On Global AI Governance (evidence for the Office of the UN Secretary General's Envoy on Technology)

 This is an open letter to the Office of the UN Secretary-General's Envoy on Technology, who are attempting to move into the space of AI governance and have put out a call for papers.

Update April 2024: see also my newer blogpost, AI Is Not a Unitary Actor: My Response to the UN Interim Report Consultation.


Dear Colleagues,

This is a brief summary of my thinking and writings on the topic of global AI governance. 

Terms and framing

First to be very clear about the quite general way I think about governance and regulation:
  • I will take regulation to mean any means by which a complex entity perpetuates a recognisable version of itself into the future. For example, breathing regulates oxygen and carbon dioxide levels in our bodies, tax policies not only generate revenue but can help regulate economies.
  • Governing is deliberate, explicit regulation. Government is just one means by which we regulate our societies and nations, though it can be a very useful one. 

Is the entire globe the right level for any governance or regulation of AI?

One of the key questions of digital governance is who should do it. This question is often framed falsely as choices between exclusive options. Framing regulation and governance as I just have, it should be evident that every company must self govern, and so must every country, and these actions will interact. Some countries have had the good fortune and foresight to find other countries with whom they can coordinate their trade, security, and digital governance quite effectively. Similarly, some disparate peoples have, through some combination of choice or history, come to cohabit a large, powerful country where relatively uniform laws and enforcement facilitate the regulation of their companies and economy.

The global level has proved useful for negotiating or at least setting expectations about necessarily transnational concerns such as rules of war, human rights, environmental protections, some trade agreements and so forth. It is quite likely that there are aspects of digital governance that would also benefit at this level of coordination, and agreed mutual enforcement, though as usual it would of course be nations that would write and enforce detailed law. The costs of distance are so low for delivering products of digital industries, that many companies (and other digital ventures–even those not legal) can very quickly become transnational, and therefore quite likely require transnational regulation, and transnational cooperation in enforcing that regulation. Further, there is some indication that even corporations in other, more-established sectors also can become a "natural monopoly" provided they invest sufficiently in software to solve logistics. Software, indeed especially AI, can help reduce or eliminate any diseconomies of scale (Bessen 2022).

My own vision concerning global regulation and governance of AI

I have a few publications concerning specifics of global and transnational governance, which I will get to in the next section. I have not yet however put all of my thoughts on this topic in one place, so I will do so here.

I do not believe there should be a single set of uniform law for the whole world. I realise this would be convenient for some companies, and as such, efficient for our economies. But each country has its own history, capacities, neighbours, resources and so forth. As such, each will have different best ways to resolve the tensions inherent in the tradeoffs of governing. Also, diversity is essential to resilience (including from regulatory capture) and agility. 

Having said that, it is evident that scale facilitates governance, including by allowing the exertion of market (and other) power to ensure compliance with local laws. I do not think the Brussels effect (whereby the EU wields some power in setting global norms of digital governance) is really unique so much as it is surprising, because historically a set of transnational markets have perhaps never been so well coordinated – better coordinated than within some countries. I expect that we will see more such unions, which will along with large nations form regulatory hubs. Some smaller nations may always be suspended between more than one hub, again for historic and geographic, realpolitik reasons. Nevertheless, to the extent possible, countries who want the benefits of transnational digital commerce will want to comply with some set of wider digital governance standards, so that doing trade within their borders will be economically worthwhile. The UN might serve to help efforts to coordinate such cooperation, but did not play such a role for the EU.

My papers on AI Governance

My most recent publication (2023), Going Nuclear? Precedents and Options for the Transnational Governance of Artificial Intelligence (with David Backovsky, for Horizons) discusses in 12 Foreign-Affairs-type pages the relevant history of AI, nuclear, and environmental global governance efforts. There are two main points:  
  1. AI is a technology, so much more like nuclear power/weapons, than the climate. This is because for any technology, its outputs and impacts are much more subject to regulatory context than nature is.
  2. The US is presently resistant to, even interfering with, global cooperation on AI governance. But in the past it has overcome similar internal national resistance in other sectors to facilitate very good orders, benefiting itself and most (perhaps all) others.
A few years ago, I was asked to write the first chapter for an Oxford University Press handbook: The Artificial Intelligence of the Ethics of Artificial Intelligence: An Introductory Overview for Law
and Regulation. (2020. Here is an open access URL, I also attach the final PDF). Precis: 
This chapter provides an overview of the nature and implications of artificial intelligence (AI), with particular attention to how they impinge on possible applications to and of law. Any artifact that transforms perception to more relevant information, including action, is AI. There is no question that AI, and digital technologies in general, are introducing massive transformations to society. Nevertheless, these impacts should be governable by less transformative legislative change. Indeed, the vast majority of AI—particularly where it has social impact—is and will remain a consequence of corporate commercial processes, and as such subject to existing regulations and regulating strategies. However, it is critical to remember that what is being held accountable is not machines themselves but the people who build, own, or operate them—including any who alter their operation through assault on their cybersecurity. It is thus important to govern the human application of technology—the human processes of development, testing, operation, and monitoring.

Most of my work on governance has been focussed on bettering EU policy, which is of course in many ways world-leading.  Here is another invited handbook chapter–here, the Handbook of AI Governance (!) and here coauthored with Mark Dempsey, Keegan McBride, Meeri Haataja (May 2022) Transnational Digital Governance and Its Impact on Artificial Intelligence (again, a green open access URL and I attach the final PDF) Precis:

The rapid pace of technological advancement and innovation has put existing governance and regulatory mechanisms to the test. There is a clear need for new and innovative regulatory mechanisms that enable governments to successfully manage the integration of digital technologies into our societies, and to ensure that such integration occurs in a sustainable, beneficial, and just manner. Artificial Intelligence (AI) stands out as one of the most debated of such innovations. What exactly is it, how should it be built and deployed, how can it be used, and how should it be regulated? Yet across the period of this debate, AI is becoming widely used and addressed within existing, evolving, and bespoke regulatory contexts. The present chapter explores the extant governance of AI and, in particular, what is arguably the most successful AI regulatory approach to date, that of the European Union. The chapter explores core definitional concepts, shared understandings, values, and approaches currently in play. It argues that not only are the Union’s regulations locally effective, but, due to the so-called “Brussels effect,” regulatory initiatives within the European Union also have a much broader global impact. As such, they warrant close consideration.

Returning briefly to the truly global, though initially done with respect to whether the EU even should be regulating AI, I wrote an article exploring the capacities of not only the EU, US, and China to build AI, but the entire rest of the world as well. We found the EU and China to be surprisingly comparable, and the rest of the world outside these three regions to be stronger than many realise.  Is There an AI Cold War? (with Helena Malikova, 2021) In addition to this data, we also present qualitative, narrative evidence of attempts at regulatory interference we have witnessed. (I have a follow-up article in progress looking at more years of data: exact leadership is volatile, but the policy-relevant analytic outcomes are consistent with this published paper.)

Regulation is a means societies use to create the stability, public goods, and infrastructure they need to thrive securely. This policy brief is intended to both document and to address claims of a new AI cold war: a binary competition between the United States and China that is too important for other powers to either ignore or truly participate in directly, beyond taking sides. We argue that while some of the claims of this narrative are based at least in part on genuine security concerns and important unknowns, evidence for its extreme binary nature is lacking. This absence of factual evidence is concerning, because related geopolitical tensions may be used to interfere with regulation of AI and agencies associated with its development. Here we first document and then analyze the extremely bipolar picture prominent policymakers and political commentators have been recently painting of the AI technological situation, portraying China and the United States as the only two global powers. We then examine the plausibility of these claims using two measures: internationally registered AI patents and the market capitalization of the companies that hold them. These two measures, while each somewhat arbitrary and imperfect, are often deployed in the context of the binary narrative and can therefore be seen as conservative choices in that they should favor exactly the “champions” of that narrative. In fact, these measures do not produce bipolar results: Chinese capacity has been exaggerated and that of other global regions deprecated. These findings call into question the motivation behind the documented claims, though they also further illuminate the uncertainty concerning digital technology security. We recommend that all parties engage in contributing to a safe, secure, and transparent regulatory landscape.

With Meeri Haataja, I have also written a series of three articles seeking to immediately inform and improve the EU's AI Act. (These are intended to be combined into one, but not soon enough that I can share that now.) 

I also wrote a series of blogposts documenting my advice to EU and UK governance, 

I've also written two relevant popular science articles on the topic of AI and its regulation for Wired (and hope to write another one soon)
Thank you for your service to the global community, and best wishes in consolidating all the evidence you are accumulating.

yours,

Joanna Bryson
[note: slightly edited for clarity 23 February 2024]

Screen shot of Going Nuclear, includes GPAI members at a Tokyo meeting


Comments