways that may not align with broader social needs.

Moreover, flexibility in defining “open” and “widely available” must allow for nuanced understandings of accessibility and control. There can, for example, be legitimate reasons to limit openness, such as protecting sensitive data, ensuring security, and respecting intellectual-property rights, while still promoting a culture of collaboration and knowledge sharing. A flexible regulatory approach would seek a balanced ecosystem where the benefits of open AI models are maximized, and potential risks are managed effectively.

B. The Benefits of ‘Open’ vs ‘Closed’ Business Models

NTIA asks:

What benefits do open model weights offer for competition and innovation, both in the AI marketplace and in other areas of the economy? In what ways can open dual-use foundation models enable or enhance scientific research, as well as education/training in computer science and related fields?[18]

An open approach to AI development has obvious benefits, as NTIA has itself acknowledged in other contexts.[19] Open-foundation AI models represent a transformative force, characterized by their accessibility, adaptability, and potential for widespread application across various sectors. The openness of these models may serve to foster an environment conducive to innovation, wherein developers, researchers, and entrepreneurs can build on existing technologies to create novel solutions tailored to diverse needs and challenges.

The inherent flexibility of open-foundation models can also catalyze a competitive market, encouraging a healthy ecosystem where entities ranging from startups to established corporations may all participate on roughly equal footing. By lowering some entry barriers related to access to basic AI technologies, this competitive environment can further drive technological advancements and price efficiencies, ultimately benefiting consumers and society at-large.

But more “closed” approaches can also prove very valuable. As NTIA notes in this RFC, it is rarely the case that a firm pursues a purely open or closed approach. These terms exist along a continuum, and firms blend models as necessary.[20] And just as firms readily mix elements of open and closed business models, a regulator should be agnostic about the precise mix that firms employ, which ultimately must align with the realities of market dynamics and consumer preferences.

Both open and closed approaches offer distinct benefits and potential challenges. For instance, open approaches might excel in fostering a broad and diverse ecosystem of applications, thereby appealing to users and developers who value customization and variety. They can also facilitate a more rapid dissemination of innovation, as they typically impose fewer restrictions on the development and distribution of new applications. Conversely, closed approaches, with their curated ecosystems, often provide enhanced security, privacy, and a more streamlined user experience. This can be particularly attractive to users less inclined to navigate the complexities of open systems. Under the right conditions, closed systems can likewise foster a healthy ecosystem of complementary products.

The experience of modern digital platforms demonstrates that there is no universally optimal approach to structuring business activities, thus illustrating the tradeoffs inherent in choosing among open and closed business models. The optimal choice depends on the specific needs and preferences of the relevant market participants. As Jonathan M. Barnett has noted:

Open systems may yield no net social gain over closed systems, can pose a net social loss under certain circumstances, and . . . can impose a net social gain under yet other circumstances.[21]

Similar considerations apply in the realm of AI development. Closed or semi-closed ecosystems can offer such advantages as enhanced security and curated offerings, which may appeal to certain users and developers. These benefits, however, may come at the cost of potentially limited innovation, as a firm must rely on its own internal processes for research and development. Open models, on the other hand, while fostering greater collaboration and creativity, may also introduce risks related to quality control, intellectual-property protection, and a host of other concerns that may be better controlled in a closed business model. Even along innovation dimensions, closed platforms can in many cases outperform open models.

With respect to digital platforms like the App Store and Google Play Store, there is a “fundamental welfare tradeoff between two-sided proprietary…platforms and two-sided platforms which allow ‘free entry’ on both sides of the market.”[22] Consequently, “it is by no means obvious which type of platform will create higher product variety, consumer adoption and total social welfare.”[23]

To take another example, consider the persistently low adoption rates for consumer versions of the open-source Linux operating system, versus more popular alternatives like Windows or MacOS.[24] A closed model like Apple’s MacOS is able to outcompete open solutions by better leveraging network effects and developing a close relationship with end users.[25] Even in this example, adoption of open versus closed models varies across user types, with, e.g., developers showing a strong preference for Linux over Mac, and only a slight preference for Windows over Linux.[26] This underscores the point that the suitability of an open or closed model varies not only by firm and product, nor even solely by user, but by the unique fit of a particular model for a particular user in a particular context. Many of those Linux-using developers will likely not use it on their home computing device, for example, even if they prefer it for work.

The dynamics among consumers and developers further complicate prevailing preferences for open or closed models. For some users, the security and quality assurance provided by closed ecosystems outweigh the benefits of open systems’ flexibility. On the developer side, the lower barriers to entry in more controlled ecosystems that smooth the transaction costs associated with developing and marketing applications can democratize application development, potentially leading to greater innovation within those ecosystems. Moreover, distinctions between open and closed models can play a critical role in shaping inter-brand competition. A regulator placing its thumb on the business-model scale would push the relevant markets toward less choice and lower overall welfare.[27]

By differentiating themselves through a focus on ease-of-use, quality, security, and user experience, closed systems contribute to a vibrant competitive landscape where consumers have clear choices between differing “brands” of AI. Forcing an AI developer to adopt practices that align with a regulator’s preconceptions about the relative value of “open” and “closed” risks homogenizing the market and diminishing the very competition that spurs innovation and consumer choice.

Consider some of the practical benefits sought by deployers when choosing between open and closed models. For example, it’s not straightforward to say close is inherently better than open when considering issues of data sharing or security; even here, there are tradeoffs. Open innovation in AI—characterized by the sharing of data, algorithms, and methodologies within the research community and beyond—can mitigate many of the risks associated with model development. This openness fosters a culture of transparency and accountability, where AI models and their applications are subject to scrutiny by a broad community of experts, practitioners, and the general public. This collective oversight can help to identify and address potential safety and security concerns early in the development process, thus enhancing AI technologies’ overall trustworthiness.

By contrast, a closed system may implement and enforce standardized security protocols more quickly. A closed system may have a sharper, more centralized focus on providing data security to users, which may perform better along some dimensions. And while the availability of code may provide security in some contexts, in other circumstances, closed systems perform better.[28]

In considering ethical AI development, different types of firms should be free to experiment with different approaches, even blending them where appropriate. For example, Claude’s approach to “Collective Constitutional AI” adopts what is arguably a “semi-open” model, blending proprietary elements with certain aspects of openness to foster innovation, while also maintaining a level of control.[29] This model might strike an appropriate balance, in that it ensures some degree of proprietary innovation and competitive advantage while still benefiting from community feedback and collaboration.

On the other hand, fully open-source development could lead to a different, potentially superior result that meets a broader set of needs through community-driven evolution and iteration. There is no way to determine, ex ante, that either an open or a closed approach to AI development will inherently provide superior results for developing “ethical” AI. Each has its place, and, most likely, the optimal solutions will involve elements of both approaches.

In essence, codifying a regulatory preference for one business model over the other would oversimplify the intricate balance of tradeoffs inherent to platform ecosystems. Economic theory and empirical evidence suggest that both open and closed platforms can drive innovation, serve consumer interests, and stimulate healthy competition, with all of these considerations depending heavily on context. Regulators should therefore aim for flexible policies that support coexistence of diverse business models, fostering an environment where innovation can thrive across the continuum of openness.

C. Dual-Use Foundation Models and Transparency Requirements

The EO and the RFC both focus extensively on so-called “dual-use” foundation models:

Foundation models are typically defined as, “powerful models that can be fine-tuned and used for multiple purposes.” Under the Executive Order, a “dual-use foundation model” is “an AI model that is trained on broad data; generally uses self-supervision, contains at least tens of billions of parameters; is applicable across a wide range of contexts; and that exhibits, or could be easily modified to exhibit, high levels of performance at tasks that pose a serious risk to security, national economic security, national public health or safety, or any combination of those matters….”[30]

But this framing will likely do more harm than good. As noted above, the terms “AI” or “AI model” are frequently invoked to refer to very different types of systems. Further defining these models as “dual use” is also unhelpful, as virtually any tool in existence can be “dual use” in this sense. Certainly, from a certain perspective, all software—particularly highly automated software—can pose a serious risk to “national security” or “safety.” Encryption and other privacy-protecting tools certainly fit this definition.[31] While it is crucial to mitigate harms associated with the misuse of AI technologies, the blanket treatment of all foundation models under this category is overly simplistic.

The EO identifies certain clear risks, such as the possibility that models could aid in the creation of chemical, biological, or nuclear weaponry. These categories are obvious subjects for regulatory control, but the EO then appears to open a giant definitional loophole that threatens to subsume virtually any useful AI system. It employs expansive terminology to describe a more generalized threat—specifically, that dual-use models could “[permit] the evasion of human control or oversight through means of deception or obfuscation.”[32] Such language could encompass a wide array of general-purpose AI models. Furthermore, by labeling systems capable of bypassing human decision making as “dual use,” the order implicitly suggests that all AI could pose such risk as warrants national-security levels of scrutiny.

Given the EO’s broad definition of AI as “a machine-based system that can, for a given set of human-defined objectives, make predictions, recommendations, or decisions influencing real or virtual environments,” numerous software systems not typically even considered AI might be categorized as “dual-use” models.[33] Essentially, any sufficiently sophisticated statistical-analysis tool could qualify under this definition.

A significant repercussion of the EO’s very broad reporting mandates for dual-use systems, and one directly relevant to the RFC’s interest in promoting openness, is that these might chill open-source AI development.[34] Firms dabbling in AI technologies—many of which might not consider their projects to be dual use—might keep their initiatives secret until they are significantly advanced. Faced with the financial burden of adhering to the EO’s reporting obligations, companies that lack a sufficiently robust revenue model to cover both development costs and legal compliance might be motivated to dodge regulatory scrutiny in the initial phases, consequently dampening the prospects for transparency.

It is hard to imagine how open-source AI projects could survive in such an environment. Open-source AI code libraries like TensorFlow[35] and PyTorch[36] foster remarkable innovation by allowing developers to create new applications that use cutting-edge models. How could a paradigmatic startup developer working out of a garage genuinely commit to open-source development if tools like these fall under the EO’s jurisdiction? Restricting access to the weights that models use—let alone avoiding open-source development entirely—may hinder independent researchers’ ability to advance the forefront of AI technology.

Moreover, scientific endeavors typically benefit from the contributions of researchers worldwide, as collaborative efforts on a global scale are known to fast-track innovation. The pressure the EO applies to open-source development of AI tools could curtail international cooperation, thereby distancing American researchers from crucial insights and collaborations. For example, AI’s capacity to propel progress in numerous scientific areas is potentially vast—e.g., utilizing MRI images and deep learning for brain-tumor diagnoses[37] or employing machine learning to push the boundaries of materials science.[38] Such research does not benefit from stringent secrecy, but thrives on collaborative development. Enabling a broader community to contribute to and expand upon AI advancements supports this process.

Individuals respond to incentives. Just as how well-intentioned seatbelt laws paradoxically led to an uptick in risky driving behaviors,[39] ill-considered obligations placed on open-source AI developers could unintentionally stifle the exchange of innovative concepts crucial to maintain the United States’ leadership in AI innovation.

IV. Regulatory Models that Support Innovation While Managing Risks Effectively

In the rapidly evolving landscape of artificial intelligence (AI), it is paramount to establish governance and regulatory frameworks that both encourage innovation and ensure safety and ethical integrity. An effective regulatory model for AI should be adaptive, principles-based, and foster a collaborative environment among regulators, developers, researchers, and the broader community. A number of principles can help in developing this regime.

A. Low-Risk vs High-Risk AI

First, a clear distinction should be made between low-risk AI applications that enhance operational efficiency or consumer experience and high-risk applications that could have significant safety implications. Low-risk applications like search algorithms and chatbots should be governed by a set of baseline ethical guidelines and best practices that encourage innovation, while ensuring basic standards are met. On the other hand, high-risk applications—such as those used by law enforcement or the military—would require more stringent review processes, including impact assessments, ethical reviews, and ongoing monitoring to mitigate potentially adverse effects.

Contrast this with the recently enacted AI Act in the European Union, and its decision to create presumptions of risk for general purpose AI (GPAI) systems, such as large language models (LLMs), that present what the EU has termed so-called “systemic risk.”[40] Article 3(65) of the AI Act defines systemic risk as “a risk that is specific to the high-impact capabilities of general-purpose AI models, having a significant impact on the Union market due to their reach, or due to actual or reasonably foreseeable negative effects on public health, safety, public security, fundamental rights, or the society as a whole, that can be propagated at scale across the value chain.”[41]

This definition bears similarities to the “Hand formula” in U.S. tort law, which balances the burden of precautions against the probability and severity of potential harm to determine negligence.[42] The AI Act’s notion of systemic risk, however, is applied more broadly to entire categories of AI systems based on their theoretical potential for widespread harm, rather than on a case-by-case basis.

The designation of LLMs as posing “systemic risk” is problematic for several reasons. It creates a presumption of risk merely based on a GPAI system’s scale of operations, without any consideration of the actual likelihood or severity of harm in specific use cases. This could lead to unwarranted regulatory intervention and unintended consequences that hinder the development and deployment of beneficial AI technologies. And this broad definition of systemic risk gives regulators significant leeway to intervene in how firms develop and release their AI products, potentially blocking access to cutting-edge tools for European citizens, even in the absence of tangible harms.

While it is important to address potential risks associated with AI systems, the AI Act’s approach risks stifling innovation and hindering the development of beneficial AI technologies within the EU.

B. Avoid Static Regulatory Approaches

AI regulators are charged with overseeing a dynamic and rapidly developing market, and should therefore avoid erecting a rigid framework that force new innovations into ill-fitting categories. The “regulatory sandbox” may provide a better model to balance innovation with risk management. By allowing developers to test and refine AI technologies in a controlled environment under regulatory oversight, sandboxes can be used to help identify and address potential issues before wider deployment, all while facilitating dialogue between innovators and regulators. This approach not only accelerates the development of safe and ethical AI solutions, but also builds mutual understanding and trust. Where possible, NTIA should facilitate policy experimentation with regulatory sandboxes in the AI context.

Meta’s Open Loop program is an example of this kind of experimentation.[43] This program is a policy prototyping research project focused on evaluating the National Institute of Standards and Technology (NIST) AI Risk Management Framework (RMF) 1.0.[44] The goal is to assess whether the framework is understandable, applicable, and effective in assisting companies to identify and manage risks associated with generative AI. It also provides companies an opportunity to familiarize themselves with the NIST AI RMF and its application in risk-management processes for generative AI systems. Additionally, it aims to collect data on existing practices and offer feedback to NIST, potentially influencing future RMF updates.

1. Regulation as a discovery process

Another key principle is to ensure that regulatory mechanisms are adaptive. Some examples of adaptive mechanisms are iterative rulemaking and feedback loops that allow regulations to be updated continuously in response to new developments and insights. Such mechanisms enable policymakers to respond swiftly to technological breakthroughs, ensuring that regulations remain relevant and effective, without stifling innovation.

Geoffrey Manne & Gus Hurwitz have recently proposed a framework for “regulation as a discovery process” that could be adapted to AI.[45] They argue for a view of regulation not merely as a mechanism for enforcing rules, but as a process for discovering information that can inform and improve regulatory approaches over time. This perspective is particularly pertinent to AI, where the pace of innovation and the complexity of technologies often outstrip regulators’ understanding and ability to predict future developments. This framework:

in its simplest formulation, asks regulators to consider that they might be wrong. That they might be asking the wrong questions, collecting the wrong information, analyzing it the wrong way—or even that Congress has given them the wrong authority or misunderstood the problem that Congress has tasked them to address.[46]

That is to say, an adaptive approach to regulation requires epistemic humility, with the understanding that, particularly for complex, dynamic industries:

there is no amount of information collection or analysis that is guaranteed to be “enough.” As Coase said, the problem of social cost isn’t calculating what those costs are so that we can eliminate them, but ascertaining how much of those social costs society is willing to bear.[47]

In this sense, modern regulators’ core challenge is to develop processes that allow for iterative development of knowledge, which is always in short supply. This requires a shift in how an agency conceptualizes its mission, from one of writing regulations to one of assisting lawmakers to assemble, filter, and focus on the most relevant and pressing information needed to understand a regulatory subject’s changing dynamics.[48]

As Hurwitz & Manne note, existing efforts to position some agencies as information-gathering clearinghouses suffer from a number of shortcomings—most notably, that they tend to operate on an ad hoc basis, reporting to Congress in response to particular exigencies.[49] The key to developing a “discovery process” for AI regulation would instead require setting up ongoing mechanisms to gather and report on data, as well as directing the process toward “specifications for how information should be used, or what the regulator anticipated to find in the information, prior to its collection.”[50]

Embracing regulation as a discovery process means acknowledging the limits of our collective knowledge about AI’s potential risks and benefits. This underscores why regulators should prioritize generating and utilizing new information through regulatory experiments, iterative rulemaking, and feedback loops. A more adaptive regulatory framework could respond to new developments and insights in AI technologies, thereby ensuring that regulations remain relevant and effective, without stifling innovation.

Moreover, Hurwitz & Manne highlight the importance of considering regulation as an information-producing activity.[51] In AI regulation, this could involve setting up mechanisms that allow regulators, innovators, and the public to contribute to and benefit from a shared pool of knowledge about AI’s impacts. This could include public databases of AI incidents, standardized reporting of AI-system performance, or platforms for sharing best practices in AI safety and ethics.

Static regulatory approaches may fail to capture the evolving landscape of AI applications and their societal implications. Instead, a dynamic, information-centric regulatory strategy that embraces the market as a discovery process could better facilitate beneficial innovations, while identifying and mitigating harms.

V. Conclusion

As the NTIA navigates the complex landscape of AI regulation, it is imperative to adopt a nuanced, forward-looking approach that balances the need to foster innovation with the imperatives of ensuring public safety and ethical integrity. The rapid evolution of AI technologies necessitates a regulatory framework that is both adaptive and principles-based, eschewing static snapshots of the current state of the art in favor of flexible mechanisms that could accommodate the dynamic nature of this field.

Central to this approach is to recognize that the field of AI encompasses a diverse array of technologies, methodologies, and applications, each with its distinct characteristics, capabilities, and implications for society. A one-size-fits-all regulatory model would not only be ill-suited to the task at-hand, but would also risk stifling innovation and hindering the United States’ ability to maintain its leadership in the global AI industry. NTIA should focus instead on developing tailored approaches that distinguish between low-risk and high-risk applications, ensuring that regulatory interventions are commensurate with the potential identifiable harms and benefits associated with specific AI use cases.

Moreover, the NTIA must resist the temptation to rely on overly prescriptive definitions of “openness” or to favor particular business models over others. The coexistence of open and closed approaches to AI development is essential to foster a vibrant, competitive ecosystem that drives technological progress and maximizes social benefits. By embracing a flexible regulatory framework that allows for experimentation and iteration, the NTIA can create an environment conducive to innovation while still ensuring that appropriate safeguards are in place to mitigate potential risks.

Ultimately, the success of the U.S. AI industry will depend on the ability of regulators, developers, researchers, and the broader community to collaborate in developing governance frameworks that are both effective and adaptable. By recognizing the importance of open development and diverse business models, the NTIA can play a crucial role in shaping the future of AI in ways that promote innovation, protect public interests, and solidify the United States’ position as a global leader in this transformative field.

[1] Dual Use Foundation Artificial Intelligence Models With Widely Available Model Weights, Docket No. 240216-0052, 89 FR 14059, National Telecommunications and Information Administration (Mar. 27, 2024) at 14063, question 8(a) [hereinafter “RFC”].

[2] See, e.g., Kristian Stout, Systemic Risk and Copyright in the EU AI Act, Truth on the Market (Mar. 19, 2024), https://truthonthemarket.com/2024/03/19/systemic-risk-and-copyright-in-the-eu-ai-act.

[3] Exec. Order No. 14110, 88 F.R. 75191 (2023), https://www.federalregister.gov/documents/2023/11/01/2023-24283/safe-secure-and-trustworthy-development-and-use-of-artificial-intelligence?_fsi=C0CdBzzA [hereinafter “EO”].

[4] See, e.g., EO at §§ 1; 2(c), 5.2(e)(ii); and § 8(c);

[5] Id. at § 3(k).

[6] Id. at § (k)(iii).

[7] Id. at § 4.6. As NTIA notes, the administration refers to “widely available model weight,” which is equivalent to “open foundation models” in this proceeding. RFC at 14060.

[8] For more on the “open” vs “closed” distinction and its poor fit as a regulatory lens, see, infra, at nn. 19-41 and accompanying text.

[9] Adaptive regulatory frameworks are discussed, infra, at nn. 42-53 and accompanying text.

[10] See Steven C. Salop & David T. Scheffman, Raising Rivals’ Costs, 73:2 Am. Econ. R. 267, 267–71 (1983), http://www.jstor.org/stable/1816853.

[11] See Steven C. Salop & David T. Scheffman, Cost-Raising Strategies, 36:1 J. Indus. Econ. 19 (1987), https://doi.org/10.2307/2098594.

[12] Cindy Gordon, Google Pauses Gemini AI Model After Latest Debacle, Forbes (Feb. 29, 2024), https://www.forbes.com/sites/cindygordon/2024/02/29/google-latest-debacle-has-paused-gemini-ai-model/?sh=3114d093536c.

[13] Id.

[14] Id.

[15] Breck Dumas, Google Loses $96B in Value on Gemini Fallout as CEO Does Damage Control, Yahoo Finance (Feb. 28, 2024), https://finance.yahoo.com/news/google-loses-96b-value-gemini-233110640.html.

[16] RFC at 14060.

[17] RFC at 14062, question 1.

[18] RFC at 14062, question 3(a).

[19] Department of Commerce, Competition in the Mobile Application Ecosystem (2023), https://www.ntia.gov/report/2023/competition-mobile-app-ecosystem (“While retaining appropriate latitude for legitimate privacy, security, and safety measures, Congress should enact laws and relevant agencies should consider measures (such as rulemaking) designed to open up distribution of lawful apps, by prohibiting… barriers to the direct downloading of applications.”).

[20] RFC at 14061 (“‘openness’ or ‘wide availability’ of model weights are also terms without clear definition or consensus. There are gradients of ‘openness,’ ranging from fully ‘closed’ to fully ‘open’”).

[21] See Jonathan M. Barnett, The Host’s Dilemma: Strategic Forfeiture in Platform Markets for Informational Goods, 124 Harv. L. Rev. 1861, 1927 (2011).

[22] Id. at 2.

[23] Id. at 3.

[24]  Desktop Operating System Market Share Worldwide Feb 2023 – Feb 2024, statcounter, https://gs.statcounter.com/os-market-share/desktop/worldwide (last visited Mar. 27, 2024).

[25]  Andrei Hagiu, Proprietary vs. Open Two-Sided Platforms and Social Efficiency (Harv. Bus. Sch. Strategy Unit, Working Paper No. 09-113, 2006).

[26] Joey Sneddon, More Developers Use Linux than Mac, Report Shows, Omg Linux (Dec. 28, 2022), https://www.omglinux.com/devs-prefer-linux-to-mac-stackoverflow-survey.

[27] See Michael L. Katz & Carl Shapiro, Systems Competition and Network Effects, 8 J. Econ. Persp. 93, 110 (1994), (“[T]he primary cost of standardization is loss of variety: consumers have fewer differentiated products to pick from, especially if standardization prevents the development of promising but unique and incompatible new systems”).

[28] See. e.g., Nokia, Threat Intelligence Report 2020 (2020), https://www.nokia.com/networks/portfolio/cyber-security/threat-intelligence-report-2020; Randal C. Picker, Security Competition and App Stores, Network Law Review (Aug. 23, 2021), https://www.networklawreview.org/picker-app-stores.

[29] Collective Constitutional AI: Aligning a Language Model with Public Input, Anthropic (Oct. 17, 2023), https://www.anthropic.com/news/collective-constitutional-ai-aligning-a-language-model-with-public-input.

[30] RFC at 14061.

[31] Encryption and the “Going Dark” Debate, Congressional Research Service (2017), https://crsreports.congress.gov/product/pdf/R/R44481.

[32] EO at. § 3(k)(iii).

[33] EO at § 3(b).

[34] EO at § 4.2 (requiring companies developing dual-use foundation models to provide ongoing reports to the federal government on their activities, security measures, model weights, and red-team testing results).

[35] An End-to-End Platform for Machine Learning, TensorFlow, https://www.tensorflow.org (last visited Mar. 27, 2024).

[36] Learn the Basics, PyTorch, https://pytorch.org/tutorials/beginner/basics/intro.html (last visited Mar. 27, 2024).

[37] Akmalbek Bobomirzaevich Abdusalomov, Mukhriddin Mukhiddinov, & Taeg Keun Whangbo, Brain Tumor Detection Based on Deep Learning Approaches and Magnetic Resonance Imaging, 15(16) Cancers (Basel) 4172 (2023), available at https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10453020.

[38] Keith T. Butler, et al., Machine Learning for Molecular and Materials Science, 559 Nature 547 (2018), available at https://www.nature.com/articles/s41586-018-0337-2.

[39] The Peltzman Effect, The Decision Lab, https://thedecisionlab.com/reference-guide/psychology/the-peltzman-effect (last visited Mar. 27, 2024).

[40] European Parliament, European Parliament legislative Resolution of 13 March 2024 on the Proposal for a Regulation of the European Parliament and of the Council on Laying Down Harmonised Rules on Artificial Intelligence (Artificial Intelligence Act) and Amending Certain Union Legislative Acts, COM/2021/206, available at https://www.europarl.europa.eu/doceo/document/TA-9-2024-0138_EN.html [hereinafter “EU AI Act”].

[41] Id. at Art. 3(65).

[42] See Stephen G. Gilles, On Determining Negligence: Hand Formula Balancing, the Reasonable Person Standard, and the Jury, 54 Vanderbilt L. Rev. 813, 842-49 (2001).

[43] See Open Loop’s First Policy Prototyping Program in the United States, Meta, https://www.usprogram.openloop.org (last visited Mar. 27. 2024).

[44] Id.

[45] Justin (Gus) Hurwitz & Geoffrey A. Manne, Pigou’s Plumber: Regulation as a Discovery Process, SSRN (2024), available at https://laweconcenter.org/resources/pigous-plumber.

[46] Id. at 32.

[47] Id. at 33.

[48] See id. at 28-29

[49] Id. at 37.

[50] Id. at 37-38.

[51] Id.

Continue reading
Innovation & the New Economy

US v. Apple Lawsuit Has Big Implications for Competition and Innovation

TOTM The lawsuit filed yesterday by the U.S. Justice Department (DOJ) against Apple for monopolization of the U.S. smartphone market (joined by 15 states and the District of . . .

The lawsuit filed yesterday by the U.S. Justice Department (DOJ) against Apple for monopolization of the U.S. smartphone market (joined by 15 states and the District of Columbia) has big implications for American competition and innovation.

At the heart of the complaint is the DOJ’s assertion that…

Read the full piece here.

Continue reading
Antitrust & Consumer Protection

Mi Mercado Es Su Mercado: The Flawed Competition Analysis of Mexico’s COFECE

Popular Media Mexico’s Federal Economic Competition Commission (COFECE, after its Spanish acronym) has published the preliminary report it prepared following its investigation of competition in the retail . . .

Mexico’s Federal Economic Competition Commission (COFECE, after its Spanish acronym) has published the preliminary report it prepared following its investigation of competition in the retail electronic-commerce market (e.g., Amazon). The report finds that: 

Read the full piece here.

Continue reading
Antitrust & Consumer Protection

Bill C-59 and the Use of Structural Merger Presumptions in Canada

Regulatory Comments We, the undersigned, are scholars from the International Center for Law & Economics (ICLE) with experience in the academy, enforcement agencies, and private practice in . . .

We, the undersigned, are scholars from the International Center for Law & Economics (ICLE) with experience in the academy, enforcement agencies, and private practice in competition law. We write to address a key aspect of proposed amendments to Canadian competition law. Specifically, we focus on clauses in Bill C-59 pertinent to mergers and acquisitions and, in particular, the Bureau of Competition’s recommendation that the Bill should:

Amend Clauses 249-250 to enact rebuttable presumptions for mergers consistent with those set out in the U.S. Merger Guidelines.[1]

The Bureau’s recommendation seeks to codify in Canadian competition law the structural presumptions outlined in the 2023 U.S. Federal Trade Commission (FTC) and U.S. Justice Department (DOJ) Merger Guidelines.  On balance, however, adoption of that recommendation would impede, rather than promote, fair competition and the welfare of Canadian consumers.

The cornerstone of the proposed change lies in the introduction of rebuttable presumptions of illegality for mergers that exceed specified market-share or concentration thresholds. While this approach may seem intuitive, the economic literature and U.S. enforcement experience militate against its adoption in Canadian law.

The goal of enhancing—indeed, strengthening—Canadian competition law should not be conflated with the adoption of foreign regulatory guidelines. The most recent U.S. Merger Guidelines establish new structural thresholds, based primarily on the Herfindahl-Hirschman Index (HHI) and market share, to establish presumptions of anticompetitive effects and illegality. Those structural presumptions, adopted a few short months ago, are inconsistent with established economic literature and are untested in U.S. courts. Those U.S. guidelines should not be codified in Canadian law without robust deliberation to ensure alignment with Canadian legal principles, on the one hand, and with economic realities and evidence, on the other.

Three points are especially important. First, concentration measures are widely considered to be a poor proxy for the level of competition that prevails in a given market. Second, lower merger thresholds may lead to enforcement errors that discourage investment and entrepreneurial activity and allocate enforcement resources to the wrong cases. Finally, these risks are particularly acute when concentration thresholds are used not as useful indicators but, instead, as actual legal presumptions (albeit rebuttable ones). We discuss each of these points in more detail below.

What Concentration Measures Can and Cannot Tell Us About Competition

While the use of concentration measures and thresholds can provide a useful preliminary-screening mechanism to identify potentially problematic mergers, substantially lowering the thresholds to establish a presumption of illegality is inadvisable for several reasons.

First, too strong a reliance on concentration measures lacks economic foundation and is likely prone to frequent error. Economists have been studying the relationship between concentration and various potential indicia of anticompetitive effects—price, markup, profits, rate of return, etc.—for decades.[2] There are hundreds of empirical studies addressing this topic.[3]

The assumption that “too much” concentration is harmful assumes both that the structure of a market is what determines economic outcomes and that anyone could know what the “right” amount of concentration is. But as economists have understood since at least the 1970s (and despite an extremely vigorous, but futile, effort to show otherwise), market structure does not determine outcomes.[4]

This skepticism toward concentration measures as a guide for policy is well-supported, and is held by scholars across the political spectrum.  To take one prominent, recent example, professors Fiona Scott Morton (deputy assistant U.S. attorney general for economics in the DOJ Antitrust Division under President Barack Obama, now at Yale University); Martin Gaynor (former director of the FTC Bureau of Economics under President Obama, now serving as special advisor to Assistant U.S. Attorney General Jonathan Kanter, on leave from Carnegie Mellon University), and Steven Berry (an industrial-organization economist at Yale University) surveyed the industrial-organization literature and found that presumptions based on measures of concentration are unlikely to provide sound guidance for public policy:

In short, there is no well-defined “causal effect of concentration on price,” but rather a set of hypotheses that can explain observed correlations of the joint outcomes of price, measured markups, market share, and concentration.…

Our own view, based on the well-established mainstream wisdom in the field of industrial organization for several decades, is that regressions of market outcomes on measures of industry structure like the Herfindahl-Hirschman Index should be given little weight in policy debates.[5]

As Chad Syverson recently summarized:

Perhaps the deepest conceptual problem with concentration as a measure of market power is that it is an outcome, not an immutable core determinant of how competitive an industry or market is… As a result, concentration is worse than just a noisy barometer of market power. Instead, we cannot even generally know which way the barometer is oriented.[6]

This does not mean that concentration measures have no use in merger screening. Rather, market concentration is often unrelated to antitrust-enforcement goals because it is driven by factors that are endogenous to each industry. Enforcers should not rely too heavily on structural presumptions based on concentration measures, as these may be poor indicators of the instances in which antitrust enforcement is most beneficial to competition and consumers.

At What Level Should Thresholds Be Set?

Second, if concentration measures are to be used in some fashion, at what level or levels should they be set?

The U.S. 2010 Horizontal Merger Guidelines were “b?ased on updated HHI thresholds that more accurately reflect actual enforcement practice.”[7] These numbers were updated in 2023, but without clear justification. While the U.S. enforcement authorities cite several old cases (cases that implicated considerably higher levels of concentration than those in their 2023 guidelines), we agree with comments submitted in 2022 by now-FTC Bureau of Economics Director Aviv Nevo and colleagues, who argued against such a change. They wrote:

Our view is that this would not be the most productive route for the agencies to pursue to successfully prevent harmful mergers, and could backfire by putting even further emphasis on market definition and structural presumptions.

If the agencies were to substantially change the presumption thresholds, they would also need to persuade courts that the new thresholds were at the right level. Is the evidence there to do so? The existing body of research on this question is, today, thin and mostly based on individual case studies in a handful of industries. Our reading of the literature is that it is not clear and persuasive enough, at this point in time, to support a substantially different threshold that will be applied across the board to all industries and market conditions. (emphasis added) [8]

Lower merger thresholds create several risks. One is that such thresholds will lead to excessive “false positives”; that is, too many presumptions against mergers that are likely to be procompetitive or benign. This is particularly likely to occur if enforcers make it harder for parties to rebut the presumptions, e.g., by requiring stronger evidence the higher the parties are above the (now-lowered) threshold. Raising barriers to establishing efficiencies and other countervailing factors makes it more likely that procompetitive mergers will be blocked. This not only risks depriving consumers of lower prices and greater innovation in specific cases, but chills beneficial merger-and-acquisition activity more broadly. The prospect of an overly stringent enforcement regime discourages investment and entrepreneurial activity. It also allocates scarce enforcement resources to the wrong cases.

Changing the Character of Structural Presumptions

Finally, the risks described above are particularly acute, given the change in the character of structural presumptions described in the U.S. Merger Guidelines. The 2023 Merger Guidelines—and only the 2023 Merger Guidelines—state that certain structural features of mergers will raise a “presumption of illegality.”[9]

U.S. merger guidelines published in 1982,[10] 1992 (revised in 1997),[11] and 2010[12] all describe structural thresholds seen by the agencies as pertinent to merger screening. None of them mention a “presumption of illegality.” In fact, as the U.S. agencies put it in the 2010 Horizontal Merger Guidelines:

The purpose of these thresholds is not to provide a rigid screen to separate competitively benign mergers from anticompetitive ones, although high levels of concentration do raise concerns. Rather, they provide one way to identify some mergers unlikely to raise competitive concerns and some others for which it is particularly important to examine whether other competitive factors confirm, reinforce, or counteract the potentially harmful effects of increased concentration.[13]

The most worrisome category of mergers identified in the 1992 U.S. merger guidelines were said to be presumed “likely to create or enhance market power or facilitate its exercise.” The 1982 guidelines did not describe “presumptions” so much as that certain mergers that may be matters of “significant competitive concern” and “likely” to be subject to challenge.

Hence, earlier editions of the U.S. merger guidelines describe the ways that structural features of mergers might inform, but not determine, internal agency analysis of those mergers. That was useful information for industry, the bar, and the courts. Equally useful were descriptions of mergers that were “unlikely to have adverse competitive effects and ordinarily require no further analysis,”[14] as well as intermediate types of mergers that “potentially raise significant competitive concerns and often warrant scrutiny.”[15]

Similarly, the 1992 U.S. merger guidelines identified a tier of mergers deemed “unlikely to have adverse competitive effects and ordinarily require no further analysis,” as well as intermediate categories of mergers either unlikely to have anticompetitive effects or, in the alternative, potentially raising significant competitive concerns, depending on various factors described elsewhere in the guidelines.[16]

By way of contrast, the new U.S. guidelines include no description of any mergers that are unlikely to have adverse competitive effects. And while the new merger guidelines do stipulate that the “presumption of illegality can be rebutted or disproved,” they offer very limited means of rebuttal.

This is at odds with prior U.S. agency practice and established U.S. law. Until very recently, U.S. agency staff sought to understand proposed mergers under the totality of their circumstances, much as U.S. courts came to do. Structural features of mergers (among many others) might raise concerns of greater or lesser degrees. These might lead to additional questions in some instances; more substantial inquiries under a “second request” in a minority of instances; or, eventually, a complaint against a very small minority of proposed mergers. In the alternative, they might help staff avoid wasting scarce resources on mergers “unlikely to have anticompetitive effects.”

Prior to a hearing or a trial on the merits, there might be strong, weak, or no appreciable assessments of likely liability, but there was no prima facie determination of illegality.

And while U.S. merger trials did tend to follow a burden-shifting framework for plaintiff and defendant production, they too looked to the “totality of the circumstances”[17] and a transaction’s “probable effect on future competition”[18] to determine liability, and they looked away from strong structural presumptions. As then-U.S. Circuit Judge Clarence Thomas observed in the Baker-Hughes case:

General Dynamics began a line of decisions differing markedly in emphasis from the Court’s antitrust cases of the 1960s. Instead of accepting a firm’s market share as virtually conclusive proof of its market power, the Court carefully analyzed defendants’ rebuttal evidence.[19]

Central to the holding in Baker Hughes—and contra the 2023 U.S. merger guidelines—was that, because the government’s prima facie burden of production was low, the defendant’s rebuttal burden should not be unduly onerous.[20] As the U.S. Supreme Court had put it, defendants would not be required to clearly disprove anticompetitive effects, but rather, simply to “show that the concentration ratios, which can be unreliable indicators of actual market behavior . . . did not accurately depict the economic characteristics of the [relevant] market.”[21]

Doing so would not end the matter. Rather, “the burden of producing additional evidence of anticompetitive effects shifts to the government, and merges with the ultimate burden of persuasion, which remains with the government at all times.”[22]

As the U.S. Supreme Court decision in Marine Bancorporation underscores, even by 1974, it was well understood that concentration ratios “can be unreliable indicators” of market behavior and competitive effects.

As explained above, research and enforcement over the ensuing decades have undermined reliance on structural presumptions even further. As a consequence, the structure/conduct/performance paradigm has been largely abandoned, because it’s widely recognized that market structure is not outcome–determinative.

That is not to say that high concentration cannot have any signaling value in preliminary agency screening of merger matters. But concentration metrics that have proven to be unreliable indicators of firm behavior and competitive effects should not be enshrined in Canadian statutory law. That would be a step back, not a step forward, for merger enforcement.

 

[1] Matthew Boswell, Letter to the Chair and Members of the House of Commons Standing Committee on Finance, Competition Bureau Canada (Mar. 1, 2024), available at https://sencanada.ca/Content/Sen/Committee/441/NFFN/briefs/SM-C-59_CompetitionBureauofCND_e.pdf.

[2] For a few examples from a very large body of literature, see, e.g., Steven Berry, Martin Gaynor, & Fiona Scott Morton, Do Increasing Markups Matter? Lessons from Empirical Industrial Organization, 33J. Econ. Perspectives 44 (2019); Richard Schmalensee, Inter-Industry Studies of Structure and Performance, in 2 Handbook of Industrial Organization 951-1009 (Richard Schmalensee & Robert Willig, eds., 1989); William N. Evans, Luke M. Froeb, & Gregory J. Werden, Endogeneity in the Concentration-Price Relationship: Causes, Consequences, and Cures, 41 J. Indus. Econ. 431 (1993); Steven Berry, Market Structure and Competition, Redux, FTC Micro Conference (Nov. 2017), available at https://www.ftc.gov/system/files/documents/public_events/1208143/22_-_steven_berry_keynote.pdf; Nathan Miller, et al., On the Misuse of Regressions of Price on the HHI in Merger Review, 10 J. Antitrust Enforcement 248 (2022).

[3] Id.

[4] See Harold Demsetz, Industry Structure, Market Rivalry, and Public Policy, 16 J. L. & Econ. 1 (1973).

[5] Berry, Gaynor, & Scott Morton, supra note 2.

[6] Chad Syverson, Macroeconomics and Market Power: Context, Implications, and Open Questions 33 J. Econ. Persp. 23, (2019) at 26.

[7] Joseph Farrell & Carl Shapiro, The 2010 Horizontal Merger Guidelines After 10 Years, 58 REV. IND. ORG. 58, (2021). https://link.springer.com/article/10.1007/s11151-020-09807-6.

[8] John Asker et al, Comments on the January 2022 DOJ and FTC RFI on Merger Enforcement (Apr. 20, 2022), available at https://www.regulations.gov/comment/FTC-2022-0003-1847 at 15-6.

[9] U.S. Dep’t Justice & Fed. Trade Comm’n, Merger Guidelines (Guideline One) (Dec. 18, 2023), available at https://www.ftc.gov/system/files/ftc_gov/pdf/2023_merger_guidelines_final_12.18.2023.pdf.

[10] U.S. Dep’t Justice, 1982 Merger Guidelines (1982), https://www.justice.gov/archives/atr/1982-merger-guidelines.

[11] U.S. Dep’t Justice & Fed. Trade Comm’n, 1992 Merger Guidelines (1992), https://www.justice.gov/archives/atr/1992-merger-guidelines; U.S. Dep’t Justice & Fed. Trade Comm’n, 1997 Merger Guidelines (1997), https://www.justice.gov/archives/atr/1997-merger-guidelines.

[12] U.S. Dep’t Justice & Fed. Trade Comm’n, Horizontal Merger Guidelines (Aug. 19, 2010), https://www.justice.gov/atr/horizontal-merger-guidelines-08192010; The U.S. antitrust agencies also issued Vertical Merger Guidelines in 2020. Although these were formally withdrawn in 2021 by the FTC, but not DOJ, they too are supplanted by the 2023 Merger Guidelines. See U.S. Dep’t Justice & Fed. Trade Comm’n, Vertical Merger Guidelines (Jun. 30, 2020), available at https://www.ftc.gov/system/files/documents/public_statements/1580003/vertical_merger_guidelines_6-30-20.pdf.

[13] 2010 Horizontal Merger Guidelines.

[14] Id.

[15] Id.

[16] 1992 Merger Guidelines.

[17]  United States v. Baker-Hughes Inc., 908 F.2d 981, 984 (D.C. Cir. 1990).

[18] Id. at 991.

[19] Id. at 990 (citing Hospital Corp. of Am. v. FTC, 807 F.2d 1381, 1386 (7th Cir.1986), cert. denied, 481 U.S. 1038, 107 S.Ct. 1975, 95 L.Ed.2d 815 (1987).

[20]  Id. at 987, 992.

[21]  United States v. Marine Bancorporation Inc., 418 U.S. 602, 631 (1974) (internal citations omitted).

[22]  Baker-Hughes, 908 F.2d at 983.

Continue reading
Antitrust & Consumer Protection

The Broken Promises of Europe’s Digital Regulation

TOTM If you live in Europe, you may have noticed issues with some familiar online services. From consent forms to reduced functionality and new fees, there . . .

If you live in Europe, you may have noticed issues with some familiar online services. From consent forms to reduced functionality and new fees, there is a sense that platforms like Amazon, Google, Meta, and Apple are changing the way they do business. 

Many of these changes are the result of a new European regulation called the Digital Markets Act (DMA), which seeks to increase competition in online markets. Under the DMA, so-called “gatekeepers” must allow rivals to access their platforms. Having taken effect March 7, firms now must comply with the regulation, which explains why we are seeing these changes unfold today.

Read the full piece here.

Continue reading
Antitrust & Consumer Protection

Test SLC (merger)

Popular Media DEFINITION The substantial lessening of competition or “SLC” test is a standard that regulatory authorities use to assess the legality of proposed mergers and acquisitions. . . .

DEFINITION

The substantial lessening of competition or “SLC” test is a standard that regulatory authorities use to assess the legality of proposed mergers and acquisitions. The SLC test examines whether a prospective merger is likely to substantially lessen competition in a given market. Its purpose is to prevent mergers that increase prices, reduce output, limit consumer choice, or stifle innovation as a result of a decrease in competition. Mergers that substantially lessen competition are prohibited under the laws of the jurisdictions that utilize this test, such as the USA, EU, Canada, the United Kingdom, Australia and Nigeria, amongst others.

Read the full piece here.

Continue reading
Antitrust & Consumer Protection

A Competition Perspective on Physician Non-Compete Agreements

Scholarship Abstract Physician non-compete agreements may have significant competitive implications, and effects on both providers and patients, but they are treated variously under the law on . . .

Abstract

Physician non-compete agreements may have significant competitive implications, and effects on both providers and patients, but they are treated variously under the law on a state-by-state basis. Reviewing the relevant law and the economic literature cannot identify with confidence the net effects of such agreements on either physicians or health care delivery with any generality. In addition to identifying future research projects to inform policy, it is argued that the antitrust “rule of reason” provides a useful and established framework with which to evaluate such agreements in specific health care markets and, potentially, to address those agreements most likely to do significant damage to health care competition and consumers.

Continue reading
Antitrust & Consumer Protection

ICLE Comments to European Commission on Competition in Virtual Worlds

Regulatory Comments Executive Summary We welcome the opportunity to comment on the European Commission’s call for contributions on competition in “Virtual Worlds”.[1] The International Center for Law . . .

Executive Summary

We welcome the opportunity to comment on the European Commission’s call for contributions on competition in “Virtual Worlds”.[1] The International Center for Law & Economics (“ICLE”) is a nonprofit, nonpartisan global research and policy center founded with the goal of building the intellectual foundations for sensible, economically grounded policy. ICLE promotes the use of law & economics methodologies to inform public-policy debates and has longstanding expertise in the evaluation of competition law and policy. ICLE’s interest is to ensure that competition law remains grounded in clear rules, established precedent, a record of evidence, and sound economic analysis.

The metaverse is an exciting and rapidly evolving set of virtual worlds. As with any new technology, concerns about the potential risks and negative consequences that the metaverse may bring have moved policymakers to explore how best to regulate this new space.

From the outset, it is important to recognize that simply because the metaverse is new does not mean that competition in this space is unregulated or somehow ineffective. Existing regulations may not explicitly or exclusively target metaverse ecosystems, but a vast regulatory apparatus already covers most aspects of business in virtual worlds. This includes European competition law, the Digital Markets Act (“DMA”), the General Data Protection Act (“GDPR), the Digital Services Act (“DSA”), and many more. Before it intervenes in this space, the commission should carefully consider whether there are any metaverse-specific problems not already addressed by these legal provisions.

This sense that competition intervention would be premature is reinforced by three important factors.

The first is that competition appears particularly intense in this space (Section I). There are currently multiple firms vying to offer compelling virtual worlds. At the time of writing, however, none appears close to dominating the market. In turn, this intense competition will encourage platforms to design services that meet consumers’ demands, notably in terms of safety and privacy. Nor does the market appear likely to fall into the hands of one of the big tech firms that command a sizeable share of more traditional internet services. Meta notoriously has poured more than $3.99 billion into its metaverse offerings during the first quarter of 2023, in addition to $13.72 billion the previous calendar year.[2] Despite these vast investments and a strategic focus on metaverse services, the company has, thus far, struggled to achieve meaningful traction in the space.[3]

Second, the commission’s primary concern appears to be that metaverses will become insufficiently “open and interoperable”.[4] But to the extent that these ecosystems do, indeed, become closed and proprietary, there is no reason to believe this to be a problem. Closed and proprietary ecosystems have several features that may be attractive to consumers and developers (Section II). These include improved product safety, performance, and ease of development. This is certainly not to say that closed ecosystems are always better than more open ones, but rather that it would be wrong to assume that one model or the other is optimal. Instead, the proper balance depends on tradeoffs that markets are better placed to decide.

Finally, timing is of the essence (Section III). Intervening so early in a fledgling industry’s life cycle is like shooting a moving target from a mile away. New rules or competition interventions might end up being irrelevant. Worse, by signaling that metaverses will be subject to heightened regulatory scrutiny for the foreseeable future, the commission may chill investment from the very firms is purports to support. In short, the commission should resist the urge to intervene so long as the industry is not fully mature.

I. Competing for Consumer Trust

The Commission is right to assume, in its call for contributions, that the extent to which metaverse services compete with each other (and continue to do so in the future) will largely determine whether they fulfil consumers’ expectations and meet the safety and trustworthiness requirements to which the commission aspires. As even the left-leaning Lessig put it:

Markets regulate behavior in cyberspace too. Prices structures often constrain access, and if they do not, then busy signals do. (America Online (AOL) learned this lesson when it shifted from an hourly to a flat-rate pricing plan.) Some sites on the web charge for access, as on-line services like AOL have for some time. Advertisers reward popular sites; online services drop unpopular forums. These behaviors are all a function of market constraints and market opportunity, and they all reflect the regulatory role of the market.[5]

Indeed, in a previous call for contributions, the Commission implicitly recognized the important role that competition plays, although it frames the subject primarily in terms of the problems that would arise if competition ceased to operate:

There is a risk of having a small number of big players becoming future gatekeepers of virtual worlds, creating market entry barriers and shutting out EU start-ups and SMEs from this emerging market. Such a closed ecosystem with the prevalence of proprietary systems can negatively affect the protection of personal information and data, the cybersecurity and the freedom and openness of virtual worlds at the same time.[6]

It is thus necessary to ask whether there is robust competition in the market for metaverse services. The short answer is a resounding yes.

A. Competition Without Tipping

While there is no precise definition of what constitutes a metaverse—much less a precise definition of the relevant market—available data suggests the space is highly competitive. This is evident in the fact that even a major global firm like Meta—having invested billions of dollars in its metaverse branch (and having rebranded the company accordingly)—has struggled to gain traction.[7]

Other major players in the space include the likes of Roblox, Fortnite, and Minecraft, which all have somewhere between 70 and 200 million active users.[8] This likely explains why Meta’s much-anticipated virtual world struggled to gain meaningful traction with consumers, stalling at around 300,000 active users.[9] Alongside these traditional players, there are also several decentralized platforms that are underpinned by blockchain technology. While these platforms have attracted massive investments, they are largely peripheral in terms of active users, with numbers often only in the low thousands.[10]

There are several inferences that can be drawn from these limited datasets. For one, it is clear that the metaverse industry is not yet fully mature. There are still multiple paradigms competing for consumer attention: game-based platforms versus social-network platforms; traditional platforms versus blockchain platforms, etc. In the terminology developed by David Teece, the metaverse industry has not yet reached a “paradigmatic” stage. It is fair to assume there is still significant scope for the entry of differentiated firms.[11]

It is also worth noting that metaverse competition does not appear to exhibit the same sort of network effects and tipping that is sometimes associated with more traditional social networks.[12] Despite competing for nearly a decade, no single metaverse project appears to be running away with the market.[13] This lack of tipping might be because these projects are highly differentiated.[14] It may also be due to the ease of multi-homing among them.[15]

More broadly, it is far from clear that competition will lead to a single metaverse for all uses. Different types of metaverse services may benefit from different user interfaces, graphics, and physics engines. This cuts in favor of multiple metaverses coexisting, rather than all services coordinating within a single ecosystem. Competition therefore appears likely lead to the emergence of multiple differentiated metaverses, rather than a single winner.

Ultimately, competition in the metaverse industry is strong and there is little sense these markets are about to tip towards a single firm in the year future.

B. Competing for Consumer Trust

As alluded to in the previous subsection, the world’s largest and most successful metaverse entrants to date are traditional videogaming platforms that have various marketplaces and currencies attached.[16] In other words, decentralized virtual worlds built upon blockchain technology remain marginal.

This has important policy implications. The primary legal issues raised by metaverses are the same as those encountered on other digital marketplaces. This includes issues like minor fraud, scams, and children buying content without their parents’ authorization.[17] To the extent these harms are not adequately deterred by existing laws, metaverse platforms themselves have important incentives to police them. In turn, these incentives may be compounded by strong competition among platforms.

Metaverses are generally multi-sided platforms that bring together distinct groups of users, including consumers and content creators. In order to maximize the value of their ecosystems, platforms have an incentive to balance the interests of these distinct groups.[18] In practice, this will often mean offering consumers various forms of protection against fraud and scams and actively policing platforms’ marketplaces. As David Evans puts it:

But as with any community, there are numerous opportunities for people and businesses to create negative externalities, or engage in other bad behavior, that can reduce economic efficiency and, in the extreme, lead to the tragedy of the commons. Multi-sided platforms, acting selfishly to maximize their own profits, often develop governance mechanisms to reduce harmful behavior. They also develop rules to manage many of the same kinds of problems that beset communities subject to public laws and regulations. They enforce these rules through the exercise of property rights and, most importantly, through the “Bouncer’s Right” to exclude agents from some quantum of the platform, including prohibiting some agents from the platform entirely…[19]

While there is little economic research to suggest that competition directly increases hosts’ incentive to policy their platforms, it stands to reason that doing so effectively can help platforms to expand the appeal of their ecosystems. This is particularly important for metaverse services whose userbases remain just a fraction of the size they could ultimately reach. While 100 or 200 million users already comprises a vast ecosystem, it pales in comparison to the sometimes billions of users that “traditional” online platforms attract.

The bottom line is that the market for metaverses is growing. This likely compounds platforms’ incentives to weed out undesirable behavior, thereby complementing government efforts to achieve the same goal.

II. Opening Platforms or Opening Pandora’s Box?

In its call for contributions, the commission seems concerned that the metaverse competition may lead to closed ecosystems that may be less beneficial to consumers than more open ones. But if this is indeed the commission’s fear, it is largely unfounded.

There are many benefits to closed ecosystems. Choosing the optimal degree of openness entails tradeoffs. At the very least, this suggests that policymakers should be careful not to assume that opening platforms up will systematically provide net benefits to consumers.

A. Antitrust Enforcement and Regulatory Initiatives

To understand why open (and weakly propertized) platforms are not always better for consumers, it is worth looking at past competition enforcement in the online space. Recent interventions by competition authorities have generally attempted (or are attempting) to move platforms toward more openness and less propertization. For their part, these platforms are already tremendously open (as the “platform” terminology implies) and attempt to achieve a delicate balance between centralization and decentralization.

Figure I: Directional Movement of Antitrust Intervention

The Microsoft cases and the Apple investigation both sought or seek to bring more openness and less propertization to those respective platforms. Microsoft was made to share proprietary data with third parties (less propertization) and to open its platform to rival media players and web browsers (more openness).[20] The same applies to Apple. Plaintiffs in private antitrust litigation brought in the United States[21] and government enforcement actions in Europe[22] are seeking to limit the fees that Apple can extract from downstream rivals (less propertization), as well as to ensure that it cannot exclude rival mobile-payments solutions from its platform (more openness).

The various cases that were brought by EU and U.S. authorities against Qualcomm broadly sought to limit the extent to which it was monetizing its intellectual property.[23] The European Union’s Amazon investigation centers on the ways in which the company uses data from third-party sellers (and, ultimately, the distribution of revenue between those sellers and Amazon).[24] In both cases, authorities are ultimately trying to limit the extent to which firms can propertize their assets.

Finally, both of the EU’s Google cases sought to bring more openness to the company’s main platform. The Google Shopping decision sanctioned Google for purportedly placing its services more favorably than those of its rivals.[25] The separate Android decision sought to facilitate rival search engines’ and browsers’ access to the Android ecosystem. The same appears to be true of ongoing litigation brought by state attorneys general in the United States.[26]

Much of the same can be said of the numerous regulatory initiatives pertaining to digital markets. Indeed, draft regulations being contemplated around the globe mimic the features of the antitrust/competition interventions discussed above. For instance, it is widely accepted that Europe’s DMA effectively transposes and streamlines the enforcement of the theories harm described above.[27] Similarly, several scholars have argued that the proposed American Innovation and Choice Online Act (“AICOA”) in the United States largely mimics European competition policy.[28] The legislation would ultimately require firms to open up their platforms, most notably by forcing them to treat rival services as they would their own and to make their services more interoperable with those rivals.[29]

What is striking about these decisions and investigations is the extent to which authorities are pushing back against the very features that distinguish the platforms they are investigating. Closed (or relatively closed) platforms are forced to open up, and firms with highly propertized assets are made to share them (or, at the very least, monetize them less aggressively).

B. The Empty Quadrant

All of this would not be very interesting if it weren’t for a final piece of the puzzle: the model of open and shared platforms that authorities apparently favor has traditionally struggled to gain traction with consumers. Indeed, there seem to be vanishingly few successful consumer-oriented products and services in this space.

There have been numerous attempts to introduce truly open consumer-oriented operating systems in both the mobile and desktop segments. Most have ended in failure. Ubuntu and other flavors of the Linux operating system remain fringe products. There have been attempts to create open-source search engines, but they have not met with success.[30] The picture is similar in the online retail space. Amazon appears to have beaten eBay, despite the latter being more open and less propertized. Indeed, Amazon has historically charged higher fees than eBay and offers sellers much less freedom in the ways in which they may sell their goods.[31]

This theme is repeated in the standardization space. There have been innumerable attempts to impose open, royalty-free standards. At least in the mobile-internet industry, few (if any) of these have taken off. Instead, proprietary standards such as 5G and WiFi have been far more successful. That pattern is repeated in other highly standardized industries, like digital-video formats. Most recently, the proprietary Dolby Vision format seems to be winning the war against the open HDR10+ format.[32]

Figure II: Open and Shared Platforms

This is not to say that there haven’t been any successful examples of open, royalty-free standards. Internet protocols, blockchain, and Wikipedia all come to mind. Nor does it mean that we will not see more decentralized goods in the future. But by and large, firms and consumers have not yet taken to the idea of fully open and shared platforms. Or, at least, those platforms have not yet achieved widespread success in the marketplace (potentially due to supply-side considerations, such as the difficulty of managing open platforms or the potentially lower returns to innovation in weakly propertized ones).[33] And while some “open” projects have achieved tremendous scale, the consumer-facing side of these platforms is often dominated by intermediaries that opt for much more traditional business models (think of Coinbase in the blockchain space, or Android’s use of Linux).

C. Potential Explanations

The preceding section posited a recurring reality: the digital platforms that competition authorities wish to bring into existence are fundamentally different from those that emerge organically. But why have authorities’ ideal platforms, so far, failed to achieve truly meaningful success?

Three potential explanations come to mind. First, “closed” and “propertized” platforms might systematically—and perhaps anticompetitively—thwart their “open” and “shared” rivals. Second, shared platforms might fail to persist (or grow pervasive) because they are much harder to monetize, and there is thus less incentive to invest in them. This is essentially a supply-side explanation. Finally, consumers might opt for relatively closed systems precisely because they prefer these platforms to marginally more open ones—i.e., a demand-side explanation.

In evaluating the first conjecture, the key question is whether successful “closed” and “propertized” platforms overcame their rivals before or after they achieved some measure of market dominance. If success preceded dominance, then anticompetitive foreclosure alone cannot explain the proliferation of the “closed” and “propertized” model.[34]

Many of today’s dominant platforms, however, often overcame open/shared rivals, well before they achieved their current size. It is thus difficult to make the case that the early success of their business models was due to anticompetitive behavior. This is not to say these business models cannot raise antitrust issues, but rather that anticompetitive behavior is not a good explanation for their emergence.

Both the second and the third conjectures essentially ask whether “closed” and “propertized” might be better adapted to their environment than “open” and “shared” rivals.

In that respect, it is not unreasonable to surmise that highly propertized platforms would generally be easier to monetize than shared ones. For example, to monetize open-source platforms often requires relying on complementarities, which tend to be vulnerable to outside competition and free-riding.[35] There is thus a natural incentive for firms to invest and innovate in more propertized environments. In turn, competition enforcement that limits a platform’s ability to propertize their assets may harm innovation.

Similarly, authorities should reflect on whether consumers really want the more “competitive” ecosystems that they are trying to design. The European Commission, for example, has a long track record of seeking to open digital platforms, notably by requiring that platform owners do not preinstall their own web browsers (the Microsoft decisions are perhaps the most salient example). And yet, even after these interventions, new firms have kept using the very business model that the commission reprimanded, rather than the “pro-consumer” model it sought to impose on the industry. For example, Apple tied the Safari browser to its iPhones; Google went to some length to ensure that Chrome was preloaded on devices; and Samsung phones come with Samsung Internet as default.[36] Yet this has not ostensibly steered consumers away from those platforms.

Along similar lines, a sizable share of consumers opt for Apple’s iPhone, which is even more centrally curated than Microsoft Windows ever was (and the same is true of Apple’s MacOS). In other words, it is hard to claim that opening platforms is inherently good for consumers when those same consumers routinely opt for platforms with the very features that policymakers are trying to eliminate.

Finally, it is worth noting that the remedies imposed by competition authorities have been anything but successes. Windows XP N (the version of Windows that came without Windows Media Player) was an unmitigated flop, selling a paltry 1,787 copies.[37] Likewise, the internet-browser “ballot box” imposed by the commission was so irrelevant to consumers that it took months for authorities to notice that Microsoft had removed it, in violation of the commission’s decision.[38]

One potential inference is that consumers do not value competition interventions that make dominant ecosystems marginally more open and less propertized. There are also many reasons why consumers might prefer “closed” systems (at least, relative to the model favored by many policymakers), even when they must pay a premium for them.

Take the example of app stores. Maintaining some control over the apps that can access the store enables platforms to easily weed out bad actors. Similarly, controlling the hardware resources that each app can use may greatly improve device performance. Indeed, it may be that a measure of control facilitates the very innovations that consumers demand. Therefore, “authorities and courts should not underestimate the indispensable role control plays in achieving coordination and coherence in the context of systemic ef?ciencies. Without it, the attempted novelties and strategies might collapse under their own complexity.”[39]

Relatively centralized platforms can eliminate negative externalities that “bad” apps impose on rival apps and consumers.[40] This is especially true when consumers will tend to attribute dips in performance to the overall platform, rather than to a particular app.[41] At the same time, they can take advantage of positive externalities to improve the quality of the overall platform.

And it is surely the case that consumers prefer to make many of their decisions at the inter-platform level, rather than within each platform. In simple terms, users arguably make their most important decision when they choose between an Apple or Android smartphone (or a Mac and a PC, etc.). In doing so, they can select their preferred app suite with one simple decision. They might thus purchase an iPhone because they like the secure App Store, or an Android smartphone because they like the Chrome Browser and Google Search. Absent false information at the time of the initial platform decision, this decision will effectively incorporate expectations about subsequent constraints.[42]

Furthermore, forcing users to make too many “within-platform” choices may undermine a product’s attractiveness. Indeed, it is difficult to create a high-quality reputation if each user’s experience is fundamentally different.[43] In short, contrary to what antitrust authorities appear to believe, closed platforms might give most users exactly what they desire.

All of this suggests that consumers and firms often gravitate spontaneously toward both closed and highly propertized platforms, the opposite of what the commission and other competition authorities tend to favor. The reasons for this trend are still misunderstood, and mostly ignored. Too often it is simply assumed that consumers benefit from more openness, and that shared/open platforms are the natural order of things. Instead, what some regard as “market failures” may in fact be features that explain the rapid emergence of the digital economy.

When considering potential policy reforms targeting the metaverse, policymakers would be wrong to assume openness (notably, in the form of interoperability) and weak propertization are always objectively superior. Instead, these platform designs entail important tradeoffs. Closed metaverse ecosystems may lead to higher consumer safety and better performance, while interoperable systems may reduce the frictions consumers face when moving from one service to another. There is little reason to believe policymakers are in a better position to weigh these tradeoffs than consumers, who vote with their virtual feet.

III. Conclusion: Competition Intervention Would be Premature

A final important argument against intervening today is that the metaverse industry is nowhere near mature. Tomorrow’s competition-related challenges and market failures might not be the same as today’s. This makes it exceedingly difficult for policymakers to design appropriate remedies and increases the risk that intervention might harm innovation.

As of 2023, the entire metaverse industry (both hardware and software) is estimated to be worth somewhere in the vicinity of $80 billion, and projections suggest this could grow by a factor of 10 by 2030.[44] Growth projections of this sort are notoriously unreliable. But in this case, they do suggest there is some consensus that the industry is not fully fledged.

Along similar lines, it remains unclear what types of metaverse services will gain the most traction with consumers, what sorts of hardware consumers will use to access these services, and what technologies will underpin the most successful metaverse platforms. In fact, it is still an open question whether the metaverse industry will foster any services that achieve widespread consumer adoption in the foreseeable future.[45] In other words, it is not exactly clear what metaverse products and services the Commission should focus on in the first place.

Given these uncertainties, competition intervention in the metaverse appears premature. Intervening so early in the industry’s life cycle is like aiming at a moving target. Ensuing remedies might end up being irrelevant before they have any influence on the products that firms develop. More worryingly, acting now signals that the metaverse industry will be subject to heightened regulatory scrutiny for the foreseeable future. In turn, this may deter large platforms from investing in the European market. It also may funnel venture-capital investments away from the European continent.

Competition intervention in burgeoning industries is no free lunch. The best evidence concerning these potential costs comes from the GDPR. While privacy regulation is obviously not the same as competition law, the evidence concerning the GDPR suggests that heavy-handed intervention may, at least in some instances, slow down innovation and reduce competition.

The most-cited empirical evidence concerning the effects of the GDPR comes from a paper by Garrett Johnson and co-authors, who link the GDPR to widespread increases to market concentration, particularly in the short-term:

We show that websites’ vendor use falls after the European Union’s (EU’s) General Data Protection Regulation (GDPR), but that market concentration also increases among technology vendors that provide support services to websites…. The week after the GDPR’s enforcement, website use of web technology vendors falls by 15% for EU residents. Websites are relatively more likely to retain top vendors, which increases the concentration of the vendor market by 17%. Increased concentration predominantly arises among vendors that use personal data, such as cookies, and from the increased relative shares of Facebook and Google-owned vendors, but not from website consent requests. Although the aggregate changes in vendor use and vendor concentration dissipate by the end of 2018, we find that the GDPR impact persists in the advertising vendor category most scrutinized by regulators.[46]

Along similar lines, an NBER working paper by Jian Jia and co-authors finds that enactment of the GDPR markedly reduced venture-capital investments in Europe:

Our findings indicate a negative differential effect on EU ventures after the rollout of GDPR relative to their US counterparts. These negative effects manifest in the overall number of financing rounds, the overall dollar amount raised across rounds, and in the dollar amount raised per individual round. Specifically, our findings suggest a $3.38 million decrease in the aggregate dollars raised by EU ventures per state per crude industry category per week, a 17.6% reduction in the number of weekly venture deals, and a 39.6% decrease in the amount raised in an average deal following the rollout of GDPR.[47]

In another paper, Samuel Goldberg and co-authors find that the GDPR led to a roughly 12% reduction in website pageviews and e-commerce revenue in Europe.[48] Finally, Rebecca Janssen and her co-authors show that the GDPR decreased the number of apps offered on Google’s Play Store between 2016 and 2019:

Using data on 4.1 million apps at the Google Play Store from 2016 to 2019, we document that GDPR induced the exit of about a third of available apps; and in the quarters following implementation, entry of new apps fell by half.[49]

Of course, the body of evidence concerning the GDPR’s effects is not entirely unambiguous. For example, Rajkumar Vekatesean and co-authors find that the GDPR had mixed effects on the returns of different types of firms.[50] Other papers also show similarly mixed effects.[51]

Ultimately, the empirical literature concerning the effects of the GDPR shows that regulation—in this case, privacy protection—is no free lunch. Of course, this does not mean that competition intervention targeting the metaverse would necessarily have these same effects. But in the absence of a clear market failure to solve, it is unclear why policymakers should run such a risk in the first place.

In the end, competition intervention in the metaverse is unlikely to be costless. The metaverse is still in its infancy, regulation could deter essential innovation, and the commission has thus far failed to identify any serious market failures that warrant public intervention. The result is that the commission’s call for contributions appears premature or, in other words, that the commission is putting the meta-cart before the meta-horse.

 

[1] Competition in Virtual Worlds and Generative AI – Calls for contributions, European Commission (Jan. 9, 2024) https://competition-policy.ec.europa.eu/document/download/e727c66a-af77-4014-962a-7c9a36800e2f_en?filename=20240109_call-for-contributions_virtual-worlds_and_generative-AI.pdf (hereafter, “Call for Contributions”).

[2] Jonathan Vaian, Meta’s Reality Labs Records $3.99 Billion Quarterly Loss as Zuckerberg Pumps More Cash into Metaverse, CNBC (Apr. 26, 2023), https://www.cnbc.com/2023/04/26/metas-reality-labs-unit-records-3point99-billion-first-quarter-loss-.html.

[3] Alan Truly, Horizon Worlds Leak: Only 1 in 10 Users Return & Web Launch Is Coming, Mixed News (Mar. 3, 2023), https://mixed-news.com/en/horizon-worlds-leak-only-1-in-10-users-return-web-launch-coming; Kevin Hurler, Hey Fellow Kids: Meta Is Revamping Horizon Worlds to Attract More Teen Users, Gizmodo (Feb. 7, 2023), https://gizmodo.com/meta-metaverse-facebook-horizon-worlds-vr-1850082068; Emma Roth, Meta’s Horizon Worlds VR Platform Is Reportedly Struggling to Keep Users, The Verge (Oct. 15, 2022),
https://www.theverge.com/2022/10/15/23405811/meta-horizon-worlds-losing-users-report; Paul Tassi, Meta’s ‘Horizon Worlds’ Has Somehow Lost 100,000 Players in Eight Months, Forbes, (Oct. 17, 2022), https://www.forbes.com/sites/paultassi/2022/10/17/metas-horizon-worlds-has-somehow-lost-100000-players-in-eight-months/?sh=57242b862a1b.

[4] Call for Contributions, supra note 1. (“6) Do you expect the technology incorporated into Virtual World platforms, enabling technologies of Virtual Worlds and services based on Virtual Worlds to be based mostly on open standards and/or protocols agreed through standard-setting organisations, industry associations or groups of companies, or rather the use of proprietary technology?”).

[5] Less Lawrence Lessig, The Law of the Horse: What Cyberlaw Might Teach, 113 Harv. L. Rev. 508 (1999).

[6] Virtual Worlds (Metaverses) – A Vision for Openness, Safety and Respect, European Commission, https://ec.europa.eu/info/law/better-regulation/have-your-say/initiatives/13757-Virtual-worlds-metaverses-a-vision-for-openness-safety-and-respect/feedback_en?p_id=31962299H.

[7] Catherine Thorbecke, What Metaverse? Meta Says Its Single Largest Investment Is Now in ‘Advancing AI’, CNN Business (Mar. 15, 2023), https://www.cnn.com/2023/03/15/tech/meta-ai-investment-priority/index.html; Ben Marlow, Mark Zuckerberg’s Metaverse Is Shattering into a Million Pieces, The Telegraph (Apr. 23, 2023), https://www.telegraph.co.uk/business/2023/04/21/mark-zuckerbergs-metaverse-shattering-million-pieces; Will Gendron, Meta Has Reportedly Stopped Pitching Advertisers on the Metaverse, BusinessInsider (Apr. 18, 2023), https://www.businessinsider.com/meta-zuckerberg-stopped-pitching-advertisers-metaverse-focus-reels-ai-report-2023-4.

[8] Mansoor Iqbal, Fortnite Usage and Revenue Statistics, Business of Apps (Jan. 9, 2023), https://www.businessofapps.com/data/fortnite-statistics; Matija Ferjan, 76 Little-Known Metaverse Statistics & Facts (2023 Data), Headphones Addict (Feb. 13, 2023), https://headphonesaddict.com/metaverse-statistics.

[9] James Batchelor, Meta’s Flagship Metaverse Horizon Worlds Struggling to Attract and Retain Users, Games Industry (Oct. 17, 2022), https://www.gamesindustry.biz/metas-flagship-metaverse-horizon-worlds-struggling-to-attract-and-retain-users; Ferjan, id.

[10] Richard Lawler, Decentraland’s Billion-Dollar ‘Metaverse’ Reportedly Had 38 Active Users in One Day, The Verge (Oct. 13, 2022), https://www.theverge.com/2022/10/13/23402418/decentraland-metaverse-empty-38-users-dappradar-wallet-data; The Sandbox, DappRadar, https://dappradar.com/multichain/games/the-sandbox (last visited May 3, 2023); Decentraland, DappRadar, https://dappradar.com/multichain/social/decentraland (last visited May 3, 2023).

[11] David J. Teece, Profiting from Technological Innovation: Implications for Integration, Collaboration, Licensing and Public Policy, 15 Research Policy 285-305 (1986), https://www.sciencedirect.com/science/article/abs/pii/0048733386900272.

[12] Geoffrey Manne & Dirk Auer, Antitrust Dystopia and Antitrust Nostalgia: Alarmist Theories of Harm in Digital Markets and Their Origins, 28 Geo. Mason L. Rev. 1279 (2021).

[13] Roblox, Wikipedia, https://en.wikipedia.org/wiki/Roblox (last visited May 3, 2023); Minecraft, Wikipedia, https://en.wikipedia.org/wiki/Minecraft (last visited May 3, 2023); Fortnite, Wikipedia, https://en.wikipedia.org/wiki/Fortnite (last visited May 3, 2023); see Fiza Chowdhury, Minecraft vs Roblox vs Fortnite: Which Is Better?, Metagreats (Feb. 20, 2023), https://www.metagreats.com/minecraft-vs-roblox-vs-fortnite.

[14]  Marc Rysman, The Economics of Two-Sided Markets, 13 J. Econ. Perspectives 134 (2009) (“First, if standards can differentiate from each other, they may be able to successfully coexist (Chou and Shy, 1990; Church and Gandal, 1992). Arguably, Apple and Microsoft operating systems have both survived by specializing in different markets: Microsoft in business and Apple in graphics and education. Magazines are an obvious example of platforms that differentiate in many dimensions and hence coexist.”).

[15] Id. at 134 (“Second, tipping is less likely if agents can easily use multiple standards. Corts and Lederman (forthcoming) show that the fixed cost of producing a video game for one more standard have reduced over time relative to the overall fixed costs of producing a game, which has led to increased distribution of games across multiple game systems (for example, PlayStation, Nintendo, and Xbox) and a less-concentrated game system market.”).

[16] What Are Fortnite, Roblox, Minecraft and Among Us? A Parent’s Guide to the Most Popular Online Games Kids Are Playing, FTC Business (Oct. 5, 2021), https://www.ftc.net/blog/what-are-fortnite-roblox-minecraft-and-among-us-a-parents-guide-to-the-most-popular-online-games-kids-are-playing; Jay Peters, Epic Is Merging Its Digital Asset Stores into One Huge Marketplace, The Verge (Mar. 22, 2023), https://www.theverge.com/2023/3/22/23645601/epic-games-fab-asset-marketplace-state-of-unreal-2023-gdc.

[17] Luke Winkie, Inside Roblox’s Criminal Underworld, Where Kids Are Scamming Kids, IGN (Jan. 2, 2023), https://www.ign.com/articles/inside-robloxs-criminal-underworld-where-kids-are-scamming-kids; Fake Minecraft Updates Pose Threat to Users, Tribune (Sept. 11, 2022), https://tribune.com.pk/story/2376087/fake-minecraft-updates-pose-threat-to-users; Ana Diaz, Roblox and the Wild West of Teenage Scammers, Polygon (Aug. 24, 2019) https://www.polygon.com/2019/8/24/20812218/roblox-teenage-developers-controversy-scammers-prison-roleplay; Rebecca Alter, Fortnite Tries Not to Scam Children and Face $520 Million in FTC Fines Challenge, Vulture (Dec. 19, 2022), https://www.vulture.com/2022/12/fortnite-epic-games-ftc-fines-privacy.html; Leonid Grustniy, Swindle Royale: Fortnite Scammers Get Busy, Kaspersky Daily (Dec. 3, 2020), https://www.kaspersky.com/blog/top-four-fortnite-scams/37896.

[18] See, generally, David Evans & Richard Schmalensee, Matchmakers: The New Economics of Multisided Platforms (Harvard Business Review Press, 2016).

[19] David S. Evans, Governing Bad Behaviour By Users of Multi-Sided Platforms, Berkley Technology Law Journal 27:2 (2012), 1201.

[20] See Case COMP/C-3/37.792, Microsoft, OJ L 32 (May 24, 2004). See also, Case COMP/39.530, Microsoft (Tying), OJ C 120 (Apr. 26, 2013).

[21] See Complaint, Epic Games, Inc. v. Apple Inc., 493 F. Supp. 3d 817 (N.D. Cal. 2020) (4:20-cv-05640-YGR).

[22] See European Commission Press Release IP/20/1073, Antitrust: Commission Opens Investigations into Apple’s App Store Rules (Jun. 16, 2020); European Commission Press Release IP/20/1075, Antitrust: Commission Opens Investigation into Apple Practices Regarding Apple Pay (Jun. 16, 2020).

[23] See European Commission Press Release IP/18/421, Antitrust: Commission Fines Qualcomm €997 Million for Abuse of Dominant Market Position (Jan. 24, 2018); Federal Trade Commission v. Qualcomm Inc., 969 F.3d 974 (9th Cir. 2020).

[24] See European Commission Press Release IP/19/4291, Antitrust: Commission Opens Investigation into Possible Anti-Competitive Conduct of Amazon (Jul. 17, 2019).

[25] See Case AT.39740, Google Search (Shopping), 2017 E.R.C. I-379. See also, Case AT.40099 (Google Android), 2018 E.R.C.

[26] See Complaint, United States v. Google, LLC, (2020), https://www.justice.gov/opa/pr/justice-department-sues-monopolist-google-violating-antitrust-laws; see also, Complaint, Colorado et al. v. Google, LLC, (2020), available at https://coag.gov/app/uploads/2020/12/Colorado-et-al.-v.-Google-PUBLIC-REDACTED-Complaint.pdf.

[27] See, e.g., Giorgio Monti, The Digital Markets Act: Institutional Design and Suggestions for Improvement, Tillburg L. & Econ. Ctr., Discussion Paper No. 2021-04 (2021), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3797730 (“In sum, the DMA is more than an enhanced and simplified application of Article 102 TFEU: while the obligations may be criticised as being based on existing competition concerns, they are forward-looking in trying to create a regulatory environment where gatekeeper power is contained and perhaps even reduced.”) (Emphasis added).

[28] See, e.g., Aurelien Portuese, “Please, Help Yourself”: Toward a Taxonomy of Self-Preferencing, Information Technology & Innovation Foundation (Oct. 25, 2021), available at https://itif.org/sites/default/files/2021-self-preferencing-taxonomy.pdf. (“The latest example of such weaponization of self-preferencing by antitrust populists is provided by Sens. Amy Klobuchar (D-MN) and Chuck Grassley (R-IA). They introduced legislation in October 2021 aimed at prohibiting the practice.2 However, the legislation would ban self-preferencing only for a handful of designated companies—the so-called “covered platforms,” not the thousands of brick-and-mortar sellers that daily self-preference for the benefit of consumers. Mimicking the European Commission’s Digital Markets Act prohibiting self-preferencing, Senate and the House bills would degrade consumers’ experience and undermine competition, since self-preferencing often benefits consumers and constitutes an integral part, rather than an abnormality, of the process of competition.”).

[29] Efforts to saddle platforms with “non-discrimination” constraints are tantamount to mandating openness. See Geoffrey A. Manne, Against the Vertical Discrimination Presumption, Foreword, Concurrences No. 2-2020 (2020) at 2 (“The notion that platforms should be forced to allow complementors to compete on their own terms, free of constraints or competition from platforms is a species of the idea that platforms are most socially valuable when they are most ‘open.’ But mandating openness is not without costs, most importantly in terms of the effective operation of the platform and its own incentives for innovation.”).

[30] See, e.g., Klint Finley, Your Own Private Google: The Quest for an Open Source Search Engine, Wired (Jul. 12, 2021), https://www.wired.com/2012/12/solar-elasticsearch-google.

[31] See Brian Connolly, Selling on Amazon vs. eBay in 2021: Which Is Better?, JungleScout (Jan. 12, 2021), https://www.junglescout.com/blog/amazon-vs-ebay; Crucial Differences Between Amazon and eBay, SaleHOO, https://www.salehoo.com/educate/selling-on-amazon/crucial-differences-between-amazon-and-ebay (last visited Feb. 8, 2021).

[32] See, e.g., Dolby Vision Is Winning the War Against HDR10 +, It Requires a Single Standard, Tech Smart, https://voonze.com/dolby-vision-is-winning-the-war-against-hdr10-it-requires-a-single-standard (last visited June 6, 2022).

[33] On the importance of managers, see, e.g., Nicolai J Foss & Peter G Klein, Why Managers Still Matter, 56 MIT Sloan Mgmt. Rev., 73 (2014) (“In today’s knowledge-based economy, managerial authority is supposedly in decline. But there is still a strong need for someone to define and implement the organizational rules of the game.”).

[34] It is generally agreed upon that anticompetitive foreclosure is possible only when a firm enjoys some degree of market power. Frank H. Easterbrook, Limits of Antitrust, 63 Tex. L. Rev. 1, 20 (1984) (“Firms that lack power cannot injure competition no matter how hard they try. They may injure a few consumers, or a few rivals, or themselves (see (2) below) by selecting ‘anticompetitive’ tactics. When the firms lack market power, though, they cannot persist in deleterious practices. Rival firms will offer the consumers better deals. Rivals’ better offers will stamp out bad practices faster than the judicial process can. For these and other reasons many lower courts have held that proof of market power is an indispensable first step in any case under the Rule of Reason. The Supreme Court has established a market power hurdle in tying cases, despite the nominally per se character of the tying offense, on the same ground offered here: if the defendant lacks market power, other firms can offer the customer a better deal, and there is no need for judicial intervention.”).

[35] See, e.g., Josh Lerner & Jean Tirole, Some Simple Economics of Open Source, 50 J. Indus. Econ. 197 (2002).

[36] See Matthew Miller, Thanks, Samsung: Android’s Best Mobile Browser Now Available to All, ZDNet (Aug. 11, 2017), https://www.zdnet.com/article/thanks-samsung-androids-best-mobile-browser-now-available-to-all.

[37] FACT SHEET: Windows XP N Sales, RegMedia (Jun. 12, 2009), available at https://regmedia.co.uk/2009/06/12/microsoft_windows_xp_n_fact_sheet.pdf.

[38] See Case COMP/39.530, Microsoft (Tying), OJ C 120 (Apr. 26, 2013).

[39] Konstantinos Stylianou, Systemic Efficiencies in Competition Law: Evidence from the ICT Industry, 12 J. Competition L. & Econ. 557 (2016).

[40] See, e.g., Steven Sinofsky, The App Store Debate: A Story of Ecosystems, Medium (Jun. 21, 2020), https://medium.learningbyshipping.com/the-app-store-debate-a-story-of-ecosystems-938424eeef74.

[41] Id.

[42] See, e.g., Benjamin Klein, Market Power in Aftermarkets, 17 Managerial & Decision Econ. 143 (1996).

[43] See, e.g., Simon Hill, What Is Android Fragmentation, and Can Google Ever Fix It?, DigitalTrends (Oct. 31, 2018), https://www.digitaltrends.com/mobile/what-is-android-fragmentation-and-can-google-ever-fix-it.

[44] Metaverse Market Revenue Worldwide from 2022 to 2030, Statista, https://www.statista.com/statistics/1295784/metaverse-market-size (last visited May 3, 2023); Metaverse Market by Component (Hardware, Software (Extended Reality Software, Gaming Engine, 3D Mapping, Modeling & Reconstruction, Metaverse Platform, Financial Platform), and Professional Services), Vertical and Region – Global Forecast to 2027, Markets and Markets (Apr. 27, 2023), https://www.marketsandmarkets.com/Market-Reports/metaverse-market-166893905.html; see also, Press Release, Metaverse Market Size Worth $ 824.53 Billion, Globally, by 2030 at 39.1% CAGR, Verified Market Research (Jul. 13, 2022), https://www.prnewswire.com/news-releases/metaverse-market-size-worth–824-53-billion-globally-by-2030-at-39-1-cagr-verified-market-research-301585725.html.

[45] See, e.g., Megan Farokhmanesh, Will the Metaverse Live Up to the Hype? Game Developers Aren’t Impressed, Wired (Jan. 19, 2023), https://www.wired.com/story/metaverse-video-games-fortnite-zuckerberg; see also Mitch Wagner, The Metaverse Hype Bubble Has Popped. What Now?, Fierce Electronics (Feb. 24, 2023), https://www.fierceelectronics.com/embedded/metaverse-hype-bubble-has-popped-what-now.

[46] Garret A. Johnson, et al., Privacy and Market Concentration: Intended and Unintended Consequences of the GDPR, Forthcoming Management Science 1 (2023).

[47] Jian Jia, et al., The Short-Run Effects of GDPR on Technology Venture Investment, NBER Working Paper 25248, 4 (2018), available at https://www.nber.org/system/files/working_papers/w25248/w25248.pdf.

[48] Samuel G. Goldberg, Garrett A. Johnson, & Scott K. Shriver, Regulating Privacy Online: An Economic Evaluation of GDPR (2021), available at https://www.ftc.gov/system/files/documents/public_events/1588356/johnsongoldbergshriver.pdf.

[49] Rebecca Janßen, Reinhold Kesler, Michael Kummer, & Joel Waldfogel, GDPR and the Lost Generation of Innovative Apps, Nber Working Paper 30028, 2 (2022), available at https://www.nber.org/system/files/working_papers/w30028/w30028.pdf.

[50] Rajkumar Venkatesan, S. Arunachalam & Kiran Pedada, Short Run Effects of Generalized Data Protection Act on Returns from AI Acquisitions, University of Virginia Working Paper 6 (2022), available at: https://conference.nber.org/conf_papers/f161612.pdf. (“On average, GDPR exposure reduces the ROA of firms. We also find that GDPR exposure increases the ROA of firms that make AI acquisitions for improving customer experience, and cybersecurity. Returns on AI investments in innovation and operational efficiencies are unaffected by GDPR.”)

[51] For a detailed discussion of the empirical literature concerning the GDPR, see Garrett Johnson, Economic Research on Privacy Regulation: Lessons From the GDPR And Beyond, NBER Working Paper 30705 (2022), available at https://www.nber.org/system/files/working_papers/w30705/w30705.pdf.

Continue reading
Antitrust & Consumer Protection

ICLE Comments to European Commission on AI Competition

Regulatory Comments Executive Summary We thank the European Commission for launching this consultation on competition in generative AI. The International Center for Law & Economics (“ICLE”) is . . .

Executive Summary

We thank the European Commission for launching this consultation on competition in generative AI. The International Center for Law & Economics (“ICLE”) is a nonprofit, nonpartisan global research and policy center founded with the goal of building the intellectual foundations for sensible, economically grounded policy. ICLE promotes the use of law & economics methodologies to inform public-policy debates and has longstanding expertise in the evaluation of competition law and policy. ICLE’s interest is to ensure that competition law remains grounded in clear rules, established precedent, a record of evidence, and sound economic analysis.

In our comments, we express concern that policymakers may equate the rapid rise of generative AI services with a need to intervene in these markets when, in fact, the opposite is true. As we explain, the rapid growth of AI markets, as well as the fact that new market players are thriving, suggests competition is intense. If incumbent firms could easily leverage their dominance into burgeoning generative AI markets, we would not have seen the growth of generative AI unicorns such as OpenAI, Midjourney, and Anthropic, to name but a few.

Of course, this is not to say that generative AI markets are not important—quite the opposite. Generative AI is already changing the ways that many firms do business and improving employee productivity in many industries.[1] The technology is also increasingly useful in the field of scientific research, where it has enabled creation of complex models that expand scientists’ reach.[2] Against this backdrop, Commissioner Margrethe Vestager was right to point out that it “is fundamental that these new markets stay competitive, and that nothing stands in the way of businesses growing and providing the best and most innovative products to consumers.”[3]

But while sensible enforcement is of vital importance to maintain competition and consumer welfare, knee-jerk reactions may yield the opposite outcomes. As our comments explain, overenforcement in the field of generative AI could cause the very harms that policymakers seek to avert. For instance, preventing so-called “big tech” firms from competing in these markets (for example, by threatening competition intervention as soon as they embed generative AI services in their ecosystems or seek to build strategic relationships with AI startups) may thwart an important source of competition needed to keep today’s leading generative-AI firms in check. In short, competition in AI markets is important, but trying naïvely to hold incumbent tech firms back out of misguided fears they will come to dominate this space is likely to do more harm than good.

Our comment proceeds as follows. Section I summarizes recent calls for competition intervention in generative AI markets. Section II argues that many of these calls are underpinned by fears of data-related incumbency advantages (often referred to as “data-network effects”). Section III explains why these effects are unlikely to play a meaningful role in generative-AI markets. Section IV concludes by offering five key takeaways to help policymakers (including the Commission) better weigh the tradeoffs inherent to competition intervention in generative-AI markets.

I. Calls for Intervention in AI Markets

It was once (and frequently) said that Google’s “data monopoly” was unassailable: “If ‘big data’ is the oil of the information economy, Google has Standard Oil-like monopoly dominance—and uses that control to maintain its dominant position.”[4] Similar claims of data dominance have been attached to nearly all large online platforms, including Facebook (Meta), Amazon, and Uber.[5]

While some of these claims continue even today (for example, “big data” is a key component of the U.S. Justice Department’s (DOJ) Google Search and adtech antitrust suits),[6] a shiny new data target has emerged in the form of generative artificial intelligence (AI). The launch of ChatGPT in November 2022, as well as the advent of AI image-generation services like Midjourney and Dall-E, have dramatically expanded the public’s conception of what is—and what might be—possible to achieve with generative-AI technologies built on massive datasets.

While these services remain in the early stages of mainstream adoption and remain in the throes of rapid, unpredictable technological evolution, they nevertheless already appear to be on the radar of competition policymakers around the world. Several antitrust enforcers appear to believe that, by acting now, they can avoid the “mistakes” that were purportedly made during the formative years of Web 2.0.[7] These mistakes, critics assert, include failing to appreciate the centrality of data in online markets, as well as letting mergers go unchecked and allowing early movers to entrench their market positions.[8] As Lina Khan, chair of the U.S. Federal Trade Commission (FTC), put it: “we are still reeling from the concentration that resulted from Web 2.0, and we don’t want to repeat the mis-steps of the past with AI”.[9]

This response from the competition-policy world is deeply troubling. Rather than engage in critical self-assessment and adopt an appropriately restrained stance, the enforcement community appears to be champing at the bit. Rather than assessing their prior assumptions based on the current technological moment, enforcers’ top priority appears to be figuring out how to rapidly and almost reflexively deploy existing competition tools to address the presumed competitive failures presented by generative AI.[10]

It is increasingly common for competition enforcers to argue that so-called “data-network effects” serve not only to entrench incumbents in those markets where the data is collected, but also confer similar, self-reinforcing benefits in adjacent markets. Several enforcers have, for example, prevented large online platforms from acquiring smaller firms in adjacent markets, citing the risk that they could use their vast access to data to extend their dominance into these new markets.[11]

They have also launched consultations to ascertain the role that data plays in AI competition. For instance, in an ongoing consultation, the European Commission asks: “What is the role of data and what are its relevant characteristics for the provision of generative AI systems and/or components, including AI models?”[12] Unsurprisingly, the FTC has likewise been bullish about the risks posed by incumbents’ access to data. In comments submitted to the U.S. Copyright Office, for example, the FTC argued that:

The rapid development and deployment of AI also poses potential risks to competition. The rising importance of AI to the economy may further lock in the market dominance of large incumbent technology firms. These powerful, vertically integrated incumbents control many of the inputs necessary for the effective development and deployment of AI tools, including cloud-based or local computing power and access to large stores of training data. These dominant technology companies may have the incentive to use their control over these inputs to unlawfully entrench their market positions in AI and related markets, including digital content markets.[13]

Certainly, it stands to reason that the largest online platforms—including Alphabet, Meta, Apple, and Amazon—should have a meaningful advantage in the burgeoning markets for generative-AI services. After all, it is widely recognized that data is an essential input for generative AI.[14] This competitive advantage should be all the more significant, given that these firms have been at the forefront of AI technology for more than a decade. Over this period, Google’s DeepMind and AlphaGo and Meta’s have routinely made headlines.[15] Apple and Amazon also have vast experience with AI assistants, and all of these firms use AI technology throughout their platforms.[16]

Contrary to what one might expect, however, the tech giants have, to date, been largely unable to leverage their vast data troves to outcompete startups like OpenAI and Midjourney. At the time of writing, OpenAI’s ChatGPT appears to be, by far, the most successful chatbot,[17] despite the large tech platforms’ apparent access to far more (and more up-to-date) data.

In these comments, we suggest that there are important lessons to glean from these developments, if only enforcers would stop to reflect. The meteoric rise of consumer-facing AI services should offer competition enforcers and policymakers an opportunity for introspection. As we explain, the rapid emergence of generative-AI technology may undercut many core assumptions of today’s competition-policy debates, which have largely focused on the rueful after-effects of the purported failure of 20th-century antitrust to address the allegedly manifest harms of 21st-century technology. These include the notions that data advantages constitute barriers to entry and can be leveraged to project dominance into adjacent markets; that scale itself is a market failure to be addressed by enforcers; and that the use of consumer data is inherently harmful to those consumers.

II. Data-Network Effects Theory and Enforcement

Proponents of tougher interventions by competition enforcers into digital markets often cite data-network effects as a source of competitive advantage and barrier to entry (though terms like “economies of scale and scope” may offer more precision).[18] The crux of the argument is that “the collection and use of data creates a feedback loop of more data, which ultimately insulates incumbent platforms from entrants who, but for their data disadvantage, might offer a better product.”[19] This self-reinforcing cycle purportedly leads to market domination by a single firm. Thus, it is argued, for example, that Google’s “ever-expanding control of user personal data, and that data’s critical value to online advertisers, creates an insurmountable barrier to entry for new competition.”[20]

Right off the bat, it is important to note the conceptual problem these claims face. Because data can be used to improve the quality of products and/or to subsidize their use, the idea of data as an entry barrier suggests that any product improvement or price reduction made by an incumbent could be a problematic entry barrier to any new entrant. This is tantamount to an argument that competition itself is a cognizable barrier to entry. Of course, it would be a curious approach to antitrust if competition were treated as a problem, as it would imply that firms should under-compete—i.e., should forego consumer-welfare enhancements—in order to inculcate a greater number of firms in a given market simply for its own sake.[21]

Meanwhile, actual economic studies of data-network effects have been few and far between, with scant empirical evidence to support the theory.[22] Andrei Hagiu and Julian Wright’s theoretical paper offers perhaps the most comprehensive treatment of the topic to date.[23] The authors ultimately conclude that data-network effects can be of different magnitudes and have varying effects on firms’ incumbency advantage.[24] They cite Grammarly (an AI writing-assistance tool) as a potential example: “As users make corrections to the suggestions offered by Grammarly, its language experts and artificial intelligence can use this feedback to continue to improve its future recommendations for all users.”[25]

This is echoed by other economists who contend that “[t]he algorithmic analysis of user data and information might increase incumbency advantages, creating lock-in effects among users and making them more reluctant to join an entrant platform.”[26] Crucially, some scholars take this logic a step further, arguing that platforms may use data from their “origin markets” in order to enter and dominate adjacent ones:

First, as we already mentioned, data collected in the origin market can be used, once the enveloper has entered the target market, to provide products more efficiently in the target market. Second, data collected in the origin market can be used to reduce the asymmetric information to which an entrant is typically subject when deciding to invest (for example, in R&D) to enter a new market. For instance, a search engine could be able to predict new trends from consumer searches and therefore face less uncertainty in product design.[27]

This possibility is also implicit in Hagiu and Wright’s paper.[28] Indeed, the authors’ theoretical model rests on an important distinction between within-user data advantages (that is, having access to more data about a given user) and across-user data advantages (information gleaned from having access to a wider user base). In both cases, there is an implicit assumption that platforms may use data from one service to gain an advantage in another market (because what matters is information about aggregate or individual user preferences, regardless of its origin).

Our review of the economic evidence suggests that several scholars have, with varying degrees of certainty, raised the possibility that incumbents may leverage data advantages to stifle competitors in their primary market or in adjacent ones (be it via merger or organic growth). As we explain below, however, there is ultimately little evidence to support such claims. Policymakers have, however, been keenly receptive to these limited theoretical findings, basing multiple decisions on these theories, often with little consideration given to the caveats that accompany them.[29]

Indeed, it is remarkable that, in its section on “[t]he data advantage for incumbents,” the “Furman Report” created for the UK government cited only two empirical economic studies, and they offer directly contradictory conclusions with respect to the question of the strength of data advantages.[30] Nevertheless, the Furman Report concludes that data “may confer a form of unmatchable advantage on the incumbent business, making successful rivalry less likely,”[31] and adopts without reservation “convincing” evidence from non-economists that have no apparent empirical basis.[32]

In the Google/Fitbit merger proceedings, the European Commission found that the combination of data from Google services with that of Fitbit devices would reduce competition in advertising markets:

Giving [sic] the large amount of data already used for advertising purposes that Google holds, the increase in Google’s data collection capabilities, which goes beyond the mere number of active users for which Fitbit has been collecting data so far, the Transaction is likely to have a negative impact on the development of an unfettered competition in the markets for online advertising.[33]

As a result, the Commission cleared the merger on the condition that Google refrain from using data from Fitbit devices for its advertising platform.[34] The Commission will likely focus on similar issues during its ongoing investigation of Microsoft’s investment into OpenAI.[35]

Along similar lines, the FTC’s complaint to enjoin Meta’s purchase of a virtual-reality (VR) fitness app called “Within” relied, among other things, on the fact that Meta could leverage its data about VR-user behavior to inform its decisions and potentially outcompete rival VR-fitness apps: “Meta’s control over the Quest platform also gives it unique access to VR user data, which it uses to inform strategic decisions.”[36]

The DOJ’s twin cases against Google also implicate data leveraging and data barriers to entry. The agency’s adtech complaint charges that “Google intentionally exploited its massive trove of user data to further entrench its monopoly across the digital advertising industry.”[37] Similarly, in its search complaint, the agency argues that:

Google’s anticompetitive practices are especially pernicious because they deny rivals scale to compete effectively. General search services, search advertising, and general search text advertising require complex algorithms that are constantly learning which organic results and ads best respond to user queries; the volume, variety, and velocity of data accelerates the automated learning of search and search advertising algorithms.[38]

Finally, updated merger guidelines published in recent years by several competition enforcers cite the acquisition of data as a potential source of competition concerns. For instance, the FTC and DOJ’s newly published guidelines state that “acquiring data that helps facilitate matching, sorting, or prediction services may enable the platform to weaken rival platforms by denying them that data.”[39] Likewise, the UK Competition and Markets Authority (CMA) warns against incumbents acquiring firms in order to obtain their data and foreclose other rivals:

Incentive to foreclose rivals…

7.19(e) Particularly in complex and dynamic markets, firms may not focus on short term margins but may pursue other objectives to maximise their long-run profitability, which the CMA may consider. This may include… obtaining access to customer data….[40]

In short, competition authorities around the globe have been taking an increasingly aggressive stance on data-network effects. Among the ways this has manifested is in basing enforcement decisions on fears that data collected by one platform might confer a decisive competitive advantage in adjacent markets. Unfortunately, these concerns rest on little to no empirical evidence, either in the economic literature or the underlying case records.

III. Data-Incumbency Advantages in Generative-AI Markets

Given the assertions canvassed in the previous section, it would be reasonable to assume that firms such as Google, Meta, and Amazon should be in pole position to dominate the burgeoning market for generative AI. After all, these firms have not only been at the forefront of the field for the better part of a decade, but they also have access to vast troves of data, the likes of which their rivals could only dream when they launched their own services. Thus, the authors of the Furman Report caution that “to the degree that the next technological revolution centres around artificial intelligence and machine learning, then the companies most able to take advantage of it may well be the existing large companies because of the importance of data for the successful use of these tools.”[41]

To date, however, this is not how things have unfolded—although it bears noting these markets remain in flux and the competitive landscape is susceptible to change. The first significantly successful generative-AI service was arguably not from either Meta—which had been working on chatbots for years and had access to, arguably, the world’s largest database of actual chats—or Google. Instead, the breakthrough came from a previously unknown firm called OpenAI.

OpenAI’s ChatGPT service currently holds an estimated 60% of the market (though reliable numbers are somewhat elusive).[42] It broke the record for the fastest online service to reach 100 million users (in only a couple of months), more than four times faster than the previous record holder, TikTok.[43] Based on Google Trends data, ChatGPT is nine times more popular worldwide than Google’s own Bard service, and 14 times more popular in the United States.[44] In April 2023, ChatGPT reportedly registered 206.7 million unique visitors, compared to 19.5 million for Google’s Bard.[45] In short, at the time we are writing, ChatGPT appears to be the most popular chatbot. The entry of large players such as Google Bard or Meta AI appear to have had little effect thus far on its market position.[46]

The picture is similar in the field of AI-image generation. As of August 2023, Midjourney, Dall-E, and Stable Diffusion appear to be the three market leaders in terms of user visits.[47] This is despite competition from the likes of Google and Meta, who arguably have access to unparalleled image and video databases by virtue of their primary platform activities.[48]

This raises several crucial questions: how have these AI upstarts managed to be so successful, and is their success just a flash in the pan before Web 2.0 giants catch up and overthrow them? While we cannot answer either of these questions dispositively, we offer what we believe to be some relevant observations concerning the role and value of data in digital markets.

A first important observation is that empirical studies suggest that data exhibits diminishing marginal returns. In other words, past a certain point, acquiring more data does not confer a meaningful edge to the acquiring firm. As Catherine Tucker put it following a review of the literature: “Empirically there is little evidence of economies of scale and scope in digital data in the instances where one would expect to find them.”[49]

Likewise, following a survey of the empirical literature on this topic, Geoffrey Manne and Dirk Auer conclude that:

Available evidence suggests that claims of “extreme” returns to scale in the tech sector are greatly overblown. Not only are the largest expenditures of digital platforms unlikely to become proportionally less important as output increases, but empirical research strongly suggests that even data does not give rise to increasing returns to scale, despite routinely being cited as the source of this effect.[50]

In other words, being the firm with the most data appears to be far less important than having enough data. This lower bar may be accessible to far more firms than one might initially think possible. And obtaining enough data could become even easier—that is, the volume of required data could become even smaller—with technological progress. For instance, synthetic data may provide an adequate substitute to real-world data,[51] or may even outperform real-world data.[52] As Thibault Schrepel and Alex Pentland surmise:

[A]dvances in computer science and analytics are making the amount of data less relevant every day. In recent months, important technological advances have allowed companies with small data sets to compete with larger ones.[53]

Indeed, past a certain threshold, acquiring more data might not meaningfully improve a service, where other improvements (such as better training methods or data curation) could have a large impact. In fact, there is some evidence that excessive data impedes a service’s ability to generate results appropriate for a given query: “[S]uperior model performance can often be achieved with smaller, high-quality datasets than massive, uncurated ones. Data curation ensures that training datasets are devoid of noise, irrelevant instances, and duplications, thus maximizing the efficiency of every training iteration.”[54]

Consider, for instance, a user who wants to generate an image of a basketball. Using a model trained on an indiscriminate range and number of public photos in which a basketball appears surrounded by copious other image data, the user may end up with an inordinately noisy result. By contrast, a model trained with a better method on fewer, more carefully selected images, could readily yield far superior results.[55] In one important example:

[t]he model’s performance is particularly remarkable, given its small size. “This is not a large language model trained on the whole Internet; this is a relatively small transformer trained for these tasks,” says Armando Solar-Lezama, a computer scientist at the Massachusetts Institute of Technology, who was not involved in the new study…. The finding implies that instead of just shoving ever more training data into machine-learning models, a complementary strategy might be to offer AI algorithms the equivalent of a focused linguistics or algebra class.[56]

Platforms’ current efforts are thus focused on improving the mathematical and logical reasoning of large language models (LLMs), rather than maximizing training datasets.[57] Two points stand out. The first is that firms like OpenAI rely largely on publicly available datasets—such as GSM8K—to train their LLMs.[58] Second, the real challenge to create cutting-edge AI is not so much in collecting data, but rather in creating innovative AI-training processes and architectures:

[B]uilding a truly general reasoning engine will require a more fundamental architectural innovation. What’s needed is a way for language models to learn new abstractions that go beyond their training data and have these evolving abstractions influence the model’s choices as it explores the space of possible solutions.

We know this is possible because the human brain does it. But it might be a while before OpenAI, DeepMind, or anyone else figures out how to do it in silicon.[59]

Furthermore, it is worth noting that the data most relevant to startups in a given market may not be those data held by large incumbent platforms in other markets, but rather data specific to the market in which the startup is active or, even better, to the given problem it is attempting to solve:

As Andres Lerner has argued, if you wanted to start a travel business, the data from Kayak or Priceline would be far more relevant. Or if you wanted to start a ride-sharing business, data from cab companies would be more useful than the broad, market-cross-cutting profiles Google and Facebook have. Consider companies like Uber, Lyft and Sidecar that had no customer data when they began to challenge established cab companies that did possess such data. If data were really so significant, they could never have competed successfully. But Uber, Lyft and Sidecar have been able to effectively compete because they built products that users wanted to use—they came up with an idea for a better mousetrap. The data they have accrued came after they innovated, entered the market and mounted their successful challenges—not before.[60]

The bottom line is that data is not the be-all and end-all that many in competition circles make it out to be. While data may often confer marginal benefits, there is little sense these are ultimately decisive.[61] As a result, incumbent platforms’ access to vast numbers of users and data in their primary markets might only marginally affect their AI competitiveness.

A related observation is that firms’ capabilities and other features of their products arguably play a more important role than the data they own.[62] Examples of this abound in digital markets. Google overthrew Yahoo, despite initially having access to far fewer users and far less data; Google and Apple overcame Microsoft in the smartphone OS market despite having comparatively tiny ecosystems (at the time) to leverage; and TikTok rose to prominence despite intense competition from incumbents like Instagram, which had much larger user bases. In each of these cases, important product-design decisions (such as the PageRank algorithm, recognizing the specific needs of mobile users,[63] and TikTok’s clever algorithm) appear to have played a far more significant role than initial user and data endowments (or lack thereof).

All of this suggests that the early success of OpenAI likely has more to do with its engineering decisions than what data it did (or did not) own. Going forward, OpenAI and its rivals’ ability to offer and monetize compelling stores offering custom versions of their generative-AI technology will arguably play a much larger role than (and contribute to) their ownership of data.[64] In other words, the ultimate challenge is arguably to create a valuable platform, of which data ownership is a consequence, but not a cause.

It is also important to note that, in those instances where it is valuable, data does not just fall from the sky. Instead, it is through smart business and engineering decisions that firms can generate valuable information (which does not necessarily correlate with owning more data).

For instance, OpenAI’s success with ChatGPT is often attributed to its more efficient algorithms and training models, which arguably have enabled the service to improve more rapidly than its rivals.[65] Likewise, the ability of firms like Meta and Google to generate valuable data for advertising arguably depends more on design decisions that elicit the right data from users, rather than the raw number of users in their networks.

Put differently, setting up a business so as to extract and organize the right information is more important than simply owning vast troves of data.[66] Even in those instances where high-quality data is an essential parameter of competition, it does not follow that having vaster databases or more users on a platform necessarily leads to better information for the platform.

Indeed, if data ownership consistently conferred a significant competitive advantage, these new firms would not be where they are today. This does not mean that data is worthless, of course. Rather, it means that competition authorities should not assume that the mere possession of data is a dispositive competitive advantage, absent compelling empirical evidence to support such a finding. In this light, the current wave of decisions and competition-policy pronouncements that rely on data-related theories of harm are premature.

IV. Five Key Takeaways: Reconceptualizing the Role of Data in Generative-AI Competition

As we explain above, data (network effects) are not the source of barriers to entry that they are sometimes made out to be. The picture is far more nuanced. Indeed, as economist Andres Lerner demonstrated almost a decade ago (and the assessment is only truer today):

Although the collection of user data is generally valuable for online providers, the conclusion that such benefits of user data lead to significant returns to scale and to the entrenchment of dominant online platforms is based on unsupported assumptions. Although, in theory, control of an “essential” input can lead to the exclusion of rivals, a careful analysis of real-world evidence indicates that such concerns are unwarranted for many online businesses that have been the focus of the “big data” debate.[67]

While data can be an important part of the competitive landscape, incumbents’ data advantages are far less pronounced than today’s policymakers commonly assume. In that respect, five main lessons emerge:

  1. Data can be (very) valuable, but beyond a certain threshold, those benefits tend to diminish. In other words, having the most data is less important than having enough;
  2. The ability to generate valuable information does not depend on the number of users or the amount of data a platform has previously acquired;
  3. The most important datasets are not always proprietary;
  4. Technological advances and platforms’ engineering decisions affect their ability to generate valuable information, and this effect swamps effects stemming from the amount of data they own; and
  5. How platforms use data is arguably more important than what data or how much data they own.

These lessons have important ramifications for competition-policy debates over the competitive implications of data in technologically evolving areas.

First, it is not surprising that startups, rather than incumbents, have taken an early lead in generative AI (and in Web 2.0 before it). After all, if data-incumbency advantages are small or even nonexistent, then smaller and more nimble players may have an edge over established tech platforms. This is all the more likely given that, despite significant efforts, the biggest tech platforms were unable to offer compelling generative-AI chatbots and image-generation services before the emergence of ChatGPT, Dall-E, Midjourney, etc.

This failure suggests that, in a process akin to Clayton Christensen’s “innovator’s dilemma,”[68] something about the incumbent platforms’ existing services and capabilities was holding them back in those markets. Of course, this does not necessarily mean that those same services or capabilities could not become an advantage when the generative-AI market starts addressing issues of monetization and scale.[69] But it does mean that assumptions about a firm’s market power based on its possession of data are off the mark.

Another important implication is that, paradoxically, policymakers’ efforts to prevent Web 2.0 platforms from competing freely in generative AI markets may ultimately backfire and lead to less, not more, competition. Indeed, OpenAI is currently acquiring a sizeable lead in generative AI. While competition authorities might like to think that other startups will emerge and thrive in this space, it is important not to confuse desires with reality. While there currently exists a vibrant AI-startup ecosystem, there is at least a case to be made that the most significant competition for today’s AI leaders will come from incumbent Web 2.0 platforms—although nothing is certain at this stage. Policymakers should beware not to stifle that competition on the misguided assumption that competitive pressure from large incumbents is somehow less valuable to consumers than that which originates from smaller firms.

Finally, even if there were a competition-related market failure to be addressed in the field of generative AI (which is anything but clear), it is unclear that the remedies being contemplated would do more good than harm. Some of the solutions that have been put forward have highly ambiguous effects on consumer welfare. Scholars have shown that, e.g., mandated data sharing—a solution championed by EU policymakers, among others—may sometimes dampen competition in generative-AI markets.[70] This is also true of legislation like the General Data Protection Regulation (GDPR), which makes it harder for firms to acquire more data about consumers—assuming such data is, indeed, useful to generative-AI services.[71]

In sum, it is a flawed understanding of the economics and practical consequences of large agglomerations of data that lead competition authorities to believe that data-incumbency advantages are likely to harm competition in generative AI markets—or even in the data-intensive Web 2.0 markets that preceded them. Indeed, competition or regulatory intervention to “correct” data barriers and data network and scale effects is liable to do more harm than good.

 

[1] See, e.g., Michael Chui, et al., The Economic Potential of Generative AI: The Next Productivity Frontier, McKinsey (Jun. 14, 2023), https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights/the-economic-potential-of-generative-AI-the-next-productivity-frontier.

[2] See, e. g., Zhuoran Qiao, Weili Nie, Arash Vahdat, Thomas F. Miller III, & Animashree Anandkumar, State-Specific Protein–Ligand Complex Structure Prediction with a Multiscale Deep Generative Model, 6 Nature Machine Intelligence, 195-208 (2024); see also, Jaemin Seo, Sang Kyeun Kim, Azarakhsh Jalalvand, Rory Conlin, Andrew Rothstein, Joseph Abbate, Keith Erickson, Josiah Wai, Ricardo Shousha, & Egemen Kolemen, Avoiding Fusion Plasma Tearing Instability with Deep Reinforcement Learning, 626 Nature, 746-751 (2024).

[3] See, e.g., Press Release, Commission Launches Calls for Contributions on Competition in Virtual Worlds and Generative AI, European Commission (Jan. 9, 2024), https://ec.europa.eu/commission/presscorner/detail/en/IP_24_85.

[4] Nathan Newman, Taking on Google’s Monopoly Means Regulating Its Control of User Data, Huffington Post (Sep. 24, 2013), http://www.huffingtonpost.com/nathan-newman/taking-on-googlesmonopol_b_3980799.html.

[5] See, e.g., Lina Khan & K. Sabeel Rahman, Restoring Competition in the U.S. Economy, in Untamed: How to Check Corporate, Financial, and Monopoly Power (Nell Abernathy, Mike Konczal, & Kathryn Milani, eds., 2016), at 23 (“From Amazon to Google to Uber, there is a new form of economic power on display, distinct from conventional monopolies and oligopolies…, leverag[ing] data, algorithms, and internet-based technologies… in ways that could operate invisibly and anticompetitively.”); Mark Weinstein, I Changed My Mind—Facebook Is a Monopoly, Wall St. J. (Oct. 1, 2021), https://www.wsj.com/articles/facebook-is-monopoly-metaverse-users-advertising-platforms-competition-mewe-big-tech-11633104247 (“[T]he glue that holds it all together is Facebook’s monopoly over data…. Facebook’s data troves give it unrivaled knowledge about people, governments—and its competitors.”).

[6] See, generally, Abigail Slater, Why “Big Data” Is a Big Deal, The Reg. Rev. (Nov. 6, 2023), https://www.theregreview.org/2023/11/06/slater-why-big-data-is-a-big-deal; Amended Complaint at ¶36, United States v. Google, 1:20-cv-03010- (D.D.C. 2020); Complaint at ¶37, United States V. Google, 1:23-cv-00108 (E.D. Va. 2023), https://www.justice.gov/opa/pr/justice-department-sues-google-monopolizing-digital-advertising-technologies (“Google intentionally exploited its massive trove of user data to further entrench its monopoly across the digital advertising industry.”).

[7] See, e.g., Press Release, European Commission, supra note 3; Krysten Crawford, FTC’s Lina Khan Warns Big Tech over AI, SIEPR (Nov. 3, 2020), https://siepr.stanford.edu/news/ftcs-lina-khan-warns-big-tech-over-ai (“Federal Trade Commission Chair Lina Khan delivered a sharp warning to the technology industry in a speech at Stanford on Thursday: Antitrust enforcers are watching what you do in the race to profit from artificial intelligence.”) (emphasis added).

[8] See, e.g., John M. Newman, Antitrust in Digital Markets, 72 Vand. L. Rev. 1497, 1501 (2019) (“[T]he status quo has frequently failed in this vital area, and it continues to do so with alarming regularity. The laissez-faire approach advocated for by scholars and adopted by courts and enforcers has allowed potentially massive harms to go unchecked.”);
Bertin Martins, Are New EU Data Market Regulations Coherent and Efficient?, Bruegel Working Paper 21/23 (2023), https://www.bruegel.org/working-paper/are-new-eu-data-market-regulations-coherent-and-efficient (“Technical restrictions on access to and re-use of data may result in failures in data markets and data-driven services markets.”); Valéria Faure-Muntian, Competitive Dysfunction: Why Competition Law Is Failing in a Digital World, The Forum Network (Feb. 24, 2021), https://www.oecd-forum.org/posts/competitive-dysfunction-why-competition-law-is-failing-in-a-digital-world.

[9] See Rana Foroohar, The Great US-Europe Antitrust Divide, FT (Feb. 5, 2024), https://www.ft.com/content/065a2f93-dc1e-410c-ba9d-73c930cedc14.

[10] See, e.g., Press Release, European Commission, supra note 3.

[11] See infra, Section II. Commentators have also made similar claims; see, e.g., Ganesh Sitaram & Tejas N. Narechania, It’s Time for the Government to Regulate AI. Here’s How, Politico (Jan. 15, 2024) (“All that cloud computing power is used to train foundation models by having them “learn” from incomprehensibly huge quantities of data. Unsurprisingly, the entities that own these massive computing resources are also the companies that dominate model development. Google has Bard, Meta has LLaMa. Amazon recently invested $4 billion into one of OpenAI’s leading competitors, Anthropic. And Microsoft has a 49 percent ownership stake in OpenAI — giving it extraordinary influence, as the recent board struggles over Sam Altman’s role as CEO showed.”).

[12] Press Release, European Commission, supra note 3.

[13] Comment of U.S. Federal Trade Commission to the U.S. Copyright Office, Artificial Intelligence and Copyright, Docket No. 2023-6 (Oct. 30, 2023), at 4, https://www.ftc.gov/legal-library/browse/advocacy-filings/comment-federal-trade-commission-artificial-intelligence-copyright (emphasis added).

[14] See, e.g. Joe Caserta, Holger Harreis, Kayvaun Rowshankish, Nikhil Srinidhi, & Asin Tavakoli, The Data Dividend: Fueling Generative AI, McKinsey Digital (Sep. 15, 2023), https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights/the-data-dividend-fueling-generative-ai (“Your data and its underlying foundations are the determining factors to what’s possible with generative AI.”).

[15] See, e.g., Tim Keary, Google DeepMind’s Achievements and Breakthroughs in AI Research, Techopedia (Aug. 11, 2023), https://www.techopedia.com/google-deepminds-achievements-and-breakthroughs-in-ai-research; See, e.g., Will Douglas Heaven, Google DeepMind Used a Large Language Model to Solve an Unsolved Math Problem, MIT Technology Review (Dec. 14, 2023), https://www.technologyreview.com/2023/12/14/1085318/google-deepmind-large-language-model-solve-unsolvable-math-problem-cap-set; see also, A Decade of Advancing the State-of-the-Art in AI Through Open Research, Meta (Nov. 30, 2023), https://about.fb.com/news/2023/11/decade-of-advancing-ai-through-open-research; see also, 200 Languages Within a Single AI Model: A Breakthrough in High-Quality Machine Translation, Meta, https://ai.meta.com/blog/nllb-200-high-quality-machine-translation (last visited Jan. 18, 2023).

[16] See, e.g., Jennifer Allen, 10 Years of Siri: The History of Apple’s Voice Assistant, Tech Radar (Oct. 4, 2021), https://www.techradar.com/news/siri-10-year-anniversary; see also Evan Selleck, How Apple Is Already Using Machine Learning and AI in iOS, Apple Insider (Nov. 20, 2023), https://appleinsider.com/articles/23/09/02/how-apple-is-already-using-machine-learning-and-ai-in-ios; see also, Kathleen Walch, The Twenty Year History Of AI At Amazon, Forbes (July 19, 2019), https://www.forbes.com/sites/cognitiveworld/2019/07/19/the-twenty-year-history-of-ai-at-amazon.

[17] See infra Section III.

[18] See, e.g., Cédric Argenton & Jens Prüfer, Search Engine Competition with Network Externalities, 8 J. Comp. L. & Econ. 73, 74 (2012).

[19] John M. Yun, The Role of Big Data in Antitrust, in The Global Antitrust Institute Report on the Digital Economy (Joshua D. Wright & Douglas H. Ginsburg, eds., Nov. 11, 2020) at 233, https://gaidigitalreport.com/2020/08/25/big-data-and-barriers-to-entry/#_ftnref50; see also, e.g., Robert Wayne Gregory, Ola Henfridsson, Evgeny Kaganer, & Harris Kyriakou, The Role of Artificial Intelligence and Data Network Effects for Creating User Value, 46 Acad. of Mgmt. Rev. 534 (2020), final pre-print version at 4, http://wrap.warwick.ac.uk/134220) (“A platform exhibits data network effects if, the more that the platform learns from the data it collects on users, the more valuable the platform becomes to each user.”); see also, Karl Schmedders, José Parra-Moyano, & Michael Wade, Why Data Aggregation Laws Could be the Answer to Big Tech Dominance, Silicon Republic (Feb. 6, 2024), https://www.siliconrepublic.com/enterprise/data-ai-aggregation-laws-regulation-big-tech-dominance-competition-antitrust-imd.

[20] Nathan Newman, Search, Antitrust, and the Economics of the Control of User Data, 31 Yale J. Reg. 401, 409 (2014) (emphasis added); see also id. at 420 & 423 (“While there are a number of network effects that come into play with Google, [“its intimate knowledge of its users contained in its vast databases of user personal data”] is likely the most important one in terms of entrenching the company’s monopoly in search advertising…. Google’s overwhelming control of user data… might make its dominance nearly unchallengeable.”).

[21] See also Yun, supra note 19 at 229 (“[I]nvestments in big data can create competitive distance between a firm and its rivals, including potential entrants, but this distance is the result of a competitive desire to improve one’s product.”).

[22] For a review of the literature on increasing returns to scale in data (this topic is broader than data-network effects) see Geoffrey Manne & Dirk Auer, Antitrust Dystopia and Antitrust Nostalgia: Alarmist Theories of Harm in Digital Markets and Their Origins, 28 Geo Mason L. Rev. 1281, 1344 (2021).

[23] Andrei Hagiu & Julian Wright, Data-Enabled Learning, Network Effects, and Competitive Advantage, 54 RAND J. Econ. 638 (2023).

[24] Id. at 639. The authors conclude that “Data-enabled learning would seem to give incumbent firms a competitive advantage. But how strong is this advantage and how does it differ from that obtained from more traditional mechanisms…”.

[25] Id.

[26] Bruno Jullien & Wilfried Sand-Zantman, The Economics of Platforms: A Theory Guide for Competition Policy, 54 Info. Econ. & Pol’y 10080, 101031 (2021).

[27] Daniele Condorelli & Jorge Padilla, Harnessing Platform Envelopment in the Digital World, 16 J. Comp. L. & Pol’y 143, 167 (2020).

[28] See Hagiu & Wright, supra note 23.

[29] For a summary of these limitations, see generally Catherine Tucker, Network Effects and Market Power: What Have We Learned in the Last Decade?, Antitrust (2018) at 72, available at https://sites.bu.edu/tpri/files/2018/07/tucker-network-effects-antitrust2018.pdf; see also Manne & Auer, supra note 22, at 1330.

[30] See Jason Furman, Diane Coyle, Amelia Fletcher, Derek McAuley, & Philip Marsden (Dig. Competition Expert Panel), Unlocking Digital Competition (2019) at 32-35 (“Furman Report”), available at https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/file/785547/unlocking_digital_competition_furman_review_web.pdf.

[31] Id. at 34.

[32] Id. at 35. To its credit, it should be noted, the Furman Report does counsel caution before mandating access to data as a remedy to promote competition. See id. at 75. That said, the Furman Report does maintain that such a remedy should certainly be on the table because “the evidence suggests that large data holdings are at the heart of the potential for some platform markets to be dominated by single players and for that dominance to be entrenched in a way that lessens the potential for competition for the market.” Id. In fact, the evidence does not show this.

[33] Case COMP/M.9660 — Google/Fitbit, Commission Decision (Dec. 17, 2020) (Summary at O.J. (C 194) 7), available at https://ec.europa.eu/competition/mergers/cases1/202120/m9660_3314_3.pdf at 455.

[34] Id. at 896.

[35] See Natasha Lomas, EU Checking if Microsoft’s OpenAI Investment Falls Under Merger Rules, TechCrunch (Jan. 9, 2024), https://techcrunch.com/2024/01/09/openai-microsoft-eu-merger-rules.

[36] Amended Complaint at 11, Meta/Zuckerberg/Within, Fed. Trade Comm’n. (2022) (No. 605837), available at https://www.ftc.gov/system/files/ftc_gov/pdf/D09411%20-%20AMENDED%20COMPLAINT%20FILED%20BY%20COUNSEL%20SUPPORTING%20THE%20COMPLAINT%20-%20PUBLIC%20%281%29_0.pdf.

[37] Amended Complaint (D.D.C), supra note 6 at ¶37.

[38] Amended Complaint (E.D. Va), supra note 6 at ¶8.

[39] Merger Guidelines, US Dep’t of Justice & Fed. Trade Comm’n (2023) at 25, available at https://www.ftc.gov/system/files/ftc_gov/pdf/2023_merger_guidelines_final_12.18.2023.pdf.

[40] Merger Assessment Guidelines, Competition and Mkts. Auth (2021) at  ¶7.19(e), available at https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/file/1051823/MAGs_for_publication_2021_–_.pdf.

[41] Furman Report, supra note 30, at ¶4.

[42] See, e.g., Chris Westfall, New Research Shows ChatGPT Reigns Supreme in AI Tool Sector, Forbes (Nov. 16, 2023), https://www.forbes.com/sites/chriswestfall/2023/11/16/new-research-shows-chatgpt-reigns-supreme-in-ai-tool-sector/?sh=7de5de250e9c.

[43] See Krystal Hu, ChatGPT Sets Record for Fastest-Growing User Base, Reuters (Feb. 2, 2023), https://www.reuters.com/technology/chatgpt-sets-record-fastest-growing-user-base-analyst-note-2023-02-01; Google: The AI Race Is On, App Economy Insights (Feb. 7, 2023), https://www.appeconomyinsights.com/p/google-the-ai-race-is-on.

[44] See Google Trends, https://trends.google.com/trends/explore?date=today%205-y&q=%2Fg%2F11khcfz0y2,%2Fg%2F11ts49p01g&hl=en (last visited, Jan. 12, 2024) and https://trends.google.com/trends/explore?date=today%205-y&geo=US&q=%2Fg%2F11khcfz0y2,%2Fg%2F11ts49p01g&hl=en (last visited Jan. 12, 2024).

[45] See David F. Carr, As ChatGPT Growth Flattened in May, Google Bard Rose 187%, Similarweb Blog (Jun. 5, 2023), https://www.similarweb.com/blog/insights/ai-news/chatgpt-bard.

[46] See Press Release, Introducing New AI Experiences Across Our Family of Apps and Devices, Meta (Sep. 27, 2023), https://about.fb.com/news/2023/09/introducing-ai-powered-assistants-characters-and-creative-tools; Sundar Pichai, An Important Next Step on Our AI Journey, Google Keyword Blog (Feb. 6, 2023), https://blog.google/technology/ai/bard-google-ai-search-updates.

[47] See Ion Prodan, 14 Million Users: Midjourney’s Statistical Success, Yon (Aug. 19, 2023), https://yon.fun/midjourney-statistics; see also Andrew Wilson, Midjourney Statistics: Users, Polls, & Growth [Oct 2023], ApproachableAI (Oct. 13, 2023), https://approachableai.com/midjourney-statistics.

[48] See Hema Budaraju, New Ways to Get Inspired with Generative AI in Search, Google Keyword Blog (Oct. 12, 2023), https://blog.google/products/search/google-search-generative-ai-october-update; Imagine with Meta AI, Meta (last visited Jan. 12, 2024), https://imagine.meta.com.

[49] Catherine Tucker, Digital Data, Platforms and the Usual [Antitrust] Suspects: Network Effects, Switching Costs, Essential Facility, 54 Rev. Indus. Org. 683, 686 (2019).

[50] Manne & Auer, supra note 22, at 1345.

[51] See, e.g., Stefanie Koperniak, Artificial Data Give the Same Results as Real Data—Without Compromising Privacy, MIT News (Mar. 3, 2017), https://news.mit.edu/2017/artificial-data-give-same-results-as-real-data-0303 (“[Authors] describe a machine learning system that automatically creates synthetic data—with the goal of enabling data science efforts that, due to a lack of access to real data, may have otherwise not left the ground. While the use of authentic data can cause significant privacy concerns, this synthetic data is completely different from that produced by real users—but can still be used to develop and test data science algorithms and models.”).

[52] See, e.g., Rachel Gordon, Synthetic Imagery Sets New Bar in AI Training Efficiency, MIT News (Nov. 20, 2023), https://news.mit.edu/2023/synthetic-imagery-sets-new-bar-ai-training-efficiency-1120 (“By using synthetic images to train machine learning models, a team of scientists recently surpassed results obtained from traditional ‘real-image’ training methods.).

[53] Thibault Schrepel & Alex ‘Sandy’ Pentland, Competition Between AI Foundation Models: Dynamics and Policy Recommendations, MIT Connection Science Working Paper (Jun. 2023), at 8.

[54] Igor Susmelj, Optimizing Generative AI: The Role of Data Curation, Lightly (last visited Jan. 15, 2024), https://www.lightly.ai/post/optimizing-generative-ai-the-role-of-data-curation.

[55] See, e.g., Xiaoliang Dai, et al., Emu: Enhancing Image Generation Models Using Photogenic Needles in a Haystack, ArXiv (Sep. 27, 2023) at 1, https://ar5iv.labs.arxiv.org/html/2309.15807 (“[S]upervised fine-tuning with a set of surprisingly small but extremely visually appealing images can significantly improve the generation quality.”); see also, Hu Xu, et al., Demystifying CLIP Data, ArXiv (Sep. 28, 2023), https://arxiv.org/abs/2309.16671.

[56] Lauren Leffer, New Training Method Helps AI Generalize like People Do, Sci. Am. (Oct. 26, 2023), https://www.scientificamerican.com/article/new-training-method-helps-ai-generalize-like-people-do (discussing Brendan M. Lake & Marco Baroni, Human-Like Systematic Generalization Through a Meta-Learning Neural Network, 623 Nature 115 (2023)).

[57] Timothy B. Lee, The Real Research Behind the Wild Rumors about OpenAI’s Q* Project, Ars Technica (Dec. 8, 2023), https://arstechnica.com/ai/2023/12/the-real-research-behind-the-wild-rumors-about-openais-q-project.

[58] Id.; see also GSM8K, Papers with Code (last visited Jan. 18, 2023), available at https://paperswithcode.com/dataset/gsm8k; MATH Dataset, GitHub (last visited Jan. 18, 2024), available at https://github.com/hendrycks/math.

[59] Lee, supra note 57.

[60] Geoffrey Manne & Ben Sperry, Debunking the Myth of a Data Barrier to Entry for Online Services, Truth on the Market (Mar. 26, 2015), https://truthonthemarket.com/2015/03/26/debunking-the-myth-of-a-data-barrier-to-entry-for-online-services (citing Andres V. Lerner, The Role of ‘Big Data’ in Online Platform Competition (Aug. 26, 2014), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2482780.).

[61] See Catherine Tucker, Digital Data as an Essential Facility: Control, CPI Antitrust Chron. (Feb. 2020), at 11 (“[U]ltimately the value of data is not the raw manifestation of the data itself, but the ability of a firm to use this data as an input to insight.”).

[62] Or, as John Yun puts it, data is only a small component of digital firms’ production function. See Yun, supra note 19, at 235 (“Second, while no one would seriously dispute that having more data is better than having less, the idea of a data-driven network effect is focused too narrowly on a single factor improving quality. As mentioned in supra Section I.A, there are a variety of factors that enter a firm’s production function to improve quality.”).

[63] Luxia Le, The Real Reason Windows Phone Failed Spectacularly, History–Computer (Aug. 8, 2023), https://history-computer.com/the-real-reason-windows-phone-failed-spectacularly.

[64] Introducing the GPT Store, Open AI (Jan. 10, 2024), https://openai.com/blog/introducing-the-gpt-store.

[65] See Michael Schade, How ChatGPT and Our Language Models are Developed, OpenAI, https://help.openai.com/en/articles/7842364-how-chatgpt-and-our-language-models-are-developed; Sreejani Bhattacharyya, Interesting Innovations from OpenAI in 2021, AIM (Jan. 1, 2022), https://analyticsindiamag.com/interesting-innovations-from-openai-in-2021; Danny Hernadez & Tom B. Brown, Measuring the Algorithmic Efficiency of Neural Networks, ArXiv (May 8, 2020), https://arxiv.org/abs/2005.04305.

[66] See Yun, supra note 19 at 235 (“Even if data is primarily responsible for a platform’s quality improvements, these improvements do not simply materialize with the presence of more data—which differentiates the idea of data-driven network effects from direct network effects. A firm needs to intentionally transform raw, collected data into something that provides analytical insights. This transformation involves costs including those associated with data storage, organization, and analytics, which moves the idea of collecting more data away from a strict network effect to more of a ‘data opportunity.’”).

[67] Lerner, supra note 60, at 4-5 (emphasis added).

[68] See Clayton M. Christensen, The Innovator’s Dilemma: When New Technologies Cause Great Firms to Fail (2013).

[69] See David J. Teece, Dynamic Capabilities and Strategic Management: Organizing for Innovation and Growth (2009).

[70] See Hagiu & Wright, supra note 23, at 23 (“We use our dynamic framework to explore how data sharing works: we find that it in-creases consumer surplus when one firm is sufficiently far ahead of the other by making the laggard more competitive, but it decreases consumer surplus when the firms are sufficiently evenly matched by making firms compete less aggressively, which in our model means subsidizing consumers less.”); see also Lerner, supra note 60.

[71] See, e.g., Hagiu & Wright, id. (“We also use our model to highlight an unintended consequence of privacy policies. If such policies reduce the rate at which firms can extract useful data from consumers, they will tend to increase the incumbent’s competitive advantage, reflecting that the entrant has more scope for new learning and so is affected more by such a policy.”); Jian Jia, Ginger Zhe Jin, & Liad Wagman, The Short-Run Effects of the General Data Protection Regulation on Technology Venture Investment, 40 Marketing Sci. 593 (2021) (finding GDPR reduced investment in new and emerging technology firms, particularly in data-related ventures); James Campbell, Avi Goldfarb, & Catherine Tucker, Privacy Regulation and Market Structure, 24 J. Econ. & Mgmt. Strat. 47 (2015) (“Consequently, rather than increasing competition, the nature of transaction costs implied by privacy regulation suggests that privacy regulation may be anti-competitive.”).

Continue reading
Antitrust & Consumer Protection