https://www.cato.org/blog/financial-regulators-shouldnt-treat-open-source-software-enemy

Jack Solowey

Washington policymakers are consumed with AI concern. Fears run the gamut from existential threats to humanity to chatbots fibbing. In recent weeks, AI entrepreneurs and policy thinkers have helped to frame one of AI’s principle risks as the possible threat posed to political stability and continuity. In a thoughtful multipart series on “AI and Leviathan,” for example, Samuel Hammond (senior economist at the Foundation for American Innovation) argues that “[d]emocratized AI is a much greater regime change threat than the internet” and “[t]he moment governments realize that AI is a threat to their sovereignty, they will be tempted to clamp down in a totalitarian fashion.”

It’s wise to expect that the prospect of dizzying changes threatening the established order will incline states toward aggressive counterreactions. Indeed, we already see early signs of this in financial regulators’ response to autonomous and self‐​executing financial tools (e.g., smart contracts on cryptocurrency blockchains). Notably, smart contracts and certain AI models share a common feature that, when paired with the ability to operate with limited human intervention, can be particularly disruptive to existing regulatory methods: open‐​source code that is freely reproducible.

Even if open‐​source AI models constitute the minority of key foundation models, the fact that enough relatively advanced AI models are readily copyable (not to mention portable and storable) poses a clear challenge to governments looking to exert control over AI. Consequently, there’s an emerging policy battle over the desirability of open‐​source AI.

Unfortunately, financial regulators have led the way in cracking down on novel, open‐​source technologies. In doing so, they risk creating dangerous precedents for the use of open‐​source software—AI-based and otherwise—in both financial applications and in tech innovation more broadly. Before continuing further down this fraught path, policymakers must carefully consider the potential benefits of open‐​source software development that will be lost to knee‐​jerk policy reactions.

Fundamentally, open‐​source software is an intellectual property question: whether the code’s authors will license the free use, copying, modification, and distribution of their software without the need to seek those authors’ permission (the authors themselves typically disclaim liability in the process). Open‐​source licenses facilitate creatively remixing software, as well as ecosystems that foster iterative improvements.

Importantly, open‐​source licenses also give code something of a life unto itself, as it can continue—through the work of developer communities—to evolve and multiply beyond the reach of the original authors.

Open‐​source software therefore can pose a challenge to government agencies accustomed to regulating products and services by regulating their providers. If the government has a problem with OpenAI’s software, they haul in OpenAI. But if they have a problem with any of the tens of thousands of open‐​source AI models, who (or what) gets named and blamed?

Open‐​source AI critics fear that averting and remediating any harms associated with AI models will be seriously hampered by a lack of namable and blamable developers with end‐​to‐​end control over the code or, in the financial services context, human professionals holding out a shingle and visibly shouldering a fiduciary duty. Yet others take precisely the opposite position on open‐​source AI, arguing that the ability to freely use, modify, and distribute AI models will be essential to tackling AI “safety” and fallibility problems. One way in which this could play out is open‐​source licenses simply allowing more minds to work on these challenges and, in turn, make the fruits of their research freely available.

Notwithstanding these hard and high‐​stakes questions, the financial regulatory leviathan already has charged headlong into criminalizing the use of certain open‐​source software when the existence of a sanctionable provider is, at the very least, contestable. The Treasury Department’s Office of Foreign Assets Control (OFAC) has been breaking new ground in sanctioning—i.e., prohibiting transactions with—open-source software itself.

Specifically, in August 2022, the OFAC added the Ethereum blockchain addresses of Tornado Cash—a tool for enhancing cryptocurrency transaction privacy—to the sanctioned persons list in connection with the tool’s alleged use by North Korean state‐​sponsored hackers to launder funds.

Tornado Cash users sued the Treasury Department to vacate the sanctions designation. They contended, among other things, that the Tornado Cash developers and token holders were not properly considered a sanctionable “entity” and that the decentralized, open‐​source, and immutable Tornado Cash software was not properly considered sanctionable “property” under relevant law. On August 17, 2023, the court found in favor of the Treasury Department on these issues.

Regardless of whether one thinks the court got it right in the case before it (plaintiffs faced challenging deference standards on interpretive questions), Tornado Cash shows an emerging government suspicion of open‐​source software, with financial regulators at the forefront.

For regulators to continue down this path would risk creating further dangerous precedent. Indeed, the Tornado Cash plaintiffs noted the chilling effect the sanctions designation had on software development. Policymakers should be wary of this chilling effect and the potential lost benefits when it comes to open‐​source financial technology, as well as open‐​source software more broadly.

In the financial context, increasing the risks of publishing open‐​source tools undermines privacy‐​enhancing technologies and the broader use of autonomous financial services that mitigate traditional intermediary risks. In addition, where the suppression extends to open‐​source AI, the potential foregone benefits include the ability of both financial institutions and individuals to run open‐​source AI models on their own hardware to improve processing speed, maintain the confidentiality of personal data, and achieve greater interoperability and customizability.

Notably, the use of more bespoke open‐​source AI models in finance could help to address regulators’ fears of herding behavior due to mono‐​models. Moreover, leveraging experimental tools for autonomous task performance (e.g., an AI agent that could help to organize one’s financial life) thus far is largely a matter of using open‐​source projects. None of this is to say that open‐​source software is always the right tool for the job, and there may ultimately be market forces that make open‐​source models less competitive. But that’s no reason for regulators to put their thumbs on the scale.

As for cutting‐​edge software more broadly, policymakers should consider the role that open‐​source software development may play in discovering and disseminating standards for better aligning AI models (i.e., averting existential risks like civilizational collapse, or worse). Policymakers must steelman the arguments for an open‐​source approach to alignment, including the example of high security standards achieved by community vetting in other open‐​source ecosystems, such as that of the Linux operating system. And even if after careful analysis it’s found that the risks of open‐​source tinkering on sufficiently advanced AI models exceed the benefits at a given moment (given the limits of alignment knowledge at that point), policymakers should not parlay that into a reason to blanket ban open‐​source AI models including those short of the technological frontier.

There are good reasons to expect advances in AI to have transformative impacts on society, including states themselves. And it should come as no surprise that incumbent authorities will react aggressively when perceiving threats to business as usual; indeed, we’ve already seen this in financial regulators sanctioning disintermediated financial tools. But fear of disruption does not justify overreaction in the financial regulatory context or elsewhere.

Notably, when Hammond identified democratized AI as a “greater regime change threat than the internet,” he highlighted that the Chinese Communist Party is already proceeding on that basis. Liberal democracies can and must do better and should have greater confidence in their adaptability to technological change. Reactive policy that targets open‐​source software development carries its own risks. And tilting against open‐​source software without careful deliberation on where that leads is one of the riskiest options of all.

If you’re interested in further discussion on these issues, please join the Cato Institute’s Center for Monetary and Financial Alternatives for a conversation on open‐​source financial technology and broader questions of crypto regulation and competitiveness next Thursday, September 7, 2023.

https://www.cato.org/blog/untested-assumptions-sec-chair-genslers-pivot-ai

Jack Solowey

Crypto startups and venture capitalists are not the only ones pivoting to artificial intelligence (AI). Recently, SEC Chair Gary Gensler delivered remarks to the National Press Club outlining his concerns about AI’s role in the future of finance.

In those high‐​level remarks, Gensler shared his anxiety that AI could threaten macro‐​level financial stability, positing that “AI may heighten financial fragility as it could promote herding with individual actors making similar decisions because they are getting the same signal from a base model or data aggregator.”

This fear largely rests on a pair of debatable assumptions: one, that the market for AI models will be highly concentrated, and two, that this will cause financial groupthink. There are important reasons to doubt both premises. Before the SEC or any regulator, puts forward an AI policy agenda, the assumptions on which it rests must be closely scrutinized and validated.

Assumption 1: Foundation Model Market Concentration

Chair Gensler’s assessment assumes that the market for AI foundation models will be highly concentrated. Foundation models, like OpenAI’s GPT‑4 or Meta’s Llama 2, are pre‐​trained on reams of data to establish predictive capabilities and can serve as bases for “downstream” applications that further refine the models to better perform specific tasks.

Because upstream foundation models are data‐​intensive and have the potential to leverage downstream data for their own benefit, Gensler is concerned that one or a few model providers will be able to corner the market. It’s understandable that one might assume this, but there are plenty of reasons to doubt the assumption.

The best arguments for the market concentration assumption are that natural barriers to entry, economies of scale, and network effects will produce a small number of clear market leaders in foundation models. For instance, pre‐​training can require a lot of data, computing power, and money, potentially advantaging a small number of well‐​resourced players. In addition, network effects (i.e., platforms with more users are more valuable to those users) could further entrench incumbents, either because big‐​tech leaders already have access to more training data from their user networks, because the model providers attracting the most users will come to access more data to further improve their models or some combination of both.

But the assumption that the market for foundation models inevitably will be concentrated is readily vulnerable to counterarguments. For one, the recent AI surge has punctured theories about the perpetual dearth of tech platform competition. With the launch of ChatGPT, OpenAI—a company with fewer than 400 full‐​time employees earlier this year—became a household name and provoked typically best‐​in‐​class firms to scramble in response. And while it’s true that OpenAI has made strategic partnerships with Microsoft, OpenAI’s rise undermined the conventional wisdom that the same five technology incumbents would enjoy unalloyed dominance everywhere forever. The emergence of additional players, like Anthropic, Inflection, and Stability AI, to name just a few, provides further reason to question the idea of a competition‐​free future for AI models.

In addition, the availability of high‐​quality foundation models with open‐​source (or other relatively permissive) licenses runs counter to the assumed future of monopoly control. Open‐​source licenses typically grant others the right to use, copy, and modify software for their own purposes (commercial or otherwise) free of charge. The AI tool builder Hugging Face currently lists tens of thousands of open‐​source models. And other major players are providing their own models with open‐​source licenses (e.g., Stability AI’s new language model) or relatively permissive “source available” licenses (e.g., Meta’s latest Llama 2). Open‐​source model availability could have a material impact on competitive dynamics. A reportedly leaked document from Google put it starkly:

[T]he uncomfortable truth is, we aren’t positioned to win this arms race and neither is OpenAI. While we’ve been squabbling, a third faction has been quietly eating our lunch. I’m talking, of course, about open source.

Lastly, Gensler’s vision of a concentrated foundation model market itself rests in large part on the assumption that model providers will continuously improve their models with the data provided to them by downstream third‐​party applications. But this too should not be taken as a given. Such arrangements are a possible feature of a model provider’s terms but not an unavoidable one. For example, OpenAI’s current data usage policies for those accessing its models through an application programming interface (API), as opposed to OpenAI’s own applications (like ChatGPT), limit (as of March 2023) OpenAI’s use of downstream data to improve its models:

By default, OpenAI will not use API data to train OpenAI models or improve OpenAI’s service offering. Data submitted by the user for fine‐​tuning will only be used to fine‐​tune the customer’s model.

Indeed, providers of base models may not always benefit from downstream data, as finetuning a model for better performance in one domain could risk undermining performance in others (a dramatic form of this phenomenon is known as “catastrophic forgetting”).

Again, this is not to say that foundation model market concentration is impossible. The point is simply that there also are plenty of reasons the concentrated market Gensler envisions may not come to pass. Indeed, a source Gensler cited put it well: “It is too early to tell if the supply of base AI models will be highly competitive or concentrated by only a few big players.” Any SEC regulatory intervention premised on the idea of a non‐​competitive foundation model market would similarly be too early.

Assumption 2: Foundation Model Market Concentration Will Cause Risky Capital Market Participant Groupthink

The second assumption underpinning Gensler’s financial fragility fear is that a limited number of model providers will lead to dangerous uniformity in the behavior of market participants using those models. As Gensler put it, “This could encourage monocultures.”

Even if one accepts for argument’s sake a future of foundation model market concentration, there are reasons to doubt the added assumption that this will encourage monocultures or herd behavior among financial market participants.

While foundation models can be used as generic tools out of the box, they also can be further customized to users’ unique needs and expertise. Finetuning—further training a model on a smaller subset of domain‐​specific data to improve performance in that area—can allow users to tailor base models to firm‐​specific knowledge and maintain a degree of differentiation from their competitors. This complicates the groupthink assumption. Indeed, Morgan Stanley has leveraged OpenAI’s GPT‑4 to synthesize the wealth manager’s own institutional knowledge.

Taking a step back, is it more likely that financial firms with coveted caches of proprietary data and know‐​how will forfeit their competitive advantages, or that they will look to capitalize on them with new tools? Beyond training and finetuning models around firm‐​specific data, firms also can maintain their edge simply by prompting models in a manner consistent with their unique approaches. In addition, firms almost certainly will continue to interpret results based on their specific strategies, cultures, and philosophies. Lastly, because there are profits to be made from identifying mispriced assets, firms would be incentivized to spot others’ inefficient herding behavior and diverge from the “monoculture”; they may even devise ways to leverage models for this purpose.

At the very least, as with model market concentration, more time and research are needed before the impact of the latest generation of AI on financial market participant herding behavior can be assessed with enough confidence to provide a sound basis for regulatory intervention.

Conclusion

Emerging technologies can, of course, be disruptive. But before regulators assume novel technologies present novel risks, they should test and validate their assumptions. Otherwise, one can reasonably doubt regulators when they proclaim themselves “technology neutral.” As SEC Commissioner Hester Peirce noted last week regarding the SEC’s proposed rules tackling a separate AI‐​related concern—conflict-of-interest risks from broker‐​dealers’ and investment advisers’ use of “predictive data analytics”—singling out a specific technology for “uniquely onerous review” is tantamount to “regulatory hazing.”

Another word of caution is warranted: even where regulators do perceive bona fide evidence of enhanced risks, they should be wary of counterproductive interventions. To name just one example, heightened regulatory barriers to entry could worsen the very concentration in the market for AI models that Gensler fears.

https://www.cato.org/blog/continue-work-stablecoin-legislation-or-risk-forfeiting-financial-future

Jack Solowey

The United States has long led global finance. Its institutions shaped critical financial infrastructure and saw the dollar become the world’s reserve currency—thanks to rule of law, property rights, and an innovative market economy at home. As the economic landscape evolves, maintaining this position is a matter of adapting to new technologies that could complement the U.S. dollar and enhance global financial plumbing. Yet, in a fit of myopia, U.S. regulators seem bent on stifling the very developments that could help extend America’s historic strengths, looking askance at recent attempts to integrate open‐​source software with finance.

Yesterday, the House Financial Services Committee’s Subcommittee on Digital Assets, Financial Technology, and Inclusion held a hearing on stablecoins (cryptocurrencies pegged to the value of an asset like the dollar). The committee deserves recognition for taking the all‐​important first step: admitting we have a problem. Nonetheless, although the witnesses largely agreed on the shortsightedness of U.S. hostility to decentralized financial technology and the need for regulatory clarity, comments from lawmakers indicated that a common‐​sense solution on stablecoins, unfortunately, remains far off.

Moreover, a bill posted on the committee’s website before the hearing—a draft stablecoin framework that first circulated last fall—needs work if it is to rein in the excessive regulatory discretion that hinders a competitive stablecoin market and undermines American developers and consumers.

To their credit—Subcommittee Chairman French Hill (R‑AR) and Committee Chairman Patrick McHenry (R‑NC) acknowledged that the bill is but a jumping off point for future revisions—an “infant” in Rep. Hill’s words (and an “ugly baby” in Rep. McHenry’s phrasing from last fall). And Chairman McHenry was candid that the bill is imperfect “in many, many ways.” More pointedly, Ranking Member Maxine Waters (D‑CA) was clear to emphasize that from her perspective, “we’re starting from scratch” and should “disregard the bill that has been posted altogether” given developments in the crypto space since earlier negotiations.

So, what should a final bill look like? To answer that question, it’s important to understand how the U.S.’s tangled web of legacy state and federal laws allow regulators to freestyle when it comes to stablecoins and intervene erratically in the market, which, according to the testimony of Columbia Business School professor Austin Campbell yesterday, is driving developers to more welcoming shores abroad. At the federal level, the Securities Exchange Commission and bank regulators have leveraged ambiguity to threaten enforcement actions against stablecoin projects and caution licensed institutions away from involvement with the crypto ecosystem.

Sensible stablecoin legislation can provide a much‐​needed signal that the U.S. is finally ready to adopt a sane approach to digital assets. However, to achieve that sanity, a stablecoin bill will need to embrace competition from new entrants. This can be accomplished by reducing the regulatory discretion that disserves U.S. businesses and users, avoiding overreactions to experimental instruments, and opening the doors to non‐​traditional market participants.

A stablecoin bill should not grant regulators open‐​ended leeway to reject the applications of stablecoin issuers. Instead, legislation should focus on objective criteria related to reserve assets and disclosures rather than vague factors like a project’s future benefits, contribution to financial stability writ large, overall convenience, or ability to promote financial inclusion.

While those are laudable goals—and consistent with the promise of stablecoins to facilitate competition, transparency, efficient payments, and financial inclusion—evaluating a given stablecoin project’s ability to achieve them ex ante would be highly subjective. Requiring stablecoin issuers to prove their merit in order to exist, instead of simply to mitigate known risks related to the quality and availability of their collateral, would hold stablecoin issuers to a higher standard than other financial institutions. Indeed, the very goals of inclusion and competition would be better served by allowing new market entrants, not creating nebulous prior restraint standards with which to reject new players.

Along these lines, a stablecoin bill should simply address stablecoins’ primary risks: that fiat asset‐​backed projects have the reserves and redemption policies they claim to. As Jake Chervinsky, Chief Policy Officer of the Blockchain Association, noted yesterday, currently “over 90% of the market capitalization for all stablecoins comes from just five custodial stablecoins.” Legislation should avoid expressing an opinion on, let alone banning or pausing, other types of instruments that are erroneously lumped together with fiat‐​collateralized stablecoins, such as crypto‐​asset‐​backed stablecoins and algorithmic stablecoins (which endeavor to maintain stable values by engineering convertibility between two digital assets from the same issuer). While algorithmic stablecoins, for example, are an unproven technology, they’re largely irrelevant to the problem of providing regulatory clarity to businesses tokenizing fiat assets. Moreover, prohibiting financial technology experimentation generally is unbecoming of the leader of the free world and an innovative market economy.

Lastly, stablecoin legislation should allow flexibility when it comes to the types of businesses issuing stablecoins. Not only should non‐​bank and state‐​chartered entities be allowed to become lawful issuers, but so too should businesses from diverse sectors, including those traditionally outside of finance. Preventing companies with other lines of business from issuing stablecoins—or affiliating with those doing so—would risk further constraining financial inclusion and competition. Inclusion goals could be hindered where trusted brands are blocked from serving the markets and communities they know best. And potential efficiency gains could be lost where networked businesses in other sectors (like social media and ecommerce platforms) are unable to bring their expertise to bear in the stablecoin market should they choose to.

Through curbing regulatory discretion, avoiding disproportionate interventions, and opening the field to new participants, Congress could help to resolve the U.S.’s unsustainable stablecoin status quo. If the U.S. wishes to remain the world’s preeminent financial market, legislative work on stablecoins must continue to ensure that our laws are open to technologies with the potential to help maintain and extend that lead.