Blog

A few words from the director:

In my view, the governance of AI will become among the most important global issues. 2020 saw many continued developments in AI governance. It is heartening to see how rapidly this field continues to grow, and exciting to be part of that growth.

This report provides a summary of our activities in 2020.

We now have a core team of 9 researchers and a network of 21 affiliates and collaborators. We are excited to have welcomed Visiting Senior Researchers Joslyn Barnhart and Robert Trager. This year we published two major reports, 15 academic publications, an AI governance syllabus, and 5 op-eds/blog posts. Our work covered many topics:

• Theory of Impact for AI governance
• The Windfall Clause
• Cooperative AI
• Clarifying the logic of strategic assets
• National security and antitrust
• AI and corporate intellectual property strategies
• AI researcher responsibility and impact statements
• Historical economic growth trends
• AI and China
• Trustworthy AI development
• And more…

As I argued in AI Governance: Opportunity and Theory of Impact, we are highly uncertain of the technical and geopolitical nature of the problem, and so should acquire a diverse portfolio of expertise. Accordingly, our work covers only a small fraction of the problem space. We are excited about growing our team and have big ambitions for further progress. We would like to thank Open Philanthropy, the Future of Life Institute, and the European Research Council for their generous support. As part of the Future of Humanity Institute, we have been immersed in good ideas, brilliant people, and a truly long-term perspective. The University of Oxford, similarly, has been a rich intellectual environment, with increasingly productive connections to the Department of Politics and International Relations, the Department of Computer Science, and the new Ethics in AI Institute.

We are always looking to help new talent get into the field of AI governance, be that through our Governance of AI Fellowship (applications are expected to open in Spring 2021), hiring researchers, finding collaborators, or hosting senior visitors. If you are interested in working with us, visit www.governance.ai for updates on our latest opportunities, or consider reaching out to Markus Anderljung (

ma***************@ph********.uk











).

We look forward to seeing what we can all achieve in 2021.

Allan Dafoe
Director, Centre for the Governance of AI
Associate Professor and Senior Research Fellow
Future of Humanity Institute, University of Oxford

‍

Research

You can find all our publications here. Our 2019 annual report is here; 2018 report here.
‍

Major Reports and Academic Publications

• “Open Problems in Cooperative AI” (2020). Allan Dafoe, Edward Hughes (DeepMind), Yoram Bachrach (DeepMind), Teddy Collins (DeepMind & GovAI affiliate), Kevin R. McKee (DeepMind), Joel Z. Leibo (DeepMind), Kate Larson, and Thore Graepel (DeepMind). arXiv. (link)
  Problems of cooperation—in which agents seek ways to jointly improve their welfare—are ubiquitous and important. They can be found at scales ranging from our daily routines—such as highway driving, scheduling meetings, and collaborative work—to our global challenges—such as arms control, climate change, global commerce, and pandemic preparedness. Arguably, the success of the human species is rooted in our ability to cooperate. Since machines powered by artificial intelligence are playing an ever greater role in our lives, it will be important to equip them with the capabilities necessary to cooperate and to foster cooperation. The authors see an opportunity for the field of Artificial Intelligence to explicitly focus effort on this class of problems which they term Cooperative AI. As part of this we co-organized a NeurIPS workshop: www.cooperativeAI.com

• “The Windfall Clause: Distributing the Benefits of AI for the Common Good” (2020). Cullen O’Keefe (OpenAI & GovAI affiliate), Peter Cihon (GitHub & GovAI affiliate), Carrick Flynn (CSET & GovAI affiliate), Ben Garfinkel, Jade Leung, and Allan Dafoe. Proceedings of the 2020 AAAI/ACM Conference on AI, Ethics, and Society (AIES20). (report) (article summary).
  The windfall clause is a policy proposal to devise a mechanism for AI developers to make ex-ante commitments to distribute a substantial part of profits back to the global commons if they were to capture an extremely large part of the global economy via developing transformative AI. The project was run by GovAI, and inspired the Partnership on AI’s launch of their Shared Prosperity Initiative.

• “The Offense-Defense Balance of Scientific Knowledge: Does Publishing AI Research Reduce Misuse of the Technology?” (2020). Toby Shevlane and Allan Dafoe. Proceedings of the 2020 AAAI/ACM Conference on AI, Ethics, and Society (AIES20). (link).
  The existing conversation within AI has imported concepts and conclusions from prior debates within computer security over the disclosure of software vulnerabilities. While disclosure of software vulnerabilities often favours defence, this article argues that the same cannot be assumed for AI research. It provides a theoretical framework for thinking about the offense-defense balance of scientific knowledge.

• “U.S. Public Opinion on the Governance of Artificial Intelligence” (2020). Baobao Zhang and Allan Dafoe. Proceedings of the 2020 AAAI/ACM Conference on AI, Ethics, and Society (AIES20). (link).
  The report presents the results from an extensive survey into 2,000 Americans’ attitudes toward AI and AI governance. The full results were published in 2019 here.

• “Social and Governance Implications of Improved Data Efficiency” (2020). Aaron Tucker (Cornell University), Markus Anderljung, and Allan Dafoe. Proceedings of the 2020 AAAI/ACM Conference on AI, Ethics, and Society (AIES20). (link).
  Many researchers work on improving the data efficiency of machine learning. What would happen if they succeed? This paper explores the social-economic impact of increased data efficiency on e.g. market concentration, malicious use, privacy, and robustness.

• “Institutionalising Ethics in AI: Reflections on the NeurIPS Broader Impact Requirement” (Forthcoming). Carina Prunkl (Ethics in AI Institute & GovAI affiliate), Carolyn Ashurst, Markus Anderljung, Helena Webb (University of Oxford), Jan Leike (OpenAI), and Allan Dafoe. Nature Machine Intelligence.
  Turning principles into practice is one of the most pressing challenges of artificial intelligence (AI) governance. In this article, we reflect on a novel governance initiative by one of the world’s most prestigious AI conferences: NeurIPS.

• “Who owns artificial intelligence? A preliminary analysis of corporate intellectual property strategies and why they matter” (2020). Nathan Calvin (GovAI affiliate) and Jade Leung (GovAI affiliate). GovAI Working Paper. (link).
  This working paper is a preliminary analysis of the legal rules, norms, and strategies governing AI-related intellectual property (IP). It analyzes the existing AI-related IP practices of select companies and governments, and provides some tentative predictions for how these strategies and dynamics may continue to evolve in the future.

• “How Will National Security Considerations Affect Antitrust Decisions in AI? An Examination of Historical Precedents” (2020). Cullen O’Keefe (OpenAI & GovAI affiliate). GovAI Technical Report. (link).
  Artificial Intelligence—like past general purpose technologies such as railways, the internet, and electricity—is likely to have significant effects on both national security and market structure. These market structure effects, as well as AI firms’ efforts to cooperate on AI safety and trustworthiness, may implicate antitrust in the coming decades. Meanwhile, as AI becomes increasingly seen as important to national security, such considerations may come to affect antitrust enforcement. By examining historical precedents, this paper sheds light on the possible interactions between traditional—that is, economic—antitrust considerations and national security in the United States.

• “The Logic of Strategic Assets” (2020). Jeffrey Ding and Allan Dafoe. Forthcoming. Security Studies. (link).
  This paper asks what makes an asset strategic, in the sense of warranting the attention of the highest levels of the state. By clarifying the logic of strategic assets, it could move policymakers away from especially unhelpful rivalrous industrial policies, and can clarify the structural pressures that work against global economic liberalism. The paper applies this analysis to AI.

• “Beyond Near- and Long-Term: Towards a Clearer Account of Research Priorities in AI Ethics and Society” (2020). Carina Prunkl (Ethics in AI Institute & GovAI affiliate) and Jess Whittlestone (CFI). Proceedings of the 2020 AAAI/ACM Conference on AI, Ethics, and Society (AIES20). (link).
  This article considers the extent to which there is a tension between focusing on the near and long term AI risks.

• “Beyond Privacy Trade-offs with Structured Transparency” Andrew Trask (DeepMind & GovAI affiliate), Emma Bluemke (University of Oxford), Ben Garfinkel, Claudia Ghezzou Cuervas-Mons (Imperial College London), Allan Dafoe. Proceedings of the 2020 AAAI/ACM Conference on AI, Ethics, and Society (AIES20). (link).
  Many socially valuable activities depend on sensitive information, such as medical research, public health policies, political coordination, and personalized digital services. This is often posed as an inherent privacy trade-off: we can benefit from data analysis or retain data privacy, but not both. Across several disciplines, a vast amount of effort has been directed toward overcoming this trade-off to enable productive uses of information without also enabling undesired misuse, a goal we term ‘structured transparency’. In this paper, we provide an overview of the frontier of research seeking to develop structured transparency. We offer a general theoretical framework and vocabulary, including characterizing the fundamental components — input privacy, output privacy, input verification, output verification, and flow governance — and fundamental problems of copying, bundling, and recursive oversight. We argue that these barriers are less fundamental than they often appear. We conclude with several illustrations of structured transparency — in open research, energy management, and credit scoring systems — and a discussion of the risks of misuse of these tools.

• “Public Policy and Superintelligent AI: A Vector Field Approach” (2020). Nick Bostrom, Allan Dafoe, and Carrick Flynn (CSET & GovAI affiliate). Ethics of Artificial Intelligence, Oxford University Press, ed. S. Matthew Liao. (link).
  The chapter considers the speculative prospect of superintelligent AI and its normative implications for governance and global policy.

• “Toward Trustworthy AI Development: Mechanisms for Supporting Verifiable Claims” (2020). Miles Brundage et al. arXiv. (link)
  This report suggests various steps that different stakeholders in AI development can take to make it easier to verify claims about AI development, with a focus on providing evidence about the safety, security, fairness, and privacy protection of AI systems. Implementation of such mechanisms can help make progress on the multifaceted problem of ensuring that AI development is conducted in a trustworthy fashion. The mechanisms outlined in this report deal with questions that various parties involved in AI development might face. (Note: The work was led by researchers at OpenAI and there were 59 contributing authors to this report. Of these, 3 were GovAI researchers and 6 were GovAI affiliates).

Other Academic Publications

• “The Suffragist Peace” (2020). Joslyn N. Barnhart (UCSD), Robert F. Trager (UCLA), Elizabeth N. Saunders (Georgetown University) and Allan Dafoe. International Organization. (link)
  Drawing on theory, a meta-analysis of survey experiments in international relations, and analysis of cross national conflict data, the paper shows how features of women’s preferences about the use of force translate into specific patterns of international conflict. When empowered by democratic institutions and suffrage, women’s more pacific preferences generate a dyadic democratic peace (i.e., between democracies), as well as a monadic peace. The analysis supports the view that the enfranchisement of women is essential for the democratic peace. The results were summarised in Foreign Affairs, by the same authors.

• “Coercion and the Credibility of Assurances” (Forthcoming). Matthew Cebul (University of Michigan), Allan Dafoe, and Nuno Monteiro (Yale University). Journal of Politics. (link).
  This paper offers a theoretical framework exploring the causes and consequences of assurance credibility and provides empirical support for these claims through a nationally-representative, scenario-based survey experiment that explores how US citizens respond to a hypothetical coercive dispute with China.

• “Coercion and Provocation” (Forthcoming). Allan Dafoe, Sophia Hatz (Uppsala University), and Baobao Zhang. The Journal of Conflict Resolution. (link).
  In this paper the authors review instances of apparent provocation in interstate relations and offer a theory based on the logic of reputation and honor. Using survey experiments they systematically evaluate whether provocation exists and what may account for it and employ design-based causal inference techniques to evaluate their key hypotheses.

• “The biosecurity benefits of genetic engineering attribution” (2020). Gregory Lewis … Jade Leung (GovAI affiliate), Allan Dafoe, et al. Nature Communications. (link).
  A key security challenge in biotechnology involves attribution: determining, in the wake of a human-caused biological event, who was responsible. The article discusses a technique which could be developed into powerful forensic tools to aid the attribution of outbreaks caused by genetically engineered pathogens.

Opinion Articles, Blog Posts, and Other Public Work

• “AI Governance: Opportunity and Theory of Impact” (2020). Allan Dafoe. Effective Altruism Forum. (link).
  This piece describes the opportunity and theory of impact of work in the AI governance space from a longtermist perspective. The piece won an Effective Altruism Forum Prize and was the most highly voted post of September.

• “A Guide to Writing the NeurIPS Impact Statement” (2020). Carolyn Ashurst (Ethics in Ai Institute), Markus Anderljung, Carina Prunkl, Jan Leike (OpenAI), Yarin Gal (University of Oxford, CS dept.), Toby Shevlane, and Allan Dafoe. Blog post on Medium. (link).
  This guide was written in light of NeurIPS — the premier conference in machine learning — introducing a requirement that all paper submissions include a statement of the “potential broader impact of their work, including its ethical aspects and future societal consequences.” The post has garnered over 14,000 views, more than the approximately 12,000 abstract submissions received by the conference.

• “Does Economic History Point Toward a Singularity?” (2020). Ben Garfinkel. Effective Altruism Forum. (link).
  Over the next several centuries, is the economic growth rate likely to remain steady, radically increase, or decline back toward zero? This piece investigates the claim that historical data suggests growth may increase dramatically. Specifically, it looks at the hyperbolic growth hypothesis: the claim that, from at least the start of the Neolithic Revolution up until the 20th century, the economic growth rate has tended to rise in proportion with the size of the global economy. The piece received the Effective Altruism Forum Prize for best post in September.

• “Ben Garfinkel on scrutinising classic AI risk arguments” (2020). Ben Garfinkel. 80,000 hours podcast. (link)
  Longtermist arguments for working on AI risks originally focussed on catastrophic accidents. Ben Garfinkel makes the case that these arguments often rely on imprecisely defined abstractions (e.g. “optimisation power”, “goals”) and toy thought experiments. It is not clear that these constitute a strong source of evidence. Nevertheless, working in AI governance or AI Safety still seems very valuable.

• “China, its AI dream, and what we get wrong about both.” (2020). Jeffrey Ding. 80,000 hours podcast. (link)
  Jeffrey Ding discusses his paper “Deciphering China’s AI Dream” and other topics including: analogies for thinking about AI influence; cultural cliches in the West and China; coordination with China on AI; private companies vs. government research.

• Talk: “AI Social Responsibility” (2020). Allan Dafoe. AI Summit London. (link)
  AI Social Responsibility is a framework for collectively committing to make responsible decisions in AI development. In this talk, Allan Dafoe outlines that framework and explains its relevance to current AI governance initiatives.

• “Consultation on the European Commission’s White Paper on Artificial Intelligence: a European approach to excellence and trust” (2020). Stefan Torges, Markus Anderljung, and the GovAI team. Submission of the Centre for the Governance of AI. (link)
  The submission presents GovAI’s recommendations regarding the European Union’s AI strategy. Analysis and recommendations focus on the proposed “ecosystem oftrust” and associated international efforts. We believe these measures can mitigate the risks that this technology poses to the safety and rights of Europeans.

• “Contact tracing apps can help stop coronavirus. But they can hurt privacy.” (2020). Toby Shevlane, Ben Garfinkel and Allan Dafoe. Washington Post. (link)
  Contact tracing apps have reignited debates over the trade-off between privacy and security. Trade-offs can be minimised through technologies which allow “structured transparency”. These achieve both high levels of privacy and effectiveness through the careful design of information architectures — the social and technical arrangements that determine who can see what, when and how.

• “Women’s Suffrage and the Democratic Peace” (2020). Joslyn Barnhart, Robert Trager, Elizabeth Saunders (Georgetown), and Allan Dafoe. Foreign Affairs. (link)
  Presenting the ideas from “The Suffragist Peace”.

• “Artificial Intelligence and China” (2020). Jeffrey Ding, Sophie-Charlotte Fischer, Brian Tse, and Chris Byrd. GovAI Syllabus. (link).
  In recent years, China’s ambitious development of artificial intelligence (AI) has attracted much attention in policymaking and academic circles. This syllabus aims to broadly cover the research landscape surrounding China’s AI ecosystem, including the context, components, capabilities, and consequences of China’s AI development.

• “The Rapid Growth of the AI Governance Field” (2020). Allan Dafoe and Markus Anderljung. AI Governance in 2019 — A Year in Review: Observations from 50 Global Experts, ed. Li Hui & Brian Tse. (link)
  This report was contributed to by 50 experts from 44 institutions, including AI scientists, academic researchers, industry representatives, policy experts, and others.

• “The Case for Privacy Optimism” (2020). Ben Garfinkel. Blog post. (link).
  This blog post argues that social privacy — from the prying eyes of e.g. family, friends, and neighbours — has increased over time, and may continue to do so in the future. While institutional privacy has decreased, it may be counteracted by the increase in social privacy.

Events

‍Webinars

‍

Workshops co-organized by GovAI

• Cooperative AI Workshop at the NeurIPS 2020 conference. Speakers included: James D. Fearon (Stanford), Gillian Hadfield (University of Toronto), William Isaac (Deepmind), Sarit Kraus (Bar-Ilan University), Peter Stone (Learning Agents Research Group), Kate Larson (University of Waterloo), Natasha Jaques (Google Brain), Jeffrey S. Rosenschein (Hebrew University), Mike Wooldridge (University of Oxford), Allan Dafoe, Thore Graepel (Deepmind).

• Navigating the Broader Impacts of AI Research Workshop at the NeurIPS 2020 conference.  Speakers: Hanna Wallach (Microsoft),  Sarah Brown (University of Rhode Island), Heather Douglas (Michigan State University), Iason Gabriel (DeepMind, NeurIPS Ethics Advisor), Brent Hecht (Northwestern University, Microsoft), Rosie Campbell (Partnership on AI), Anna Lauren Hoffmann (University of Washington), Nyalleng Moorosi (Google AI), Vinay Prabhu (UnifyID), Jake Metcalf (Data & Society), Sherry Stanley (Amazon Mechanical Turk), Deborah Raji (Mozilla), Logan Koepke (Upturn), Cathy O’Neil (O’Neil Risk Consulting & Algorithmic Auditing), Tawana Petty (Stanford University), Cynthia Rudin (Duke University), Shawn Bushway (University at Albany), Miles Brundage (OpenAI & GovAI affiliate), Bryan McCann (formerly Salesforce), Colin Raffel (University of North Carolina at Chapel Hill, Google Brain), Natalie Schluter (Google Brain, IT University of Copenhagen), Zeerak Waseem (University of Sheffield), Ashley Casovan (AI Global), Timnit Gebru (Google), Shakir Mohamed (DeepMind), Aviv Ovadya (Thoughtful Technology Project), Solon Barocas (Microsoft), Josh Greenberg (Alfred P. Sloan Foundation), Liesbeth Venema (Nature), Ben Zevenbergen (Google), Lilly Irani (UC San Diego).

• We hosted a CNAS-FHI Workshop on AI and International Stability in January.

‍

Selected publications by research affiliates

• “Economic Growth under Transformative AI: A guide to the vast range of possibilities for output growth, wages, and the labor share” (2021). Philip Trammell (GPI) and Anton Korinek (UVA and GovAI affiliate). Global Priorities Institute Working Paper. (link)

• “Neither Private Nor Fair: Impact of Data Imbalance on Utility and Fairness in Differential Privacy” (2020). Tom Farrand, Fatemehsadat Mireshghallah (UCSD), Sahib Singh (Ford), Andrew Trask (DeepMind & GovAI affiliate). Proceedings of the 2020 Workshop on Privacy-Preserving Machine Learning in Practice. (link)

• “COVID-19 Infection Externalities: Trading Off Lives vs. Livelihoods” (2020). Zachary A. Bethune (University of Virginia) and Anton Korinek (University of Virginia & GovAI affiliate). NBER Working Paper. (link)

• “Nonpolar Europe? Examining the causes and drivers behind the decline of ordering agents in Europe” (2020). Hiski Haukkala (University of Tampere & GovAI affiliate). International Politics. (link)

• “All the News that’s Fit to Fabricate: AI-Generated Text as a Tool of Media Misinformation” (2020). Sarah E. Kreps (Cornell), Miles Mcain (Stanfaord), and Miles Brundage (OpenAI & GovAI affiliate). SSRN. (link)

• “Messier than Oil: Assessing Data Advantage in Military AI” (2020). Husanjot Chahal (CSET), Ryan Fedasiuk (CSET), and Carrick Flynn (CSET & GovAI affiliate). CSET Issue Brief. (link)

• “The Chipmakers: U.S. Strengths and Priorities for the High-End Semiconductor Workforce” (2020). Will Hunt (CSET) and Remco Zwetsloot (CSET & GovAI affiliate). CSET Issue Brief. (link)

• “Antitrust-Compliant AI Industry Self-Regulation” (2020). Cullen O’Keefe (OpenAI & GovAI affiliate). Working Paper. (link)

• “Have Your Data and Use It Too: A Federal Initiative for Protecting Privacy while Advancing AI.” (2020). Roxanne Heston (CSET) and Helen Toner (CSET & GovAI affiliate). Day One Project. (link)

• “Americans’ Perceptions of Privacy and Surveillance in the COVID-19 Pandemic.” (2020). Baobao Zhang (Cornell & GovAI affiliate), Sarah Kreps (Cornell), Nina McMurry (WZB Berlin Social Science Center), and R. Miles McCain (Stanford University). PLoS ONE. Replication files. Coverage in Bloomberg and IEEE Spectrum; shared with the World Health Organization. (link)

‍

Team and Growth

Our team has grown substantially. In 2020 we welcomed Robert Trager and Joslyn Barnhart as Visiting Senior Research Fellows and Eoghan Stafford as a Visiting Researcher. We ran another round of the GovAI Fellowship and welcomed 7 Fellows, with an acceptance rate of around 5%.  Our management team also evolved, with Alexis Carlier joining as a Project Manager following Jade Leung’s departure.

We continue to receive a lot of applications and expressions of interest from researchers across the world who are eager to join our team. In 2021, we plan to continue our GovAI Fellowship programme, engaging with PhD researchers primarily in Oxford, and hiring additional researchers.

Daron Acemoğlu is an economist and the Elizabeth and James Killian Professor of Economics and Institute Professor at the Massachusetts Institute of Technology (MIT), where he has taught since 1993. He was awarded the John Bates Clark Medal in 2005 and co-authored Why Nations Fail: The Origins of Power, Prosperity, and Poverty with James A. Robinson in 2012.

Diane Coyle, CBE, OBE, FAcSS is an economist, former advisor to the UK Treasury, and the Bennett Professor of Public Policy at the University of Cambridge, where she has co-directed the Bennett Institute since 2018. She was vice-chairman of the BBC Trust, the governing body of the British Broadcasting Corporation, and was a member of the UK Competition Commission from 2001 until 2019. In 2020, she published Markets, State, and People: Economics for Public Policy.

Joseph Stiglitz is an economist, public policy analyst, and a University Professor at Columbia University. He is a recipient of the Nobel Memorial Prize in Economic Sciences (2001) and the John Bates Clark Medal (1979). He is a former senior vice president and chief economist of the World Bank and is a former member and chairman of the US President’s Council of Economic Advisers. His most recent book, Measuring What Counts; The Global Movement for Well-Being came out in 2019.

‍

You can watch a recording of the event here or read the transcript below:

‍

Allan Dafoe:

Welcome to our inaugural webinar on the governance and economics of AI. It is extremely exciting to see so many audience members from around the world. I see in the chat Portugal, Shanghai, Brazil represented, so that’s great. I am Allan Dafoe, the director of the Centre for the Governance of AI which is organizing this series. We are based at the Future of Humanity Institute at the University of Oxford. For those of you who don’t know about our work, we study the opportunities and challenges brought by advances in AI so as to advise policy to maximize the benefits and minimize the risks. We understand AI as broadly referring to the cluster of technologies associated with machine intelligence, especially the recent progress in machine learning, but also including advances in computing power, sensors, robotics and our digital infrastructure. The term governance, which may not be familiar to many of you, refers both descriptively to the ways that decisions are in fact made about the development and deployment of AI, but also to the normative aspiration that those decisions emerge from institutions that are effective, equitable and legitimate.

We have a special interest in understanding the long run impact of artificial intelligence.  Over the past few years, it has become increasingly common for economists to identify AI as a general-purpose technology or GPT, as I expect we’ll hear about more today. If AI turns out to be anything like previous transformative GPTs such as electricity and the internal combustion engine, then we can expect massive changes in our culture, politics, and in the character of war.

More speculatively, AI might even turn out to be something more than another GPT in a long line of GPTs. A number of scholars, including those attending today, have begun to explore more radical possibilities and their associated challenges such as massive labor displacement, extreme inequality, rapidly accelerating economic growth, and the maintenance of human oversight of highly intelligent artificial systems. This webinars series will continue these conversations.  In the coming months, we will host a conversation on challenges for US-China cooperation and the governance of AI, on the impact of AI on democracy, on forecasting methodology and insights for trends in AI, as well as many more discussions of the economics of AI.

This series is put on in partnership with Anton Korinek of the University of Virginia — who is sharing the screen with me — who will be moderating today’s event. Anton is one of the leading economists who has been thinking seriously about the economic implications of advanced AI. Anton first came to my attention because of his excellent paper coauthored with Joseph Stiglitz, also here with us today, on the implications of AI for income distribution and unemployment. In this paper they discuss with subtlety and insight the many challenges to making technological progress broadly beneficial due to failures in insurance markets for technological displacement, and the costs and feasibility of redistribution. From my conversations with Anton, I’ve learned a lot more about the economics of AI and I encourage you all to follow his work. I will now turn the mic over to Anton to introduce and moderate this event.

‍

Anton Korinek:

Let me thank the GovAI team, Markus Anderljung and Anne le Roux, for making this event possible. Let me also thank Allan for hosting us and for the kind introduction. I have followed Allan’s work for a number of years. What I find really admirable is that he focuses on how to put into practice many of the policy proposals that economists like myself only consider in theory.

In the economics of AI, a big theme is that smart machines may be a substitute for human labor rather than complementing us. And that this may be unlike what earlier technological revolutions entailed. The fear is that this will progressively lead to a decline in relative and perhaps even absolute demand for labor, driving down wages, and when wages cannot fall, causing unemployment. This would exacerbate inequality, poverty, and social and political tension.

Well, just like doctors learn the most about the human body when it is sick or injured, economists learn the most about the economy when it is in crisis.

When Allan and I first spoke about this webinar series, we felt that it would be a fitting theme for our inaugural event to invite three of the world’s top thinkers on the economics of AI to share with us what they have learned from the ongoing pandemic and what lessons this provides us for how we as a society can prepare for the advent of ever smarter machines.

Aside from devastating health effects, Covid-19 has led to hundreds of millions of jobs lost around the world — probably one of the largest negative labor demand shocks in human history, although it was a policy induced and temporary one. It has also led to unprecedented government actions to support the jobless while simultaneously giving rise to significant political tension. So one important question is what can we learn to prepare for potential future labor demand shocks that may arise from automation?

Another issue is that Covid-19 has also spurred a massive technological transition into the virtual world in which the marginal cost of distribution is zero. So instead of holding an in-person conference on our topic today (and I would very much enjoy being with you in person), we have to live stream this event on the web. There are some obvious benefits: it democratizes the attendance, but it also risks exacerbating the superstars phenomenon and exacerbating inequality in our world. Another really important question is: what we can learn from Covid-19 about a future that is increasingly digital? More broadly, let me ask our panelists: what lessons have you learned from the pandemic that we can carry over to the governance of AI?

Without further ado, let me introduce three superstars who are our panelists today, Daron Acemoğlu, Diane Coyle, and Joseph Stiglitz.

Daron Acemoğlu is the Elizabeth and James Killian Professor of Economics and Institute Professor at the Massachusetts Institute of Technology where he has taught since 1993, and he is also a winner of the John Bates Clark medal. Has coauthored a book on Why Nations Fail: The Origins of Power, Prosperity, and Poverty.

Daron Acemoğlu:

It’s a great pleasure to be here even though we cannot all be in the same room. And I think Anton gave an excellent introduction to what I wanted to say, which is that we are living through a transformative moment and this is true for many dimensions of our lives, but two that are particularly important are the future of technology, especially related to AI, and the future of institutions. Both because the current state of institutions is shaping how we react to the crisis, but this is also a window for potentially transformative changes for the future. I’m going to spend my eight minutes or so equally on these two points. First on the AI.  During this hour of need, we are all grateful for the digital technologies that enable us not to be completely isolated from the rest of the world, but there are also dangers as well as opportunities for how we use AI. To understand that, I think it’s useful to look at what has happened in the labor market and why over the last three decades.

Here [indicating a slide] I’m showing the labor share in the US, but the pattern is similar in other OECD countries. But the US is simpler and sharper. You see a huge decline in the labor share in national income from around 2000. There is some decline going on before then, but it’s small, especially when you look at industry share, composition adjusted, it’s a very, very remarkable decline of almost 10 points in the course of about 15 years. So what’s going on with this? Well, the explanation that Pascual Restrepo and I have pushed for over the last several years with our researchers that this is mostly about automation. Partly AI, but really mostly the forerunners of AI.

One way of seeing that is in the next three graphs, and those are going to be the background on which I’ll put some thoughts on the future of AI and on this current crisis. Here, when you look at the left graph, what you see is the private sector wage bill growth in the United States. That’s so I can measure inclusive measure of labor demand growth in the private sector in the US. It’s a remarkable picture. It will be more remarkable if I also showed you wage inequality and the wages of the bottom, but it’s essentially a picture of shared growth for about four decades, which is that labor demand is growing for about above 2% a year, every year, very steadily. Wages are more or less keeping up, and then when you come to the post 1990 period, which is the right panel, you see a completely different picture.

First, the growth of labor demand becomes anemic and then it essentially stops after 2000. There’s really no growth in the wage bill or overall labor demand in the US private sector. Where is this coming from? Pascual and I argue this is from sources like monopsony, monopoly, rent sharing have also played a role. But mostly this is about the types of technologies that we have adopted. Particularly if you look at the again the four decades after world war II, the red line is what we call technological displacement, technologies that are reducing the labor share substituting machines —  mostly numerically controlled machines, specialized software, robotics, and very recently AI — that substitute these types of machines for labor. The blue is when you find new ways of doing tasks that increase labor demand. These are industries where the labor share is actually going up. You see that the blue and the red are roughly balanced and the yellow in the middle is essentially the sum of the two. So, the reason why labor demand is growing very steadily and the labor share is constant is because automation technologies are being counterbalanced by new tasks and other human friendly technology. Fast forward to the last 30 years, you see a completely different picture. The blue line here is now about 30% to 40% slower than the previous one. So there’s much less of these human friendly technologies. The displacement curve is much faster, about 30-40% faster than what it was before 1987. We are doing much more automation and much less human friendly technology. Why is that?

Well, there are a number of reasons for this. I don’t have the time to get into all of them right now, but I want to highlight one of them. This is not the most important one, but it’s one of the top three, but it’s easy to talk about and it highlights the other point that I want to make, which is the inefficiency of this capital labor substitution. If you look at the US tax code, labor taxes have been roughly constant, but capital taxes, especially on software and equipment, which are the purple and the red curves here, have been much, much lower essentially getting into the zero territory. So we are subsidizing the use of capital while at the same time taxing the use of labor. And that’s encouraging a lot of automation and much of that marginal automation is actually not super productive.

So against this background, what are we going to experience in this crisis period? I think one of the things that we are already seeing, we don’t have hard data on this, but the surveys are very clear, firms are using more and more AI in order to substitute for workers because the lockdown is making the labor supply even harder for firms, and the demand for machines is increasing exponentially. So against the background of very fast and perhaps already excessive automation, now there is a danger that we’re going to go and repeat exactly this pattern, not enough use of AI for helpings human and too much for replacing rather than the more balanced pattern of the four decades after World War II. But of course, if you’re going to hope anything will work out and not the worst case scenario of how we use technology, and what the implications of that for wages, unemployment, labor share, income distribution, we have to turn to institutions.  Institutions can actually help us redirect technology in the right way. I have argued in a lot of my work over the last two and a half decades that the path of technology is not preordained. It is firms, workers, and scientists’ choices and especially regulators’ choices in redirecting technology, and in this instance AI technology, that is going to play a critical role.

So can we hope that we have the right institutions to guide us in the right way? Actually, that’s when any type of cautious optimism that one might have becomes more jaded because we have actually seen a spectacular failure of institutions during the current crisis. This is really a combination of two things: one is that we have seen an erosion of expertise, technology, and autonomy in institutions. I think the sorry state of the CDC — which was actually very successful a short while ago during the Ebola crisis — which has been an utter failure during this crisis is related to that. But we are also seeing the role of institutions become much, much more difficult because of a collapse of trust in institutions. If you look at the trust in government and state from the world value survey, you end up with a very paradoxical and disturbing pattern. You know, in autocracies such as China, Turkey, Singapore, you have relatively high trust in state institutions and government and in democracies, including some places that have done extremely well during this crisis including Taiwan and South Korea, but also in the United States, you have very low and falling trust in institutions. That’s really making life much more complicated.

But then what does the future hold? There’s no doubt in my mind the Covid-19 crisis has created what Jim Robinson and I have called a critical juncture. There will be changes in institutions because their inadequacy has been laid bare. There are many possible futures for these institutions. I have outlined that in some talks and articles, but since time is short, let me not go into each one of them in detail. We may do nothing, which will be completely tragic. We may try to emulate China, which would also be tragic because we couldn’t emulate their good parts, such as a competent bureaucracy living for 2,500 years under an authoritarian hierarchical system. We would end up with emulating their bad parts such as a lack of respect for civil liberties and autocracy and repression. We could turn to large tech companies motivated by the failure of our government and perhaps they are better for us than the failing government. But I think there is another option which is to remake our welfare state.

The current crisis has highlighted that we need new responsibilities for the state combating inequality, climate change, pandemics, better regulation. But I think a lot of people are worried whether that’s going to happen starting from the current sorry state. They’re worried as Hayek was after the Beveridge report, which led him to write Road to Serfdom, whether once the state becomes very powerful, economically much larger, much more administratively in control of wages and allocation of resources, if that’s going to be a tenable state. Well, this is actually my last slide and I’ll conclude that this is actually what James Robinson and I tackled in our new book, The Narrow Corridor. We came up with a framework for arguing why Hayek was actually wrong and there was a way for society to adapt to greater state power as long as they deepen democracy, and we outlined the dynamics of that. Since time is short, let me not get into the details of that with the hope that somebody will ask during the Q&A session and I can provide more explanation about what the main thesis is and why despite all of the difficulties that we’re facing, I’m saying that a little bit of cautious optimism might be possible. Let me conclude here and pass it to other panelists and come back to these issues during the Q&A session. Thank you.

Anton Korinek:

Thank you very much, Daron, for your insightful remarks. And let me now turn it over to Diane. But before doing so, let me also make an announcement to all of us attending the webinar. Please feel free to click on the link “ask a question” at the bottom of your screen and to add any questions that you may have for our panelists. You can also upvote existing questions that other people have already posed.

Diane Coyle is an economist and former advisor to the UK treasury and the Bennett professor of public policy at the University of Cambridge where she has also codirected the Bennett Institute since 2018. She was vice chairman of the BBC trust, the governing body of the British broadcasting corporation and was a member of the UK competition commission from 2001 to 2019 and she has just published a book on Markets, State, and People: Economics for Public Policy.

Diane Coyle:

Thank you. Hello everybody. It’s a great pleasure to have this opportunity. We panelists haven’t coordinated beforehand, but I think what I’m going to say complements Daron’s comments without overlapping with them. I’ve got about eight minutes, and I want to make three main points. The first is that the crisis has crystallized tensions around expertise and to what extent modelling can or should inform policy choices. I think we need to reflect on the lessons for AI because machine learning systems are technocrats par excellence. The second point is that the companies that can operate AI at scale are being strengthened by the crisis and they’re going to emerge even more powerful. So we need to double down on the policies that will make them accountable and make the markets in which they operate contestable. The third point is that data is everything and we’re going to need to understand much better the creation and the distribution of value in data value chains and the trade-offs between private and collective benefit.

Let me start with the first of those: expertise. Machine learning systems have been programmed and trained to act just like Homo Economicus. They maximize some well specified objective functions subject to constraints, and they use the rules of logic. In the pre-crisis world, this was already problematic when machine learning systems were starting to be deployed in public policy decisions. We live in a very complex socioeconomic system, there are multiple conflicting aims and tradeoffs. Just as target setting can distort public sector behavior, AIs can game the objectives that they’re set. And there are what political scientists call “incompletely theorized agreements,” what others might call political fudge, which means that quite often we don’t want to specify too clearly what objective we’re aiming for in order to achieve some consensus on actions. We’re already seeing during the crisis the kinds of problems that economists are very familiar with using models to forecast. You get model drift and you get what’s called the Lucas critique where structural breaks mean that the relationships that you’ve modelled breakdown. In some domains, this doesn’t matter — if you’re thinking about algorithms determining online shopping offers, the fact they’ve broken down doesn’t matter, they’ll fix those quickly. There are also some quite reasonably narrowly specified domains in which AI is proving really useful at the moment in biomedical discovery. So I think the main lesson of the pandemic is actually about the limitations of the kind of model we can build and train. And I think we’re really far from the complex interactions that policymakers need to consider now. The genetics and other aspects of the virus itself, individual and group susceptibility, social and economic conditions, behavioral responses to the pandemic and lock down policies, responses to climate and so on. This is really, this is really complicated.  I think this is a real lesson in learning the limitations of what we can and should be trying to do with AI in policy.

The second point is about power. Before the crisis, a number of countries around the world were recommending tougher competition and regulatory policies toward big tech. Now big tech is getting even bigger. This is a moment for governments to hold their nerve, as our society is dependent on digital companies as never before. And that means that they even more than before, they need to be held accountable for the market power and also political power that they hold. We need a lot more thought about the governance of AI. And I welcome Allan’s comments about that in his introduction. Can we avoid a geopolitical arms race? What are the national and global institutions that will deliver accountability? With some of my colleagues in the Bennett Institute, we’re starting a project looking at the history of different governance frameworks for new technologies. It’s not straightforward. It depends on the cost of entering the technology. If the technology changes the governance frame, what needs to change as well? It depends on the market context and what the interaction between public and private sectors looks like with developing new technologies and so on. I was a member of Jason Furman’s panel in the UK looking at competition in digital markets. Just before the lockdown, the government announced that our recommendation for a digital markets unit to be set up will go ahead and that needs to happen. It needs to happen in other countries. I think the more we can get regulatory alignment and alignment of competition policies between countries, the more effective we will all be. We also need to reflect about the skills base and national capabilities in AI. If you’re sitting as I am in Europe and London, being in between the United States and China — who have the leading capabilities at the moment — when they seem to be embarking on a phase of geopolitical rivalry and AI is part of that, that’s not a very comfortable position. And so for everybody, for a number of reasons, thinking about sharing the skills needed to use AI and deploying it and building up those national capabilities will be important.

My final set of comments is about data. It’s one of the barriers to entry that we identified in the Furman review. We recommended looking at and enforcing interoperability and enforcing rules about APIs and some data sharing. The key issue emerging in the pandemic is health and location data. I think that’s been really unfortunately shaped by the narrative that data is personal. Almost no data is personal. There might be quite a lot that we want to be kept private, but that’s a different matter because the information content of data is almost always relational and contextual. Daron has an excellent paper about negative externalities of potential privacy loss from the provision and sharing of data. But there are also substantial potential positive externalities from aggregation and the sharing of data with colleagues. I have a policy paper on that and we are working on an academic paper also. In the current context of the pandemic, my health and location status has substantial implications for other people as a very large externality. That to my mind outweighs concerns about privacy, not data security, but privacy in the sense of not sharing data at all in the context of the huge civil liberties removal from lockdowns in various countries. We shouldn’t have to be relying on the good will of companies like Google and Apple to provide limited data on what’s happening during the lockdown, and the APIs that they are developing. The democratic public interest in this is too large. I think that the Covid example is an instance of a much broader debate that we need to start having about data, about the positives and the negatives, the individual and social, about how to capture that value and how to distribute the benefits. Also about what kinds of institutions can be trusted to govern data and data access both in terms of security and privacy, but also in terms of the rights of access to various forms of information that can be used for the good of individuals and the good of the public. And I will stop there. Thank you.

Anton Korinek:

Thank you so much for your insightful remarks. Let me now hand the microphone over to Joseph Stiglitz, who is an economist, public policy analyst and University Professor at Columbia University. He is the recipient of the Nobel Prize in Economics in 2001 and the John Bates Clark medal in 1979. He is also a former Senior Vice President and Chief Economist of the World Bank and the former member and Chairman of the US president’s Council of Economic Advisors. His most recent book is Measuring What Counts; The Global Movement for Well-Being. So Joe, the floor is yours.

Joseph Stiglitz:

Thank you very much, Anton. It’s really good to be here. I again join the others in saying that I wish it could be in person. I agree with the point that you made in the beginning that we’ve learned a lot about our society, about our economy, about our government from this pandemic. It’s like pathology in medicine, you learn a lot from putting the system under stress. And I think at least in the United States, we found things didn’t go quite as well as we would have hoped. We’ve seen a lack of resilience. Our private sector, our markets couldn’t even produce masks and protective gear and distribute them to where they were needed. We’ve seen the importance of government. We all turn to government in times of disaster. And this is clearly a time of disaster. We’ve seen that 40 years of denigrating the role of government has actually worked. It’s worked in weakening the institutions. Daron pointed out the weakening of the CDC, which had been a very strong institution, the abandonment of the White House Office of Pandemics. We had created institutional structures designed to prepare us for a pandemic, but then a weakening of institutions led to the abandonment of those institutions.

As we look at what has happened, it’s natural to think about this in relationship to other crises, and we can see some shared underlying factors. In the 2008 crisis, the last crisis, we saw a weakening of the state, with financial deregulation being one of the conditions leading to the crisis. Then again we saw short-sighted behavior on the part of the banks leading to the crisis. Here, it’s short-sighted behavior on the part of firms relating to an economic system that lacks resilience.

We’ve also seen in this crisis that this is not an equal opportunity disease. It goes after those with poor health. Those with poor health are disproportionately people who are poor, especially in the United States where we have not recognized the right of access to healthcare as a basic human right. I’ll make some comments later on that. Wealth inequality is clearly part of the preconditions that have exposed the United States so strongly to the disease, and one of the reasons why we’ve had the highest rate of death. The problem of inequality is going to be exacerbated by the crisis. And I’ll try to explain that.

But the topic of the seminar is about AI. AI is a major structural change in the economy. One of the things that we’ve seen over a long period of time, and that is going to be exacerbated by the pandemic, is that markets don’t handle these kinds of large structural changes well. That’s not one of the strengths of markets, and it inevitably requires government assistance to manage that. What we’ve seen is maybe not so optimistic. There are three things which are going to reinforce, I hope, what has already been said. The first is that the long-standing weaknesses of the American economy, but also other economies, have been exposed. The second is that there’s a clear possibility of further adverse effects from the pandemic. But third, echoing what’s been said, that it’s not inevitable. It’s a matter of policy. And then the final question which I won’t get to, I hope we get to in the Q&A, is one that Daron raised at the end of his discussion: the question is whether we actually do what we could do. And that is a matter of how our democratic institutions respond.

Let me begin by noting one aspect of the pandemic, that it has led to a fundamental shift to the cost of labor versus machines or robots. Daron pointed out very clearly that what has been happening is a shift in technology, labor-replacing versus labor-augmenting innovation. That is one of the reasons why the labor market is not working well, why the output share of labor has gone down. This pandemic has emphasized, even increased, the virtues of robots. Robots don’t get the coronavirus (though, obviously, computers do get computer viruses). And there is an ongoing war in both spheres with some uncertainty and some hope that the good guys, the antivirals, will win over the virals. Robots, even if they do get viruses, don’t need to be socially distanced. And all of this adds to the shadow price of labor. It makes labor less attractive relative to capital. And that will exacerbate, I worry, some of the trends that Daron talked about. There was an interesting article this morning in the New York Times about a city in the UK where robots are being used for deliveries. They had already set up a company before the pandemic, but after found a vastly new market. If this is so, it will mean the problems with unemployment and inequality that we’ve been facing before Covid-19 will be even worse.

There is a failure to design an adequate response in the United States to growing unemployment. The unemployment rate in the United States is clearly already at 20%, and a broader measure of unemployment, which we call U-6, is clearly north of 25%. And this growing unemployment is in spite of massive spending — almost $3 trillion fiscal support, and an equivalent amount of monetary support. What is equally disturbing is an unwillingness on the part of some to continue to support this spending, even though it’s obviously needed. That’s obviously very worrying, it’s a clear sign that some aspects of our solutions may not be working as well as they should.

At one level, one can say it’s not a surprise that things didn’t work out as well as we would have hoped. Everything had to be done in a rush. But the fact is that countries all over the world had to do it in a rush. In some countries, the institutions actually worked. That’s a hopeful side. In New Zealand, not only did they avoid the massive increase of unemployment that they had in the United States, the disease was almost brought down to zero, and there’s strong social cohesion. Other democracies have done so as well. As Daron said, some of the more authoritarian countries have also brought down disease numbers, but in ways that obviously wouldn’t be acceptable to us. But the good news is that there are countries like New Zealand and South Korea who are democracies that have brought it down, have gotten the disease under control.

What is most disconcerting is the marginally different perceptions, the beliefs about the disease and its consequences and what to do about it, reflecting and deepening pre-existing divides. And that goes to the point that Daron and Diane emphasized – the importance of trust in science, trust in experts. In large parts of our society there is that lack of trust, and that’s been exposed very strongly by the pandemic. To me that suggests that we may not be able to respond appropriately to the enormous social and economic challenges that AI may present going forward. Now, some have suggested that the pandemic will, in the short run, reduce the problems posed by AI and robotization because it is causing onshoring. But I think that’s an overly optimistic note. Onshoring will be done by robots or by machines more broadly. The jobs that have been lost to robotization and de-industrialization won’t be regained.

In fact, as I said earlier, the short-run impacts are going to be just the opposite. Those who can do remote work will work remotely – the high-tech workers have been relatively little affected. We’ve gone on with our teaching on Zoom. It’s the others, the people who work in the restaurants that have faced job losses. In the short run, the problems of inequality to which I refer are likely to get worse. The disease has exposed and is likely to exacerbate these inequalities.

In the United States, the disease has also exposed the weaknesses of our whole system of social protection. The fact that America has the least adequate system of social protection, such as paid sick leave. This really illustrates the worries about our institutions. Congress recognized the importance of paid sick leave. We don’t want people who are sick with COVID-19 going to work. And since almost half of all Americans are living paycheck to paycheck, if they get sick and there is not paid sick leave, they have to go to work. Congress passed a law requiring paid sick leave just for COVID-19, but then, under the lobbying of major companies, those companies with more than 500 workers were exempted. Now that reflects a kind of short-sightedness on the part of the companies — you can say it also reflects a lack of humanity. It reflects an inadequacy in our political process that they would let this group win the day. These companies employ almost 50% of all workers in the private sector.

Another example is that we have asked workers to go to work without protective gear. We have an agency within the government called OSHA, that’s supposed to protect workers, but it has still not issued regulations concerning the disease. I referred earlier to the lack of resilience in our economy, but it’s a lack of resilience for which the poor pay the highest price. Indeed, the rapid restructuring of the economy, accelerating change already going on, such as in retail, will create a pool of unemployed that would, even in a normal recession, take some time to work off.

The next point I want to make is the same point that Diane emphasized: the restructuring of the economy has advantaged large digital firms. Firms which have large elements of monopoly power, related in part to superstar and network effects. The problem of the lack of competition in this key sector – something that I talk about in my book, People, Power and Profits – is getting worse as a result of the crisis. So too will the problems of inequality, which are linked to this monopoly power, and to some of the other effects I talked about. In the medium term, we shouldn’t have a problem there, but our politics may lead us to have one: We will need massive investments for the green transition; there are gaps left by underinvestment in the last 20 years in our infrastructure. These gaps should necessitate more job creation than we will be losing. But that will require government revenue, and that’s the question – will we have the political will to make these investments?

I have even more worries about Africa. The cheap labor that enabled export growth in manufacturing goods was at the center of the development strategy in East Asia. And that won’t be working in Africa. As I said, we shouldn’t have a problem in the medium term. And in the longer term too, we shouldn’t have a problem. We should be able to use our tax system and intellectual property rights system to ensure the benefits are shared by all. This is particularly important in light of COVID-19, which can be viewed as a large negative technology shock. Negative technology shocks or similar events give rise to distributive battles: who will bear the cost of the reduced standard of living? Such distributive battles can be particularly ugly in countries lacking a certain degree of underlying social solidarity such as we’ve seen in the United States.

I want to end on a couple more positive notes. The first is that we should be able to steer innovation. Steering innovation to what has been called intelligence-assisting innovation rather than labor-replacing innovation. Maybe that itself is a problem which AI could be trained to do. Daron emphasized the problems of misguided incentives of encouraging labor-replacing innovation. There are others too: The fact that monetary policy has kept the cost of capital down to a negative real interest rate obviously exacerbates the problem of the incentives to have human-replacing robots. But if we can steer innovation in another direction, then the problems that we have with AI will be mitigated.

The second more positive note is that government has never intervened more strongly in the economy. Never has there been so much spending and so much lending, where in the midst of this pandemic, they’re making life and death decisions over enterprises.

We are shaping the economy or failing to do so. The choices we make now will have long-lasting effects. So we have the potential to use conditionality on public lending programs in ways that can really reshape our economy and make us better able to handle the problems of inequalities we’re facing, and some of the governing problems we’re facing. The problem is, will we have the institutions that will direct this money to try to create the post-pandemic society and economy that we like? So far in the United States the answer is no. So far in other countries, the answer is partially yes. Let me stop there and we can have a discussion.

Anton Korinek:

Thank you so much Joe. Let me now also bring all panelists on screen. We have just heard three really thoughtful perspectives on the effects of the pandemic on our economy and also how to think about our societal response to other large shocks.

I thought I would start the panel discussion by posing a perhaps somewhat personal question. What has surprised you over the past few months, and are there any specific lessons that you feel you have learned that give you a new perspective on how easily or how our economy and our society can adapt to large shocks? And how should this inform how we react to the prospect of ever more automation?

Daron Acemoğlu:

Let me make one remark which will be a partial answer and also a riff off what Diane said because I think it’s going to be an illustration of the power and the dangers of technology and our governance challenges. Before this crisis I was probably close to one extreme on issues of privacy, in that I saw the control of data by governments and the control of data by companies as a real threat to democracy. I have partially changed my mind in the way that Diane already anticipated. It is clear in the midst of the pandemic that data sharing, use of data on infections, and contact tracing are all critical for saving lives. So how do you square that with the issues that I worried about? In fact, I think this is a critical test case for some of the issues that both the other panelists and I talked about. I’ve been somewhat frustrated by conversations I’ve had over the last few weeks with computer scientists, who a year ago would have not paid sufficient attention to issues of privacy and their importance to democracy, who now object to use of data sharing in order to combat the pandemic. I think all of these conflicted responses are an implication of our inability to visualize and understand and imagine a better governance for data.

I think in an ideal world, what we would say is that of course right now we have to use all the data we can in order to combat the pandemic. But then do that with a proactive plan for doubling down on protecting privacy as soon as the pandemic is over. That means both controlling the use and abuse of data by governments and controlling and containing the use and abuse of data by companies. Now the question I think is that some people come down with a very different conclusions because they have different views on what is feasible institutionally. For example, if you believe that once you open the gates to companies or governments using private data, you can never take that back, you’re going to be much more cautious. If you think that our institutions are so badly failed at the moment that we can never double down on protecting privacy and strengthening democracy, you might have a very different view. I think this privacy issue is a test case and I do actually still retain a cautious optimism that recognizing the issues and publicly debating them and understanding what sorts of institutions can deal with them will open the way to a better governance of data. That is actually very related to the governance of AI.

One of the questions that I saw is do we need broad institutions that protect us in terms of inequality, public safety, and democracy, or do we need technology-specific institutions? I do very much believe that broad institutions are the first line of defense, but we do need technology-specific governance structures and data is one of them. And AI is another, both because of its ability to change the political discourse, to transform privacy and political activism by individuals, and also because of its labor market effects. Because I think replacing labor with machines, sometimes it’s very productivity enhancing, but it also has external effects because it really damages the very fabric of society. It has to be balanced out with other social objectives. Thank you.

‍

Diane Coyle:

I think Daron is absolutely right to point out that this is a critical moment for thinking about the kinds of institutions that we trust to handle data and technology more generally. So that kind of thinking is really important. But to answer your question more directly, the thing that struck me is the way that people sit in intellectual silos, even in the face of a major crisis like this. We’re a self-selected sample of people who talk to computer scientists a lot, and so already crossing disciplinary boundaries in that way. I’ve been quite struck in all the discussions I’ve observed in UK government and elsewhere that medics talk about medical issues, geneticists talk about genetic issues, the epidemiologists and the economists, maybe they’re starting to talk to each other. This really highlights to me the importance of thinking about ways to integrate social science and different strands of science because the problems that we’re facing — be it the covid pandemic or climate change or geopolitical disruption — these don’t fit into narrow silos. That surprised me and concerned me. I hope we can also take this opportunity to do some more of that joining up because if we’re putting a lot of effort into medical innovation only and not into the social context of the institutions so that people would trust the health system that will deliver it, then we’re going to fail in tackling this crisis.

Joseph Stiglitz:

I agree very much that the key is creating institutions. I’m optimistic that we can create them, but let me express a concern that one has to go a little bit beneath that. The question is why isn’t there trust in our institutions? And why should there be some skepticism? Well that goes back to, you might say, the word power or inequality in our society. If we think that Facebook is in one way or another going to write the rules, we’re not going to feel comfortable with the rules that come out. And if we think our society has a lot of inequality, which it does, and that we have a political system where that economic inequality translates into political inequality, then we’re not going to trust the institutions that emerge out of the political process that are supposed to protect us. They’ll be protecting the one 10th of 1%. That’s why I’ve always said at the root, we have to begin by dealing with the underlying problems of inequality, the problems of ensuring that we have competition, and of course, that’s interactive. How do we do that without good institutions?

Let me give one more example that’s a little bit different from the data problem, but one that is of great concern to me both before the pandemic, but made very clear by the pandemic, and that’s misinformation. The concern about spreading misinformation about the pandemic response, which has been a major problem. What’s interesting about that is before the pandemic, Facebook and other technology companies said they didn’t have the technology to address problems of misinformation. None of us really believed it because AI has the technology, not necessarily to do it perfectly, but to do it reasonably well. Then finally, when it became clear that it was our country’s health was at risk, they did come forward and say they were going to take down misinformation about responses to pandemic. But they feel very hesitant to take down misinformation about the pandemic put up by political leaders. So again, a political and institutional decision which is obviously a problem.

Finally, let me say, in response to your question about what has surprised me, one of the things that surprised me was the willingness to come up with a sizeable response, on the one hand, and the magnitude of the failures in the design of response on the other, which I find quite colossal. It wasn’t like they didn’t know about the alternatives that were being discussed. And the third is the willingness of one of the two parties not to have a comprehensive program and not to have a sustained program, saying we ought to pause now.  The social divisions in our society that are forming over this issue are actually a surprise.  We can’t even, on this particular issue, come to some agreement about reality.

Anton Korinek:

Thank you Joe. A number of people have posed questions in our question box around a familiar theme that automation will on the one hand create more abundance, but on the other hand, we are concerned about whether the resulting prosperity will be shared or whether will just benefit the few. What is your take on this question and do you view it differently in a post-COVID world? Are you more optimistic or maybe more pessimistic on how we can resolve this tension? Let me maybe go in inverse order now, let me start with Joe and then Diane and Daron.

Joseph Stiglitz:

Absolutely, the fact that we have more resources means in principle every group in our society could be better off. I alluded very briefly in my introductory remarks to the fact that we can use intellectual property rights, taxation, we have lots of incentives that we can use to make sure that the benefits are shared. Part of that is competition policy to make sure that you don’t have an agglomeration of market power. There are lots of things that we could do.  I guess I have an ambiguous reaction coming out of the pandemic whether we will. On the one hand, I certainly get a very strong feeling that a lot of people have realized that the pandemic has exposed the magnitude of inequality in our society and a lot of discussions of inequality, unfairness, and a lot of resolve to deal with that. On the other hand, the point I made before, the kind of divisions in our society that have led some of the people who should be the most strong advocates of pro-equality policies to actually resist the kinds of policies that would enable us to more effectively deal with the problems.

‍

Diane Coyle:

If you think about the 19th century, technological change and automation brought about a long period of great inequality and low wage growth. Then if you think about the 1950s and 60s, which saw a lot of automation, we had the opposite outcome. We had reduced inequality, lots of good jobs for middle class people, and rapid wage growth. And so the question is how can you steal yourself into that mid 20th century pattern rather than that late 19th century pattern? One of the keys for me is about the skills that you need. And anybody who deals with the big data sets and AI now knows that actually handling the data is a craft skill, and people don’t have any very systematic ways of passing on that skill. It’s a learning by doing system. You learn it at the feet of the master and you gain those skills yourself. And so what we need to do is both make the technology itself more routine and change the provision of the supply of labor, the people with skills, and make sure that it becomes less of an inequality machine than it has been to date. And the pandemic is probably an opportunity to start to create some of those skills and think about that because governments are going to have to think about it, how to avoid the scarring on the large groups of young people coming into the labor market and needing to find themselves a good career and a good job prospects. So on balance, I think I’m probably a little bit optimistic about that, but this is very uncertain. Who knows?

Daron Acemoğlu:

I think this is a really interesting question. I’ve thought a lot about it. I’m going to just ever so slightly disagree with the other panelists in the sense that I think even though automation is an enormous engine for productivity growth, it is also potentially very disastrous if the attitude is “automate everything in sight.” And the reason for that is threefold.

First, it isn’t actually true that automation always increases productivity. Automation has the promise of increasing productivity, but if it involves substituting machines that are only slightly more profitable than labor, it doesn’t increase TFP. And if there are policy distortions such as the ones I hinted at, and there are many others related to labor market structure, it may actually reduce TFP.

Second, my belief on the basis of my work and data analysis, is that periods such as the one that Diane said, middle class wage growth, broadly shared prosperity, stable or sometimes even declining inequality, though they coincide with automation, critically depend on other technological changes, periods that have mostly automation and no other technological changes have never brought that kind of prosperity. And the reason why economists have often not been as clear on this is because we have imposed on the data a way of looking and models that have only one type of technology, blinding ourselves to the critical question of which types of technologies are doing what. So automation can increase productivity but it’s generally a force towards greater inequality and slower wage growth. It needs to be counterbalanced by other technology.

That brings me to reiterate what I said earlier, technology policy, redirecting technological change away from just automation, especially for AI which has so much promise to be complementary to humans, is critical. My reason for being very worried about “let’s just do AI on everything and get rid of the troublesome humans which are now proving to be more troublesome because they can get Covid-19” is because I think we have also no great experience of generating shared prosperity based on redistribution. There was a question on predistribution and I very much agree with that question. Predistribution is critical. So it’s very difficult both for political reasons but also for social reasons to create a harmonious, well-functioning democratized society when everybody depends on bread and circuses out of the hands of the governments or this new version UBI.

We really need people to be earning less unequally distributed wages, and that means middle class wages generated by the technological working conditions, and bargaining situations in the workplace. That’s going to become more and more difficult if automation just gets out of control. Of course, redistribution helps, especially for a social safety net, providing public services and keeping in touch through progressive means, what types of incentives that are at the top of the distribution, but it can never replace the market system generating more equal wages. And that will never be possible if we double down on automation because if you just have more and more automation technologies, bargaining power cannot survive. If workers ask for higher wages, firms will just shift to machines that are getting better, whereas humans are not getting better. So it’s absolutely critical that we have ways of investing our ingenuity, especially in the field of AI, to make humans more productive as well, not just machines. Thank you.

Anton Korinek:

Thank you, Daron. And you have just touched upon the next question that a member of the audience has posted which was on predistribution versus redistribution. So I wanted to ask Diane and Joe if they could also share their thoughts on this question with us.

‍

Diane Coyle:

I don’t really have a lot to add on that. I mean the point about increasing labor skills is the point about predistribution shaping the configuration of labor supply and demand. And that’s exactly why I put emphasis on that in my previous answer. One other point to make perhaps is looking back again at that history, the role of institutional innovation. So we’ve been talking about automation, but we might also want to think about in this context what kinds of new institutions might we want to see emerging out of this? And they might not be able to deliver financial redistribution or pre-distribution, but they can alter things like the distribution of social capital, the distribution of natural capital among people. And, you know, income matters a lot, finance matters a lot, but these are also other assets that people really need.

Joseph Stiglitz:

I just want to say I strongly agree with what both Diane and Daron said. I think there should be a focus on pre-distribution or what used to be called just market income. I want to just add that it’s a comprehensive issue of the rules of the game and the investments. What do I mean by that? We’ve talked about competition policy, also corporate governance policy. One of the sources of inequality are the CEOs being able to shape what the firm does and getting more for themselves and making decisions about labor-saving innovation versus other kinds of innovation. If we add more representation of workers on boards, we might get different decisions. They might not view labor as an irritant but as part of the objective of society.

One of the striking things about the pandemic, as I mentioned, was that employers did not provide sick leave or protective gear. In many cases it was only the unions that succeeded in getting that kind of protective gear. So that’s an extreme manifestation of the lack of social responsibility or short-sightedness on the part of corporations. And I mentioned before, monetary policy, which changes the incentives of labor, of intelligence-assisting innovation, which strengthens the productivity of labor and increases the demand for labor rather than labor-replacing innovation. There are a whole set of policies that go to shape how our economy works, and which affect the market distribution of income. And we ought to be focusing a lot more on that.

Anton Korinek:

Thank you. We are almost at the end of our webinar and time is always way too short. But I wanted to ask our panelists if they are willing to leave us with just a very short thirty second parting thought on this theme of what we can learn from the pandemic for the future of governing AI. And let me go through in alphabetic order again. So with Daron first.

Daron Acemoğlu:

Let me agree with one thing that Joe said, which is the ability of most Western societies and beyond to actually respond to the crisis with large stimulus packages to deal and in a general agreement within society despite a lot of misinformation that you have to deal with this problem both at the level of containing the virus and healthcare systems being bolstered, I think are hopeful signs that when push comes to shove, there will be some agreement on key issues. That is the only sort of straw we can cling to in terms of remaking institutions in the future.

Diane Coyle:

I certainly think the mood has changed. People are ready for a different kind of system. They’re very aware of the many inequalities that have been exposed and exacerbated by this crisis. So that makes this an opportunity. And there is a cliche, don’t let a good crisis go to waste, grab that opportunity. My concern about it is that we have already recently let a good crisis go to waste in 2008. We did far less than I expected coming out of that. All of us who are engaged in this debate really need to make sure we grab that opportunity now.

Joseph Stiglitz:

I agree very strongly again and in fact, maybe it’s one of those instincts where the second time around you actually learn the lesson that you should have learned the first time around. And the lesson is very much that we need a better balance of the market and the state. We put too much on the view that markets will solve all problems, and we didn’t realize how you need to have regulation, you need to have public investment in science, you need to have good institutions, you need to have trust in experts, you need to build up trust in these institutions rather than bringing them down. You won’t be able to get that unless you have societies with more solidarity, and that kind of solidarity will only be achieved if we get a society with more shared prosperity, more equality. So that agenda of equality is both the object of what we’re trying to get, but also a necessary condition to get the kind of society that we want.

‍

Anton Korinek:

Thank you very much, Daron, Diane, Joe, for sharing your thoughts with us. Thank you to everybody in the audience who has joined us today. I should also let you know that all three of our panelists today have agreed to give a full webinar on more specific topics in the coming months. So please check back on our website frequently as we announce future events. I hope to see all of you back with us soon. Goodbye.

By Jesse Peters and Tessa Trapp

Blogpost 27/2024

Transparency and environmental policy are two key issues in the upcoming European Parliament elections. In this regard, the General Court’s (‘the Court’) ruling on 13 March 2024 in the case of ClientEarth and Leino-Sandberg v Council provides some highly relevant insights. The Court annulled two Council decisions refusing to disclose the Council Legal Service’s opinion on the 2021 proposal to amend the Aarhus Regulation. While the Court’s critical approach to the Council’s justifications for secrecy is to be applauded, and the outcome of the case is certainly to be welcomed, this post suggests that an alternative route to reach the same conclusion would have been more desirable. The Court now seems to deliberately gloss over the document’s potential legal and political significance, turning a blind eye to the heated and ongoing debate on the Union’s (non-)compliance with the Aarhus Convention. Instead of downplaying the relevance of the document’s content, we argue that a more principled emphasis on demanding openness in the realm of environmental policy would have led the Court to the same outcome but would have also made the Union’s transparency framework more robust, in line with the objectives of the Aarhus Convention.

The EU and the Aarhus Convention

The requested document was produced by the Council’s Legal Service in the process of amending the Aarhus Regulation, which presents one aspect of the Union’s implementation of the Aarhus Convention. The Aarhus Convention is an international agreement, which the Union approved in 2005, aiming to improve public access to information, public participation in decision-making, and access to justice in environmental matters. The Aarhus Regulation, adopted in 2006, applies the various provisions of the Convention to the Union institutions. At the time, the internal review mechanism of Article 10 of the Regulation was considered the most promising creation, which allows non-governmental organisations and other natural and legal persons to request reconsideration of certain administrative acts or omissions by the adopting institution. Through this administrative review mechanism, the Union aimed to provide a legal avenue for applicants who do not qualify for standing under Article 263(4) TFEU due to the restrictive criteria of direct and individual concern. The Union thereby aimed to meet the requirements of Article 9(3) and (4) of the Aarhus Convention, which obliges to allow members of the public broad access to effective review mechanisms to challenge acts and omissions that contravene environmental law.

In 2011, the Aarhus Convention’s Compliance Committee (ACCC) already indicated that the restrictive scope of challengeable acts via the internal review mechanism of the Aarhus Regulation might not be sufficient to ensure the Union’s compliance with the Convention’s access to justice obligations. Due to the refusal of the Union courts to depart from their restrictive case law on the standing of natural persons under Article 263(4) TFEU established in Plaumann (and clarified later for example in Greenpeace, Danielsson, UPA, Jègo-Quéré, or Carvalho), as well as their narrow interpretation of relevant provisions of the Aarhus Regulation (for example in Stichting Milieu, LZ or Trianel), the ACCC eventually adopted a decision in 2017, confirming the Union’s non-compliance with Article 9(3) and (4) of the Convention.

The main aspects of the Union’s non-compliance were that only acts of individual scope, adopted under environmental law, and having legally binding and external effects could be challenged via the internal review mechanism (see the ACCC’s 2017 Decision, particularly paras 94-104) and that members of the public other than NGOs could not request such review (paras 92-93). This led to most internal review requests being declared inadmissible.

Following this established non-compliance, the Commission proposed amendments to the Regulation, which would now allow for the challenge, within the internal review mechanism, of acts and omissions regardless of their personal scope that more generally contravene environmental law, and that have legal and external effects (for more detailed considerations of these amendments, see for example Brown, Leonelli, or Pagano). In February and again in July 2021, the ACCC assessed these particular proposed changes positively. An agreement on the amendments was reached in the trilogue negotiations in July 2021, and in October 2021, the amendments were officially adopted in Regulation (EU) 2021/1767.

The Document Request and the Judgment 

It is within this revision and negotiation process that the legal opinion at the core of the dispute in ClientEarth and Leino-Sandberg v Council comes into play. The currently only partially available version of the requested document contains a (legal) analysis of the findings of non-compliance of the ACCC, as well as a proposal for next steps to be taken, also in light of the (at the time) upcoming Meeting of the Parties to the Aarhus Convention (MoP). The crucial question then is why the Council, after providing only very restricted access to the requested legal opinion, still refuses to grant full access to this document. This question is all the more pertinent as the relevant negotiations have been closed and the changes to the Regulation have already long been adopted, leading the Court to quickly dismiss the argument that disclosure could undermine an ongoing decision-making process (Judgment, para 100).

The Council feared that full disclosure of the document would have two negative consequences for the Union. In its view, disclosure would threaten its ability to receive high-quality advice from its Legal Service because disclosing the full analysis invites external pressure and litigation due to its broad scope. Furthermore, disclosure would in the eyes of the Council hurt the Union’s ability to act effectively on the international stage. Both of these concerns relate to grounds protected by the Access to Documents Regulation, which contains exceptions to the general rule that Union institutions need to disclose documents.

The Legal Advice Exception

With regard to the Council’s first concern, the main dispute centred on the question of whether the document contained information sensitive enough to argue that disclosing it would endanger the Council’s ability to receive frank, objective, and comprehensive advice. Ever since the ECJ’s Turco ruling, institutions withholding access under this ground need to do more than describe an abstract worry. Instead, they need to “give a detailed statement of reasons” why they believe the legal advice in question is “of a particularly sensitive nature or [has] a particularly wide scope” (para 69).

To that effect, the Council in this case cited ‘external pressure’ and the large number of cases brought before the Union courts as evidence of the contentious nature of the subject matter (Judgment, paras 63 and 71). In such a controversial area, disclosing a broad legal discussion of the Union’s compliance with the Aarhus Convention in light of the proposed amendments could add fuel to the fire, and in turn, make members of the Council Legal Service hesitant to present their honest opinions in the future.

The Court deemed the argument based on the existence of ‘external pressure’ completely unsubstantiated (Judgment, para 65). This observation is to be applauded, given that the ‘external pressure’ in question amounted to nothing more than quite measured comments by NGOs and academics, including on this blog (Council Replypara 37). Especially in legislative procedures, it is striking that the Council views critical engagement with the Union’s policies as ‘external interference’ rather than healthy signs of public engagement in the democratic process.

The second concern, regarding the broad nature of the legal analysis, and the related risk of litigation, was taken more seriously by the Court, as it acknowledged the many legal challenges against the Union’s compliance with the Aarhus Convention. However, the Council did not explain specifically how disclosing the document at hand would negatively influence such procedures. Indeed, how could legal advice that was not negative about the Commission’s proposal make it more difficult to defend the eventually adopted Regulation in court (Judgment, para 75)? Finally, the Court stressed that the amendment of the Aarhus Regulation could not and did not entail consequences for the standing criteria laid down by Article 263 TFEU. Thus, disclosing legal advice on the relation between the internal review mechanism and the remedies provided by the Treaties was considered unproblematic (Judgment, paras 84-85).

The International Relations Exception

The second ground for refusal by the Council related to the Union’s international relations. In the case law on this exception, institutions have generally presented two main rationales for secrecy (see Peters and Ankersmit for an overview). The first concerns information that reveals strategic objectives and tactical considerations, because external actors could in turn use that information to the detriment of the Union. The second main reason stems from the fact that certain documents are shared with the Union on a confidential basis and disclosing them could hurt the climate of confidence.

The Council in this case employed the first rationale, stressing that revealing the legal analysis would ‘compromise the Union’s position vis-à-vis the other parties to the Aarhus Convention’ (Judgment, para 107). In line with previous case law such as In ‘t Veld v Council, the Court required more than a mere fear, but rather an argument showing ‘how disclosure could specifically and actually undermine’ the Union’s interest in international relations (Judgment,para 108). Given that the ACCC itself had in fact recommended the adoption of the amendment to the Aarhus Regulation, and the Council’s Legal Service opinion in question was not negative to or critical of the amendment (paras 115-116), the Court failed to see how disclosure could weaken the Union’s position in negotiations with the Convention parties.  

Simply a Piece of Uncontroversial Legal Advice? 

In general, the Court’s critical approach to the Council’s fears signifies a positive development in the case law concerning access to documents. As has been argued before by Leino-Sandberg, Union institutions generally showcase an attitude of ‘exasperation and foot-dragging’ when it comes to publishing legal advice. Moreover, in previous cases, the Court itself has been dangerously deferential to any justification presented under the ‘international relations’-exception. The fact that the Court carefully scrutinised the Council’s arguments and did not take the presented worries for granted is a laudable approach that brings the Union more in line with its own commitment to transparency (Article 1(2) TEU).

Still, the judgment relies on an assumption that can be viewed critically. The Court seems to infer that the concerned legal analysis cannot invite external pressure, litigation, or tough negotiations with Aarhus Convention parties, mainly because it does not take a negative stance towards the legislative proposal. However, based on the available information (and lacking knowledge of the full document), this assumption seems far from self-evident.

While the judgment only contains the positive comments of the ACCC on the 2021 amendments to the Aarhus Regulation (Judgment, paras 10, 18, and 92), the actual negotiations surrounding the Union’s compliance with the Convention are far from settled. Indeed, the ACCC in 2021 determined that while the amended Regulation constituted a ‘significant positive development’, certain remaining hurdles to the Union’s compliance with Articles 9(3) and (4) of the Convention would now depend predominantly on whether the relevant provisions are interpreted consistently with the objectives and obligations of the Convention (see the ACCC’s 2017 Report, paras 117-119).

Moreover, another concrete issue of the Aarhus Regulation’s review mechanism, concerning the impossibility of challenging state aid decisions, was raised in a different complaint and ACCC report, and has not been addressed by the 2021 amendment to the Regulation. In the last MoP in 2021, a new decision on the Union’s compliance on this matter was postponed, as the Union extraordinarily requested more time to “analyse the implications and assess the options available” (see paras 54-55, 57).

It thus appears that the dilemma at the core of the negotiations to which the legal advice of the Council related, seems anything but resolved. While we await the Council to provide the requested document in full in order to know for sure what the content of the advice really is, the various communications from the Council allow some theorising.

What we know for sure is what the secret document does not address, as the Council explained in the hearing in the case that the document (1) does not cover political or strategic aspects of the Commission’s proposal and the Union’s position in the Aarhus Compliance negotiations, (2) does not cover the aspect of the state aid exception, and (3) does not relate to any other future international agreement (Report for the Hearing in Case T-683/21).

Furthermore, reading between the lines of the Council’s rather vague statements in the written reply to the document request and the hearing, one can hypothesise what the document does address. It seems to concern the Union’s compliance with the Aarhus Convention’s access to justice obligations of Article 9(3) and (4) in a much more general way and in relation to the limitations posed not only by the then-to-be-amended Aarhus Regulation but also by the Union’s overarching system of legal remedies under primary law. Indeed, according to the Council, the document “contain[s] an elaborate analysis, including questions relating to primary law”, concerning “the system of internal review as established under this regulation in relation to the system of legal remedies as provided for under Article 263 [TFEU]”, and the “legal feasibility of solutions that the European Union could implement to address the alleged non-compliance with the Aarhus Convention” (Council Reply, paras 50, 52, 69 and 70). As such, even more sensitive, the Council in the hearing explained that the advice seems to cast doubt on the Union’s compliance with Article 9(3) and (4) of the Convention, potentially by interpreting the Aarhus Regulation and Union primary law in a way contrary to what the ACCC was expecting in their 2017 and 2021 reports (Report for the Hearing in Case T-683/21).

Thus, while the Court rejected the Council’s worries in relation to the sensitivity of the requested document, it does not seem unlikely that the Council within this document reflected on intricate matters of Union law and the relationship with international obligations.

A More Principled Way to Reach the Same Conclusion

Although it is thus not implausible that the document contains politically and legally charged information, this does not mean that the Council withheld access to it rightly. While the Court, in line with case law such as ClientEarth (ISDS), coupled its review of the refusal to disclose with the sensitivity or strategic nature of the legal opinions, we argue that a more principled line of argumentation would have been more desirable.

As argued previously by Peters and Ankersmit, the Court could have distinguished policy areas characterised by a zero-sum logic and areas characterised by a positive-sum logic. In the former realm, secrecy is classically viewed as a necessary evil to avoid adversaries from gaining too much insight into the Union’s internal deliberations. As alluded to by the Ombudsman, disclosure of information could indeed be dangerous if certain ‘key strategic interests’ are at play, such as military strategies or critical infrastructure. In contrast, the development of collaborative policies in fields like environmental law is typically spurred on, rather than hurt, by transparency and openness. The typical mutual benefits from cooperation in these areas even hinge on the trust parties obtain by being able to check on each other. Likewise, MoPs are generally open and transparent, whereas the Aarhus Convention also contains a pledge to uphold a high degree of transparency for environmental information (Article 4).

The Court could have interpreted the Access to Documents Regulation in light of these considerations by making this distinction between areas where the need for secrecy differs widely. As a result, the Council’s fears would not justify secrecy. It cannot be said to be in the Union’s interest to hide legal advice as a strategic move to escape critical debates on the Union’s compliance with a crucial pillar of the system of international environmental law, the success of which relies on genuine cooperation and mutual trust amongst the parties. In our view, such a principled approach is to be preferred over implicitly increasing the level of scrutiny in the review, as it makes the Union’s transparency framework more robust, in line with the objectives of the Aarhus Convention.

To conclude, we suggest that the Council’s legal advice at the core of this judgment clearly contains information that the public should be able to access, even if this information continues to have strategic significance. How controversial the content of the previously hidden legal advice actually is, should be clarified soon, when the Council follows up on the judgment and discloses the full document.

The authors would like to thank Professor Päivi Leino-Sandberg for providing us with additional context on the case, as well as the Report for the Hearing in Case T-683/21. This document is not (yet) published online.

Stephanie Bell is a Research Fellow at the Partnership on AI affiliated with the AI and Shared Prosperity Initiative. Her work focuses on how workers and companies can collaboratively design and develop AI products that create equitable growth and high quality jobs. She holds a DPhil in Politics and an MPhil in Development Studies from the University of Oxford, where her ethnographic research examined how people can combine expertise developed in their everyday lives with specialized knowledge to better advocate for their needs and well-being.

Katya Klinova is the Head of AI, Labor, and the Economy Programs at the Partnership on AI. In this role, she oversees the AI and Shared Prosperity Initiative and other workstreams which focus on the mechanisms for steering AI progress towards greater equality of opportunity and improving the working conditions along the AI supply chain. She holds an M.Sc. in Data Science from University of Reading and MPA in International Development from Harvard University, where her work examined the potential impact of AI advancement on the economic growth prospects of low- and middle-income countries.

Robert Seamans is an Associate Professor at New York University’s Stern School of Business. His research focuses on how firms use technology in their strategic interactions with each other, and also focuses on the economic consequences of AI, robotics and other advanced technologies. His research has been published in leading academic journals and been cited in numerous outlets including The Atlantic, Forbes, Harvard Business Review, The New York Times, The Wall Street Journal and others. During 2015-2016, Professor Seamans was a Senior Economist for technology and innovation on President Obama’s Council of Economic Advisers.

‍

You can watch a recording of the event here or read the transcript below:

‍

Anton Korinek 00:12

Welcome. I’m Anton Korinek. I’m a professor of economics at the University of Virginia and a research fellow at the Centre for the Governance of AI, which is organizing this event.

Today’s topic is redesigning AI for shared prosperity. We have three distinguished speakers: Katya Klinova and Stephanie Bell, who will be giving the main presentation, followed by discussion by Rob Seamans.

Let me first introduce Katya and Stephanie, and I will introduce Rob right before his discussion. Stephanie Bell is a research fellow at the Partnership on AI, affiliated with the AI and Shared Prosperity Initiative. Her work focuses on how workers and companies can collaboratively design and develop AI products that create equitable growth and high-quality jobs. Stephanie holds a DPhil in politics and an MPhil in development studies from Oxford where [she conducted] ethnographic research which examined how people can combine expertise developed in their everyday lives with specialized knowledge to better advocate for their needs and well-being.

Katya Klinova is the Head of the AI, Labour, and the Economy program at the Partnership on AI. In this role, she oversees the AI and Shared Prosperity Initiative—which has developed the report that the two will be presenting today—and other work streams, which focus on mechanisms for steering AI progress towards greater equality of opportunity and towards improving the working conditions along the AI supply chain. She holds a Master of Science in data science from the University of Reading and an MPA in international development from Harvard where her work examined the potential impact of AI on economic growth for low-income and middle-income countries.

The concern underlying today’s webinar is that AI poses a risk of automating and degrading jobs all around the world, which would create harmful effects for vulnerable workers’ livelihoods and well-being. The question is: how can we deliberately account for the impact on workers when designing and commercializing AI? [How can we ensure] AI benefits workers’ prospects while also boosting companies’ bottom lines and increasing overall productivity in the economy? With this short introduction, let me hand the mic over to Stephanie and Katya.

Katya Klinova 03:15

Anton, thank you so much for hosting us. Thank you to the Centre for the Governance of AI—it is an absolute pleasure to be here. Thanks to everyone who is joining us for this hour to talk about redesigning AI for shared prosperity.

As Anton said, the work that we’re presenting today is part of the AI and Shared Prosperity Initiative. Recently, we released the Initiative’s agenda, which is our plan for research and action. You can download this agenda at partnershiponai.org/shared-prosperity. The [agenda] is not just authored by Stephanie and I. [Rather], it is a result of a multi-stakeholder collaboration by the steering committee, which consists of distinguished thinkers from academia, from industry, from civil society, and from human rights organizations. It was also supported by the research group that included someone very dear to us who is now at the Future of Humanity Institute—Avital Balwit. We want to say thank you to Avital and to everyone who supported this work.

The goal of the AI and Shared Prosperity Initiative is to make sure that AI adoption advances an abundance of good jobs, not just for a select profile of workers, but for workers who have different skills and different demographics all around the world. To advance that goal we decided, under the guidance of the steering committee, on the method of introducing shared prosperity targets. [These shared prosperity targets] are measurable commitments by the AI industry to expand the number of good jobs in the broader economy. These targets can be adopted voluntarily, or they can be adopted with regulatory encouragement.

The agenda that we’re [discussing] today is our plan for developing these targets and thinking through their accompanying questions. The agenda is structured in two parts, which are going to be broadly mirrored by our talk today. We begin by [describing] the background against which AI advancement is happening today. [Next,] we introduce the proposal for shared prosperity targets. Then, we analyse the critical stakeholders and their interests and concerns when it comes to adopting, encouraging, or even opposing the shared prosperity targets. Our presentation [will follow this format] today. [Additionally,] we’ll briefly discuss why set targets [can] expand access to good jobs for AI, the structure of the shared prosperity targets proposal [itself], and the key stakeholders, their interests, and the constraints that they’re facing.

Let’s begin by discussing the motivation for this work. Many of you have seen this graph before. It has certainly been shown in this very seminar before by David Autor. [This graph] shows the polarization of wages, which is especially pronounced for men in the US, though women also very much experience [this polarization]. This graph [reveals] that despite the economy growing almost threefold in real terms since the sixties, not everyone has [benefitted from] that growth. You can see that people with graduate degrees have experienced a [large] wage growth, while other skill groups and educational attainment groups did not. [In fact], wages stagnated or even declined in real terms, which is quite staggering. [This pattern] definitely cannot be called “shared prosperity.” The risk and the worry are that AI will exacerbate or continue this wage polarization, [because] AI can be a skill-biased technology, [meaning it is] biased in favour of people with higher levels of educational attainment.

A [much-discussed] and very important solution is upskilling, or re-skilling, which we should [certainly] invest in. [This requires] educating people to help them navigate changing skill demands in the labour market. Nobody will ever argue against [improving] education [quantity and quality]. However, we need to be aware that if the overall demand for human labour goes down in the long term, upskilling itself will not fix the [core] issue: the scarcity of good jobs. [No matter] how much we retrain people, if there’s a declining number of good jobs in the economy, [retraining] will always be a losing battle.

The graphs you’re looking at are from a paper by Acemoglu and Restrepo, which shows that automation has picked up in the last 30 years, [which is a departure from historical trend]. [Historically,] automation existed: automation is not something that only AI introduced into our economy. [However,] automation was displacing humans from tasks [at the same rate] new tasks [were created]. [However, this has not been true over the last 30 years] and the risk is that AI will continue this trend [of rapid displacement]. The last [concern] that I want to mention is that the impacts [of automation] tend to be global. There are no mechanisms for global redistribution of [automation’s] gains, which tend to be concentrated in a few countries’ firms and accrue to just a handful of individuals.

There is a memorable anecdote that I want to share with you. You’re looking at a picture of self-order kiosks introduced in fast food restaurants. Once the investment [in these kiosks] had been made in California and the rest of the United States, the cost of deploying [this technology] everywhere around the world became so low that no matter how low the wages were in some of the low-income and middle-income countries, workers [simply] couldn’t compete with the technology. The picture you’re looking at was taken in South Africa. Even before COVID, the unemployment [rate] in [South Africa] was 29%; it was not the time to be eliminating formal sector jobs. [However,] the march of technology knows no [limits].

[Given AI’s] global impact and [our current inability to] redistribute AI’s gains globally—either through taxation or other transfers—we need to think ahead and [consider how we can ensure] AI supports a globally inclusive economic future. One [frequent] recommendation, [supported] by a growing [body] of literature, is that AI [should] complement, rather than replace, human labour. While this sounds intuitive, in practice it can be difficult to differentiate between technology that complements labour and [technology] that replaces [labour]. [Our concept of] shared prosperity targets addresses exactly [this question]: how do you differentiate between labour-complementing and labour-displacing technology?

What makes this differentiation hard? In economic terms, the definition assumes that you know exposed outcomes from the technological advancements. A technology is called labour saving if it reduces the overall labour demand in the economy and [a technology] is called labour using if it creates growth in overall labour demand in the economy. [However,] it’s very difficult to know [how a technology will impact] labour demand. Early research in [technology] can be used [in] many different applications down the road. Some of those [applications] can be labour saving and some of those can be labour complementing.

Deployment contexts very much matter. The same application [of technology] can be used for different purposes: in the workplace, something can be used to augment peoples’ productivity or to surveil them. [While the technology is applied in the same way,] how [the technology] is used depends on the values and the orientation of the employer actually introducing [the technology] into the workplace. It’s also difficult to map micro[economic] impacts of a given technology to the macro[economic] trends in the economy, because the economy is a dynamic system with [many] interacting parts. It is very difficult to predict ex ante [how a technology investment will] impact that dynamic system, just like it is difficult to predict how a business or technology investment will impact the climate, because climate is also a very complex, dynamic system. And yet, people came up the idea of tracking carbon emissions as a proxy for their impact on global warming. The shared prosperity targets are inspired by the carbon emission targets, in their pursuit of finding appropriate proxies that would, despite all of these constraints and sources of uncertainty, introduce a good enough measure to [determine] if a product or a system is likely to be labour displacing or labour complementing down the line.

I want to spend some time unpacking the connection between the micro[economic] impact of introducing technology in the given workplace and the macro[economic] consequences in the rest of the economy. Of course, there are direct consequences [of introducing technology]: there are people who [a firm] might be firing or hiring directly as [they] introduce technology in the workplace because now [the firm] needs fewer people of certain skill groups and more people of other skill groups. This is very intuitive.

We [also] want to make sure we’re not missing [technology’s] broader impacts, which can [occur] up or down the value chain. [For example,] after [a firm] introduces a new technology, [they] might require a different volume of interim inputs from [their] suppliers. [These suppliers] in turn might hire or fire workers to reduce or expand their workforce. [These are all examples of] indirect effects.

If introducing a new technology into the production process improves goods’ and services’ quality or lowers their prices, then some of the gains [of technology] are passed along to the consumers. The consumers are now, in real terms, richer: they can spend their free income on something else in the economy, which may create new jobs. We want to keep these indirect impacts in mind when we’re talking about the impact of technology.

Finally, [changes in] labour demand not [only impact] the [size of] the workforce—whether it is expanded or downsized—but also the quality of jobs and their level of compensation. Under lower demand for labour, jobs can become more precarious or [worse] paid, even if the total size of the workforce does not change. The ambiguity that I [described] between labour-displacing and labour-complementing technology gets even more complicated when people start describing their technology as “labour augmenting.” As of today, anybody can claim this title of “worker augmenting,” whether the technology grows productivity of workers and makes them more valuable to the labour market or the technology squeezes [every] last bit of productivity from [workers] using exploitative measures like movement tracking and not allowing [workers] to take an extra break. [The distinction] can be [extremely] blurry.

Shared prosperity targets would allow [genuine] producers of worker-augmenting technology to credibly differentiate themselves: if [producers] adopt ways to measure their impact on the availability of good jobs in the economy, then they would have receipts to show for calling themselves worker augmenting [rather than] worker exploiting. Shared prosperity targets are a proposal for firm-level commitments to produce labour-friendly AI while also keeping broader economic effects in mind.

There are three components that we want to track with shared prosperity targets: the impact on labour income in the form of job availability and distribution, job compensation, job quality i.e. worker well-being, and job distribution. [Job distribution describes whom] new, good jobs are available to and for whom are [good jobs] are getting scarcer. These groups can be split by skills by geographic location or by demographic factors. [Now, I’ll] turn it over to Stephanie to talk about incorporating workers’ voices into designing shared prosperity targets.

Stephanie Bell 20:21

Great, thanks so much Katya. Thank you as well to Anton and to FHI for having us here today to talk about the Shared Prosperity Agenda. We’re really excited to be in this conversation with you all.

In thinking about the next phase of applied research with workers, [we’re considering] the [most important] areas in the context of the shared prosperity targets—that Katya just mentioned—to ensure we’re taking into account workers’ priorities and needs. There’s been substantial research as to what constitutes a good job, or “decent work,” in the words of the ILO. There’s been research much more recently into the impact of artificial intelligence on different aspects of worker well-being and worker power.

Setting [shared prosperity] targets [requires] finding a sufficient amount of depth to address [workers’] real needs while also creating targets that are sufficiently clear and straightforward for companies to implement. [A framework] that covers all [worker concerns] is going to be less useful than one that is focused on workers’ high-priority needs. Our goals include [focusing] on job quality— [which we’ll] track within the shared prosperity targets—by identifying possible mechanisms, as well as required conditions, for workers themselves to participate in AI design and AI development. [Workers have] largely been left out of this process. [We aim to] identify places for workers to participate directly in this process and identify how technologies can not just not harm workers, but [rather actively improve] workers’ ability to do their jobs and also boost their workplace satisfaction. This would be a tremendous advancement in terms of the trajectory of these technologies.

Our approach [relies on] qualitative field research at different field sites around the world. Given the context of COVID, this is largely going to take place digitally, using [approaches] like diary studies, contextual observation, and semi-structured interviews to learn what workers have observed about the implementation of AI in their workplace as well as any [insights they have which will allow us to develop the] job quality component of the shared prosperity targets. Some might question the necessity of actively incorporating workers and [ensuring] we talk to a variety of people in different industries, occupations, and geographies. The rationale is that workers, regardless of their wage, formal skills, training, or credentialing, are experts in their own roles. [Workers] know the most about what it takes to get their jobs done: what the tasks are and how they experience their working conditions. By going directly to workers, we have an opportunity to understand their needs and make sure that we’re addressing their well-being and their power within workplaces, [rather than relying on] managers or company leaders as proxies and potentially misidentifying workers’ interests or missing the nuances of [workers’ experiences]. Finally, [incorporating workers is critical to our] process’ integrity. The entire point of this initiative [is to address workers’ needs]. Leaving [workers] out of the conversation would surely be malfeasance on our part, if we’re trying to make sure that we’re creating a set of targets that really does meet the needs of people whose voices are often left out of these conversations. We frequently have and [witness] conversations about the future of work that never [address] the future of workers, [and we’re trying to remedy this problem through our] work.

24:55

[Let’s transition to the] section of the agenda focused on key stakeholders’ interests and constraints. The first group that we’d like to give an overview of is workers themselves. We have two major areas of concern. [The first] is the impact of AI on worker power and worker well-being. In what ways are these technologies degrading or potentially benefiting workers? Katya mentioned [examples] like worker exploitation, aggressive surveillance, and privacy invasions. Another [area of concern is how] these systems impact workers’ ability to organize on the job, improve their working conditions, and grow their ability to participate in workplace decision-making. The [less contact] workers have with other humans during work, [the fewer opportunities workers have] to discuss their job quality and the harder it is for workers to effect change within their workplace. For example, you can introduce an effective scheduling software system that’s able to anticipate customer demand and then tailor your shift scheduling efficiently. [However,] this can radically disrupt workers lives by calling [workers] in at the last minute, forcing them to rearrange childcare, or causing them to worry that their job is in jeopardy if they aren’t able to match those needs. What we would want is for workers to be able to advocate for themselves—to have the opportunity to have a conversation with their supervisor, to make sure that their job is one that they can perform without having to worry about last-minute disruptions to their lives. However, once these decisions are no longer stemming from human-to-human conversations, you open up the opportunity for what Mary Gray called “algorithmic cruelty” to be the decision-making power within a workplace.

Stephanie Bell 27:05

The second area that we’re focused on is how worker voice [can direct] AI development and deployment. As I mentioned earlier, workers have a tremendous amount of expertise in their own tasks and [insight into how they can] improve their efficiency and productivity. For example, perhaps there are opportunities to improve safety or working conditions using technology. Depending on who’s [raising the concerns that] technology is designed to address, very different technologies are implemented. We believe that we [must] take workers seriously: they are impacted by these technologies and [their insights can be] quite generative and a real benefit to AI development companies.

Then the big question is: what are the mechanisms for change? We’ve identified three major [avenues through which] workers can create opportunities for their participation, the first of which is unions and worker organizations. This is probably an obvious [approach] to this audience, but always worth noting. However, [it is tenuous to rely on] unions and worker organizations as the [sole] avenue for change: around the world, we’re at a historically low unionization rate, which means that workers might not be in a position of power when they’re coming to these conversations.

Second, companies often take into account user and stakeholder research and testing, if not with the actual workers in a given company, than with workers who are in some way similar to them. [Workers could better participate in technological decisions if they had the] opportunity to contribute to the [research and testing] processes in a way that actually had teeth, [for example, by saying,] “This is this is a step too far in terms of its impact on me and my co-workers.” [Alternatively, workers might say,] “Hey, there’s a design feature that you hadn’t thought about, that would be really useful to build.” We see real opportunity for workers to collaborate with AI designers as well as their corporate leadership to be able to create “win-win” situations.

Finally, I think there are opportunities [for worker empowerment] within corporate governance and ownership structures. While this area is less defined in the context of artificial intelligence, historically, there are [successful models] like codetermination, cooperative ownership, shadow boards, and worker boards in which company leaders get the opportunity to have a sense of what workers think of a given product.

The second audience to discuss is businesses. One of the big questions in this work is: what would a business get out of committing [to shared prosperity targets]? As Katya pointed out, there is an opportunity for [businesses] to differentiate and gain credibility, especially when [they create] a genuinely worker-augmenting product as opposed to a worker-exploiting, worker-surveilling, or worker-replacing product.

There are also opportunities in the product development cycle. On the left side [of this slide] is a simplified graphic about the AI product development cycle. Ideally, AI-developing businesses find ways to commercialize research and develop workplace AI products which they sell to AI-deploying businesses—which are frequently a different set of businesses entirely. Those AI-deploying businesses purchase and implement AI products and then offer feedback both through their purchases and through direct communication with the firm from which organizations they buy their technology.

[However, this idealized model] isn’t how the development cycle actually functions. Instead, [a great deal of development] is driven by research breakthroughs. This isn’t necessarily a bad thing. [However,] research breakthroughs [require] use case identification and many of these use case identifications follow the format that Katya has already described: they are quite anti-worker in that they automate tasks even in instances where it doesn’t increase productivity or exploit workers for the sake of [maximizing profits]. While this is not the whole universe of use cases, one reason [for the trend toward worker exploitation] is that businesses are [not] engaging with and listening to [workers about how products impact them]. The more [businesses] build in conversations with workers—and frontline workers in particular—the more opportunity [businesses] have to identify additional use cases and different ways these technologies can be implemented. Oftentimes, [engaging with workers] can [allow businesses and developers to] expand the productivity frontier, [rather than] swapping out a human worker—one form of productivity—for a robot or an algorithm—another form of productivity.

Other business stakeholders are involved as well, [including] researchers, developers and product managers, many of whom get into this kind of work because of the intellectual challenge and the opportunity but don’t want their products to harm other people. [This challenge creates an] opportunity for conversations between workers within tech companies. [Another stakeholder] is artificial intelligence investors. Investors, and particularly large institutional investors, frequently invest in spaces where it [is profitable] to have a robust labour market. Investment in automation technology creates problems for other investments within [these investors’] portfolios. We speak about this in more detail in the agenda.

The last audience that I’ll talk about is government. We saw three major opportunities for government to participate in steering AI [in a better direction] for society and to support workers in addressing their particular challenges. Right now, [there is a great deal] of government investment in basic research and [along] the commercialization chain. [However, this investment doesn’t have] any kind of constraints on technologies whose most obvious use cases are going to be harmful for society and consolidate gains within the hands of a few.

[First,] there are opportunities to assess the way that governments deploy their research funding and procurement processes to support an AI trajectory that is broadly socially beneficial in an economic sense. Second, there’s [been a focus on] identifying opportunities to support workers who would be navigating challenges created by AI. What would be the role of government if the trajectory were different— [if we working toward mitigating AI-related risks]? [In this scenario, we may still implement] reskilling and universal basic income. The question is: how do we [avoid] creating a problem that we have to solve down the road, if [right now] we have an opportunity to [prevent] some of the most devastating impacts? Finally, low-income and middle-income countries have some very specific challenges that they need to work through. As Katya showed with her earlier example, these technologies, once created, have a very low marginal cost to implement anywhere that the company is operating, which could result in massive labour market disruption [without distributed gains] because there are no redistribution mechanisms. We think substantial work needs to take place in this space to ensure that low-income and middle-income countries don’t end up continuing to bear the brunt of the growth of the Global North. Katya, I’ll hand it over to you to cover international institutions and civil society stakeholders.

Katya Klinova 35:47

Thank you, Stephanie. I want to highlight a [section] from our chapter on international organizations and civil society. As Daron Acemoglu once said, “AI can become the mother of all inappropriate technologies for low and middle-income countries.” I showed you a photo from Twitter when I was talking about spill over effects of automation in developing countries because there are no graphs and no data that measure the magnitude or the extent of these spill over effects. We need much more research and attention to understand [these effects well]. If there is one thing we’ve learned from globalization, it’s that the expansion of trade can produce incredibly large gains and those gains can be quite concentrated. There are very real losers from [free trade], and [certain] populations can be hurt very badly. We shouldn’t repeat this story. Now, the expansion of very powerful technology opens up the possibilities for automation and the frontier of the kinds of activities that can be automated in a way that is not globally controllable. We need to be much more attentive to these trends. The role of the international organizations can be very meaningful in balancing, pacing, and understanding the cross-border impact of [technology].

So, with that, we’ll hand it back to Anton. Just to remind everyone: if you’d like to read the full agenda, it is on our website. You can also email or tweet me and Stephanie—please get in touch if you’d like to be involved with this work. Thank you very much.

Anton Korinek 37:53

Thank you so much, Katya and Stephanie, for a very clear and inspiring presentation. Let me invite all the members of the audience to contribute questions for Katya and Steph in the Q&A box. You can also upvote questions asked by others to express your interest in them.

Robert Seamans has kindly agreed to be our discussant. Rob is an associate professor at New York University’s Stern School of Business. His research focuses on how firms use technology in their strategic interactions with each other and also focuses on the economic consequences of AI, robotics, and other advanced technologies. His research has been published in leading academic journals and has been cited in numerous outlets, including the Atlantic, Forbes, HBR, The New York Times, Wall Street Journal, and others. And in 2015-2016, Rob was a senior economist for technology and innovation on President Obama’s Council of Economic Advisers. Let me hand the mic over to Rob.

Robert Seamans 39:38

Anton, thank you very much for inviting me to discuss this paper. Let me start off by saying that I’ve been following the Partnership for AI for a number of years. I think it’s an excellent organization and I like the impact that that the organization has been having. This particular initiative is very important work and very ambitious.

I’m going to start at a fairly high level with a definition. We’re using the term artificial intelligence, AI, which sounds very fancy. I think it’s useful to dumb it down. Here’s my definition of artificial intelligence: [AI] is a group of computer software techniques. At the end of day, AI is highly sophisticated software and its algorithmic techniques rely on a lot of data. I’m not a computer scientist, [AI] is outside the realm of what I can [create]. However, that doesn’t mean that I can’t talk about AI—one does not need to be an expert in a specific technology in order to think through its effects on economy and society. As a perhaps tortured analogy, I don’t know how to build a car; I certainly don’t know how to fix an engine. I would probably even have trouble changing the oil in my car. But that doesn’t prevent me from thinking deeply about how changes in cars might affect the economy and society. The same is true [for AI] and for any technology.

AI and robotic technologies are developing and commercializing rapidly. They’ll likely lead to innovation and productivity growth, which is the good news. But according to some, the effects on human labour are unclear and potentially a cause for concern. I’ll spend half an hour half my time the first part—[the good news]— and half my time on that on that very last bullet point—[the bad news].

First, [I’ll discuss] some basic facts. AI has been developing very rapidly. There have been many breakthroughs. Here is one example: this picture tracks the progress in ImageNet’s ability in image recognition. So, the y-axis shows error rate: the lower you go, the better off you are; the lower you go, the more progress there is. The x-axis shows what’s happening over time. Over time, ImageNet’s [capability in] image recognition is dramatically improving. By 2015 or 2016, ImageNet surpasses human capacity in image recognition. This is one piece of evidence that [suggests] we have rapid breakthroughs [occurring] in the lab. Moreover, these rapid breakthroughs have led to these technologies’ commercialization. The panel on the left shows venture capital funding for mostly US-based AI start-ups—[the data] comes from Crunchbase. You can see a dramatic increase [in funding] starting roughly in 2010. Why is this useful to point this out? Well, venture capitalists have very strong incentives to make sure that they’re getting these investments right. They believe that breakthroughs in the lab have commercial applications which provides some evidence [for the emergence of] commercial applications of AI.

It’s also useful to talk about what’s happening with robotics. Robots, of course, have been around longer than [AI] and to date have had more of an impact [than AI], particularly on manufacturing. There are some things happening with robots that are [useful to understanding] what might happen with AI. I’m going to talk about robots a little bit in my remarks.

The panel on the right looks at worldwide robot shipments. [There were] about 100,000 units sold annually until about 2010. Then, there is a dramatic increase, and by about 2016, three times that amount [were sold annually]. Once again, [this demonstrates] rapid commercialization of a new technology.

While it’s probably too early to say, I would bet there’s going to be a lot of productivity growth as a result of AI, as we’re already seeing with robots. Graetz and Michaels had a fantastic paper came out in The Review of Economics and Statistics in 2018. According to their study, robots added an average of 0.4 percentage points of annual GDP growth in the seventeen countries that they were studying between 1993 and 2007. This was about a tenth of GDP growth for those countries during that time period. I think that’s as good of a benchmark as we can expect. I suspect we’ll get a similar, if not greater, boost from AI, though frankly, it’s still too early to tell.

We can look at robots to get excited about what AI can do. We can also look at prior episodes of automation. All prior episodes of automation, and particularly steam engines and electrification, have led to growth. One of favourite studies is by David Autor and Anna Salomons from 2018 in the Brookings Papers on Economic Activity. They’re looking at episodes of productivity growth and the effect that had on labour. The last column [shows productivity’s] net effect—when you have productivity boosts, you see an increase in labour.

What I find interesting, though, is that there’s a fair amount of heterogeneity in the supply chain. The direct effect is negative, the final demand effect is positive, the upstream effect is positive, and the downstream effect is noisy. There are two big takeaways from this. [The first takeaway] is that the net effect is positive for labour. The second takeaway is that there’s a lot of heterogeneity. We can think about other sources of heterogeneity, for example, within a given firm across different occupations in that firm.

Let’s [consider AI]. Katya and Anton, [in their] paper “AI and Shared Prosperity,” write that future advances in AI, which automate away human labour, may have stark implications for labour markets and inequality. I agree with that statement and I [think] there are [two important components to highlight]. The first is that AI is automating away human labour. And the [second important component to consider is] inequality. [My—albeit nascent—work] may [provide] early evidence [for these points]. [My research] suggests that [at this stage] firms are using AI for augmentation rather than for replacement. However, there’s also early evidence that the augmentation is disproportionately benefiting [only] some [people: automation’s gains are not widely shared]. [This provides evidence for AI’s] heterogenous [impact] across occupations.

Over the past several years I’ve worked with Jim Bessen and a couple of other co-authors to survey AI-enabled start-ups. In the first wave of [surveys], we asked these AI start-ups, “What is the goal of this AI-enabled product that you’re creating? What is this product’s KPI when you’re trying to sell to customers?” [Start-ups] could [choose one or more from a range of possible answers]. Most [start-ups answered that their products were aimed at] making better predictions or decisions, managing and understanding data, and gaining new capabilities. [These answers suggested that technologies] augmented, rather than replaced, human labour. [On the slide,] I’ve highlighted in red the answers “automate routine tasks” and “reduce labour costs,” [as these answers suggest] replacement. However, these [replacement-indicating reasons were not among the] top reasons that these firms gave. [While there may be] some evidence that AI is being used to replace human workers, [most] technologies are being used to augment work.

[Now, let’s consider] inequality. [We’ll turn to] a paper [I wrote] with Ed Felten, a computer science professor at Princeton, and Manav Raj, a PhD student I work a lot with it. We came up with an AI Occupational Exposure Score: for each occupation in the US, we’ve come up with a way to describe how exposed that occupation has been to AI. Now let’s [segment these occupations] into three [categories]: low-income, middle-income, and high-income occupations. Let’s look at employment growth and wage growth over a ten-year period. The positive coefficient for high-income workers’ unemployment growth suggests that as the high-income group is more exposed to AI, they will see larger employment growth. The same holds true for wage growth: as these occupations are more exposed to AI, they will see faster wage growth. [In contrast], the [opposite is true] for low-income workers’ employment: [as AI growth increases, employment growth will decrease]. This suggests AI may be exacerbating inequality.

So, what’s the solution? You’ve heard about one solution from the authors of the [Shared Prosperity Agenda]: they’re developing a framework to enable ethically minded companies to create and deploy AI systems. I think that this is a very good solution. The first question that I have is: “How likely is it that firms would self-regulate to adopt such a framework?” I’ve asked a [similar] question, with Jim Bessen and co-authors, by surveying AI start-up firms. We [surveyed] firms [to learn] how many of them have adopted a set of ethical AI principles. I expected, ex ante, a low number of firms [would have ethical AI principles]. We learned that about 60% of these firms said that they did [have ethical AI principles]. Perhaps unsurprisingly, there’s a fair bit of heterogeneity across different industries.

Coming up with a framework is a useful solution, because firms actually will adopt these. Now, under what conditions might firms be most likely to adopt these principles? Based on correlational evidence from the survey that we did, [represented on the slide in] columns one and two, we found that when AI start-up firms are collaborating closely with a large tech firm—like Microsoft, Google, or Amazon—these smaller firms are much more likely to adopt these AI ethical principles. One potential [strategy this evidence suggests could be effective] is to [develop a] framework [for ethical AI] and specifically target large companies to adopt this framework first so that smaller companies will follow.

Katya and Stephanie highlighted some of the larger macroeconomic and labour market trends [and described] increasing inequality and declining union membership. Other [trends include] falling labour force participation and rising industry concentration. These high-level macroeconomic trends are important to keep in mind because [they differentiate this episode of automation] from prior episodes. The conditions under which electrification or steam [power were introduced] were very different than they are now. [These changing conditions] are useful to keep in mind.

How do we measure corporate shared prosperity targets? While it’s easy to [call] something measurable, it’s much harder to actually measure it. [This is something I hope to] push the authors on.

[Finally, let’s consider] customers [as a lever of change]. Stephanie described the three different mechanisms that could be used to push for change, but she [didn’t discuss] customers or “end-users.” We know that customers are an important stakeholder and can [cause] firms to adopt [ethical] standards. [Consider a] perhaps tortured analogy. When I’m purchasing eggs, I care about how the chickens were treated and I’m willing to pay a [small] premium [for better treatment]. You could imagine the same kind of mechanism at play here. When customers—people like you and me—are willing to pay a little bit more for a product that’s been certified as treating workers [fairly, that can motivate firms to adopt ethical standards]. [Stephanie and Katya] can add customers to the set of stakeholders that they’re thinking about.

Anton Korinek 55:21

Thank you so much, Rob, for the insightful discussion and comments. Let me give Katya and Steph an opportunity to offer their reactions.

Katya Klinova 55:31

Rob, thank you so much. These were great comments that we will use in the future. I couldn’t agree more with you. For the record, we’re not anti-automation; historically, automation has gone [well]. [However,] as you described, the societal and economic conditions [today] are different [than they were historically]. We cannot [uncritically] rely on past successes to be confident about our future success. I completely agree with your point about measurement: [now, the work is to understand how to measure these targets we’ve developed].

Stephanie Bell 56:34

Thank you so much for these comments Rob. I think they’re extremely insightful and helpful as we forge ahead with all of this. I really appreciated your point about customers and their role [in motivating ethical behaviour]. The evidence—at least that I’ve seen—seems mixed about the degree to which people willing to trade off on their surplus of productivity in order to help workers. However, there are people who are willing to buy free-range and cage-free chicken eggs and there are people who prefer Patagonia to North Face as a result of [Patagonia’s] supply chain and environmental principles. I think [there’s certainly a group of people who] are an audience for this work and we’re thinking hard about how we engage [this group].

The other part of your comments that I really appreciated was [your examination of] automation versus productivity and what that [distinction means] for workers. As I’ve [read] the [literature] on AI’s impact on workers, [I’ve learned] the degree to which many so-called augmentation technologies are a new-fangled Taylorism or Fordism. [Augmentation technologies employ a] very old managerial style [that is] technologically enabled to be much more aggressive. [For example,] injury rates in Amazon warehouses that have AI-enabled robotics are much higher [than in warehouses without robots]. [AI isn’t creating a] Terminator scenario— [rather, AIs are] colliding into people on the warehouse floor. [The danger] is about increasing work and job intensity to the point where people are badly injured from repetitive stress injuries. [As we consider] measuring [these metrics, we have to place a] fine line between augmentation and exploitation.

Anton Korinek 58:35

Thank you, Katya and Steph. The agenda that you laid out proposes [a path] to ensure that progress in AI will be broadly shared with workers. I agree that [this is very important] for the short-term and medium-term future.

At GovAI we are also interested in preparing for a long-term future in which potentially all jobs can be performed more effectively by machines [than by people]. In this future scenario, it would cost less to pay a machine than a human for the same type of work. Equilibrium wages would not even cover the basic subsistence [needs of humans]. Steph has already hinted at this possibility during her presentation. One of the members of the audience, Vemir Michael, phrased it this way: “In the long term, can shared prosperity be [managed] within the company environment, as workers [self-advocate]? [Or, will there need to be] deeper [governance, in the form of a] government structure? [Or, must there] be a societal shift?”

So, let me ask you: how do you think about this potential long-term future in which all jobs may be automated? How does it square with the agenda that you are advancing? How do you [negotiate] the tension between using work as the vehicle to deliver shared prosperity and the fear—that some in the field have—that there may be no work in the future?

Katya Klinova 1:00:12

[It’s essential to] make sure there is work in the interim [period], [given that during this interim,] work will still be the main vehicle for distributing prosperity around the world and the main source of income for the majority of the global population. [This work availability] is actually a precondition for long-term AI progress. If the decline in labour demand and the elimination of good jobs happens too quickly, there will be so much social discontent that it could preclude technological progress from happening. We need to pace the development of technology [with] the development of social institutions that enable redistribution. Eventually, we may need to decouple people’s prosperity, dignity, and well-being from their employment. Right now, [however,] we are in a society in which [work and well-being] are tightly coupled. We cannot pretend we’ve already figured out how decoupling can be done painlessly and globally. Even the boldest proposals [don’t propose] large [redistributions: they are in the range of] $10,000 to $13,000 per year. I don’t want to say anything at all against UBI—I think social safety nets are incredibly important and very much needed. We just need to be realistic about [what is] sufficient in the interim. [This] interim period is a precondition for success in a future in which nobody needs to work to survive.

Stephanie Bell 1:02:35

I fully agree [with Katya]. The devil really is in the details when [considering] the feasibility of different approaches to the trajectory of [AI] and its impact on people’s livelihoods. I think, based on my previous work in democratic theory and trust building across different social groups, and considering the current political environment, that we are more likely to convince an important subset of capitalist companies to ever so slightly decrease their bottom line than to put in place large-scale redistribution. And that’s just [considering redistribution] in a given nation state, let alone across nation states. [Redistribution requires] functioning democratic governments. Unfortunately, right now, we’re seeing many governments—which would consider themselves to be democratic—backtracking. Given this, what does a transition period look like? What is the best way to work toward a jobless future? How do we ensure that [our path to this future is] humane for everybody involved? Unfortunately, I’m not optimistic that near-term redistribution is the solution.

Anton Korinek 1:04:08

Thank you, Steph and Katya. Let me read the next question from Markus: “Could you say more about the role of policy in shifting AI technology in a labour-augmenting direction?”

I’ll add my own follow-up question for Rob specifically. The agenda for redesigning AI for shared prosperity has focused on making AI more worker friendly, [especially] in the private sector. I think we all agree this is an important starting point. Rob, you also have considerable experience in public policy settings. I wanted to ask you [how you would approach creating] public policies to support shared prosperity. What would be your advice on how to best go about making public policy work useful and appealing to policymakers?

Robert Seamans 1:05:27

I agree with much, though maybe not all, of what Stephanie and Katya have said. [While they didn’t cover this,] I worry about [a specific segment of] AI policy [focused on] addressing inequality. The first reason [I’m concerned relates to what] Katya said earlier: it’s very difficult to know ex-ante if technology will be labour displacing or labour augmenting. We can only [make this distinction] ex-post. I don’t think it makes any sense to try to create a policy focused on taxing certain technologies because we think [these technologies] are going to be labour replacing—I worry about the distortions that [tax] would impose. The second reason [I worry about this policy] is that the larger trends we’ve touched on, like declining union membership, increasing inequality, declining labour force participation, and increasing industry concentration are first-order concerns. We want to be addressing these before coming up with policy that’s specific to AI. That being said, I like The Partnership on AI’s approach because it gets firms to engage in self-regulation, which I think [is a better approach than] government-imposed [regulation]. There is a role for government to play, as a convener of different firms, stakeholders, workers, and customers to arrive at a set of principles that firms might be more willing to adopt rather than less willing to adopt.

Anton Korinek 1:08:00

Katya and Steph, would you like to add your thoughts on policy?

Katya Klinova 1:08:07

It’s, of course, scary if the government begins taxing something that is likely to be labour displacing ex ante. However, the government does fund a great deal of technology R&D which can [affect development] in the private sector. If the government, [in addition to implementing] other policies, starts thinking [ahead] and lays the groundwork for labour-complementing technologies, it [could steer] AI away from becoming excessively automating. Interest rate policy and immigration policy can influence the supply of labour and [impact how likely firms are to] invest in automation. We want the government to be aware of AI’s capacity to [increase] inequality by benefiting high-skilled workers and to think through what [it] can do to create conditions in which the private sector [makes] investments in labour-complementing technology.

Stephanie Bell 1:09:58

I wholeheartedly agree with Rob’s point: a tax that targets AI specifically is likely to cause quite a few distortionary effects, as many of the problems that emerge from AI also emerge from other technologies. To the extent that [we focus] on dealing with the impacts of technological change on workers well-being, worker power, and worker livelihood, a more encompassing set of regulations or approaches would be [warranted].

[Currently, a great deal of] AI research is targeted at human parity metrics: how well can this technology replace a human doing the same task? That’s a very different kind of metric than one focused on what we can achieve when technology is working together with a person on a [given] task. Using something other than a human parity metric to measure success could help the government [steer] AI research to be more augmenting and potentially less exploitative.

A second thought [concerns Katya’s comment on the] taxation scheme. Capital and labour are treated differently in tax schemes around the world. If [government] makes it much cheaper—at least in terms of accounting gains—for a company to purchase software or a robot to do a given task, then [the government is] disadvantaging a worker who could be doing those tasks instead. If aggressive depreciation gives [firms] tax advantages— [as happens in] the United States—on any piece of equipment or a capital investment, but all labour-related [expenses] incur a payroll tax, then [the government creates] two different incentives to replace labour.

Finally, I think many of these problems [stem from] labour law. Places like the United States in particular would benefit from having more stringent laws to protect workers from workplace injuries and exploitation and to safeguard workers’ livelihoods, wages, and hours. Putting [these protections] in place, either through additional rules or heavier fines for breaking these laws, would steer companies away from using more exploitative technologies.

Robert Seamans 1:12:44

I completely agree with these comments, Stephanie and Katya. In particular, I think the point about the different ways that capital and labour are taxed is very important.

[Let’s consider a scenario] that I would like to get your reaction to. Let’s say the Partnership on AI successfully comes up with the framework that you’re in the process of developing. Might [implementing] a policy that the government can only purchase from firms that have adopted this framework [create an incentive] for firms to adopt [these principles]?

Katya Klinova 1:13:27

I would love that. Right now, the government procures a lot of technology. If the government recognized long-term decline in labour demand as [important], how would they [evaluate] criteria for which technology to buy from whom? Would they just decide based on marketing [information] on the website that says, “this technology augments workers”? Or would [the government] ask for disclosures? [If so,] what kind of disclosures would they be looking for? What would be measured? We think this framework could be useful—even if not mandated as law—as a [means] to inform decision makers who handle government procurement of technology.

Stephanie Bell 1:14:58

A question from Michelle: “What do each of you believe will be the biggest challenge in redesigning AI for shared prosperity? Will [the challenge] be [from] a specific industry? The engagement from a specific stakeholder group? Or [will it be] something else? Is there a [consensus] on the largest challenges among your research team?”

I should also add that Michelle is asking how she can best continue the conversation, so perhaps tell us again how to find out more about the Shared Prosperity Agenda on the PAI’s website.

Katya Klinova 1:15:38

Michelle, thank you for the question. For you and for everyone that would like to stay in touch with us, there is a form to leave your email and sign up for updates on these discussions and conversations. All of this is on partnershiponai.org/shared-prosperity.

We will see right now if there is agreement on the biggest challenge. The immediate challenge for us is to figure out a reliable, robust way to measure [our goals] that would be intellectually honest and substantive but at the same time be intuitive and simple enough to explain so that a lot of people could get behind it. [Referring back to the] example of eggs in the store—there is a one simple label you’re looking for, the “free range” label. [To consider another example,] carbon emission targets are a very complex system that was proxied so that it was easy to understand and get behind, though it [still] took two decades to build momentum behind corporate carbon emission targets. And governments [still have difficulty] deciding which investments are environmentally sustainable or not.

We don’t have decades for this work because AI progress, and its impacts on labour, are happening [so quickly]. How do we quickly [develop] a metric that is substantive but intuitive? This is the question that keeps me up at night.

I should add that Anton is this Senior Advisor to the initiative and on our steering committee. I couldn’t have done [this work] without his support.

Stephanie Bell 1:17:58

I agree with Katya: getting this set of metrics right is going to be our biggest challenge because developing intellectually honest and rigorous metrics is challenging. [Another challenge is] finding a way to translate the rigor into something that’s easily implementable, especially for companies who don’t have a team of in-house macroeconomists and microeconomists. [We have to distil our metrics] so that companies can [understand these metrics], support [our] cause, and [feel capable implementing these targets]. Our work over the next couple of years will be to figure out how to make [metrics] that are coherent and actionable.

Anton Korinek 1:18:51

Thank you, Katya and Steph. Now let me ask you, perhaps as a concluding question, if one of the members in our audience is an AI developer, what tangible next steps would you recommend that they take to advance shared prosperity through their work?

Katya Klinova 1:19:24

If you read the companion paper “AI and Shared Prosperity” on our website, we lay out steps that could be [useful for] AI developers. If you would give us feedback on [if these steps] are working for you and whether they’re helpful or not, that would be [very] appreciated. [Another way to help would be to] spread the word: AI developers and innovators at-large have a responsibility to think about their economic impacts on labour and on the distribution of good jobs. I do not think that this notion of [developer and innovator reasonability] is broadly accepted. [You can advance the cause by helping this] become more of a norm and an expectation.

Stephanie Bell 1:20:15

[I echo] everything that Katya just said. [It’s important to] push for economic impact as a fundamental part of AI ethics. AI has advanced impressively along a number of different tracks. For whatever reason, the economic impact of these technologies is not a part of that conversation. The more we’re able to bring awareness to how [AI’s economic impacts affect] people’s livelihoods, the better the opportunity we have for success in [steering AI in a] in a positive [direction].

Anton Korinek 1:20:55

Let me say thank you to Katya, Steph and Rob, not only for your presentations and the discussion, but also for the thoughtful conversation that we have had thereafter.

Thank you and we hope to see you at our next webinar.

‍

In Brief‍

Today, the Centre for the Governance of AI (GovAI) has relaunched as a nonprofit organisation.

Our mission remains the same: We are building a global research community, dedicated to helping humanity navigate the transition to a world with advanced AI. Our core research activities will also remain largely the same. However, owing to the greater flexibility afforded by our new nonprofit structure, we will also be expanding our field-building activities. 

You can see the other pages on our new website more detailed descriptions of our mission and history, research, team, events, governance structure and approach to conflicts of interest, and opportunities for involvement. We are currently hiring for Chief of Staff and Research Fellow roles and accepting applications to our Summer Fellowship program.‍

‍Our Activities‍

We will maintain our core activities of producing, supervising, and coordinating research; running a twice-annual fellowship program for junior AI governance researchers; hosting both internal research seminars and a public seminar series; advising decision-makers; and offering career advice, connections, and informal mentoring to promising people entering the field. In the coming year, we will also host our inaugural AI governance conference, begin awarding student research prizes, and explore an expansion of our policy-advising work.

Over approximately the next two years, we are planning to experiment and learn more about how we can provide the most value in our current form. We may adjust or halt current or planned activities if we are not sufficiently convinced of their impact. We may also trial additional activities such as grantmaking and scholarship programs or a program reminiscent of the Forethought Fellows program.‍

‍

Our Structure and Team‍

We are based in Oxford, in the same building as the Future of Humanity Institute, Global Priorities Institute, and Centre for Effective Altruism, with a global network of collaborators and affiliates spread across several institutions. See here for a description of our governance structure.

Our Acting Director, Ben Garfinkel, leads the organization. Ben is a Research Fellow at the Future of Humanity Institute and the head of its AI Governance Team. He has been involved with GovAI and its predecessor organizations for five years: he was a founding member of the Yale University research group that evolved into GovAI.

Our President, Allan Dafoe, advises the organization and collaborates on research projects. Allan is also the founder and previous Director of GovAI. He currently heads DeepMind’s Long-Term AI Strategy and Governance Team.

The rest of the core team consists of:

• Markus Anderljung, as Head of Policy and Research Fellow.
• Joslyn Barnhart, as Applied Research Lead.
• Alexis Carlier, as Head of Strategy.
• Noemi Dreksler, as Survey Researcher.
• Anton Korinek, as Economics of AI Lead.
• Anne le Roux, as Operations Manager.
• Robert Trager, as Strategic Modelling Team Lead.
• Eoghan Stafford, as Strategic Modelling Researcher.

Toby Shevlane will also soon join the team as a Research Fellow.

Our Advisory Board consists of Ajeya Cotra, Allan Dafoe, Helen Toner, Tasha McCauley, and Toby Ord.

Our broader affiliate community consists of Miles Brundage, Tantum Collins, Diane Cooke, Jeffrey Ding, Sophie-Charlotte Fischer, Carrick Flynn, Ulrike Franke, Hiski Haukkala, William Isaac, Jade Leung,  Cullen O’Keefe, Jonas Schuett, Stefan Torges, Andrew Trask, Brian Tse, Waqar Zaidi, Baobao Zhang, and Remco Zwetsloot. 
‍

Opportunities‍

We are also currently accepting applications for two roles: Chief of Staff and Research Fellow.

The Chief of Staff would serve as the “central node” within GovAI, reporting only to the Director. A range of crucial responsibilities, most of which currently sit with the Director, would be delegated to the Chief of Staff. We believe that the right candidate could significantly increase the organization’s long-run impact and ability to expand its activities.

Research Fellows will be expected to produce research that bears on important problems and open questions in AI governance. We are interested in candidates from a range of disciplines, who have a demonstrated ability to produce excellent research and care deeply about the long-run impacts of AI. The role would offer significant research freedom, access to a broad network of experts, and opportunities for collaboration.

We are also currently accepting applications for our Summer Fellowship Program. This program provides an opportunity for early-career individuals to spend three months working on an AI governance research project, learning about the field, and exploring different ways to contribute.
‍

Acknowledgements‍

We are grateful to Open Philanthropy, the Centre for Emerging Risk Research, and Effective Altruism Funds for financial support; to the Centre for Effective Altruism for providing us with temporary fiscal sponsorship; to the Future of Humanity Institute and the University of Oxford for having provided an excellent initial home; and to the countless individuals who have supported or been part of GovAI over the years.