A Call For Action on AI Raises More Questions
Good intentions are not enough to stop calls for an AI policy response from going awry.
I was rather excited when I saw this morning, Melbourne time, the news that a group of leading economists and AI researchers had released a call for action. “Oh good,” I thought, “I love acts.” And sixteen Nobel laureates? That’s basically all of them.
At the risk of spoiling it for you, the text of the statement is rather less thrilling.
1. AI may become radically more powerful over the next 10 years.
2. This could drive an unprecedented transformation of our economy, larger than the Industrial Revolution, but unfolding over a vastly shorter time frame. It could bring risks, including large-scale job displacement, as well as opportunities such as major gains in living standards.
3. Economists, policymakers and technology leaders must act now to understand the economics of transformative AI and to build the incentives, guardrails, and institutions needed to steer AI in a direction that complements humans and benefits society.
Unobjectionable, really. Even the most accelerationist of AI researchers has to agree that the rapid progress in AI we’re seeing could drive an unprecedented transformation of the economy. It could bring risks. I took my son to a birthday party at a trampoline park the other day—many things could bring risks.
I also find myself nodding along with the call to action. Yes, we should act now to understand the economics of AI. Yes, we should build things needed to steer AI in a direction that benefits society. I’m not sure we needed more than 200 of our brightest minds to come up with that. Thinking Machines shared a similar vision several days before, all the more striking because they have already released several products that support an open and competitive market for intelligence. I think maybe we could have saved the domain—wemustact.ai—for when we have enumerated what some of those actions are. But those are minor quibbles from a man without a laureate. I don’t see how you can hate from outside the club.
No, my concern is instead about how it’s going to be received. If the audience were simply economists and technology leaders, the signatories did not need to sign a public note. If senior academics at Stanford and MIT think that more research is needed to understand the impact of AI, they are in an excellent position to fund and support that research. The same goes for senior leaders at Google and OpenAI and Anthropic. I’m sure they are doing so already.
Instead I think the public letter should be read as a political act. Not in a partisan sense, but rather as a message from the world of academia and technology directed to the world of policy makers, advisors and politicians themselves. I suspect that the nuance in the statement, short as it is, will not cross the divide. The careful caveats will be read away, as will the call to “understand”, and the message received will primarily be one of urgency. We have to do something!
In the absence of a clear theory of the case, fanning the political urge to do something! can be counterproductive. New technology, even technology that is far simpler and narrower in scope than Artificial Intelligence, often presents the policy class with a particular challenge. Social media, cloud applications and online marketing were technically well understood in 2016, for example, when the EU’s General Data Protection Regulation was adopted. But the regulation, intended to protect consumers, nonetheless led to a rise in market concentration. And a lot of annoying pop-ups.
There is no shortage of politicians, in the US and elsewhere, keen to Do Something. Senator Sanders and Representative AOC proposed a data center moratorium in March. Representative Greg Casar has proposed an entire platform of anti-AI populism, including a tax on tokens. The Trump Administration itself has oscillated, issuing an Executive Order earlier in June before placing an export ban on Fable 5 that many people, including some of the signatories to this call for action, criticized as rash or arbitrary.
From a regulatory perspective, the urge to Do Something right now seems particularly fraught. We are in a moment of great uncertainty and endless possibility. Six weeks ago, the world was going to belong to Anthropic in perpetuity. Two weeks later: open source models will win. The value will accrue to the ontology layer. Alex Karp says CEOs don’t trust the labs. Perhaps this week it will be OpenAI’s compute lead and usage resets that prompt calls for regulation or action.
A regulatory regime—or even a set of institutions—conceived in any one of those moments may well be unfit for the moment to follow.
I think many of the signatories would agree. They might even point to the language of the statement itself: we must act now to understand. The list includes many people across academia and industry who are leading work to do exactly this, with the goal of explaining in deep, granular detail the impact of AI on the economy and society. In the next few months and years we will likely see more of this work, and it will allow us to talk with new authority about effects—on market power, on labor markets, on welfare—that are now mostly speculative. These insights could form the basis of a more informed public and political conversation about AI.
You could even put them in an open letter, and ask nobel laureates, leading economists and AI researchers to sign their name to it. Proof that, even in a field moving as fast as AI, it’s possible to get ahead of yourself.
The GDPR parallel is the one that should worry the signatories most, because it proves the exact failure mode they're inviting. A regulation designed to protect consumers ended up entrenching the incumbents who could afford compliance and killing the smaller competitors who couldn't. The AI version of that outcome is worse because the compliance cost of AI regulation would be compute, and compute is already the scarcest resource in the industry. Any regime that adds friction to compute access hands the advantage directly to the three companies who already have it.
The letter asks for institutions without naming what those institutions would do. That's not caution. That's how you get a token tax and a data centre moratorium, because a politician who's been told to Do Something will do the first Something available.
https://scenarica.substack.com/p/the-shareholders-brake