Some cheerful notes on the US Senate Hearing on Oversight of AI

Earlier this week, a hearing was held at the US Senate on the topic Oversight of AI: Insiders' Perspectives. Here is the full 2h 13 min video recording of the event, and here is a transcript. I strongly recommend seeing or reading the whole thing.

As regards the subject-matter content of the hearing, large parts of it can only be described as deeply troubling, provided one cares about the human civilization and the human race not being destroyed in the sort of AI catastrophe that may well become the endpoint of the ongoing and reckless race between leading tech companies towards creating superintelligent AI.¹ Nevertheless the meeting cheered me up a bit, because I think it is of tremendous importance that the topics discussed reach the ears both of powerful politicians and of the general public. In addition, the following two observations had a really heartening effect on me.

1. My admiration for Senator Richard Blumenthal is on a steady increase. When he chaired an earlier session, in May 2023, on a similar topic, he was apparently unprepared to seriously take in the idea of AI-caused human extinction, and misunderstood it as being a labor market issue. Here is what he then said to OpenAI's CEO Sam Altman:

You have said - and I'm gonna quote - development of superhuman machine intelligence is probably the greatest threat to the continued existence of humanity. End quote. You may have had in mind the effect on jobs.

This is understandable. Extinction of humanity is such a far-out concept that it can be hard to take in if you are not used to it. But over the next few hours and months, Blumenthal did take it in, and in this week's hearing he showed excellent undertanding of the issues at stake. He really does take the issues seriously, and seems to be a force for good concerning the need to involve government in mitigating AI risk. Also, not every 78-year old top politician in the United States shows such a steep learning curve.

2. Of the four witnesses, two of them - Helen Toner and William Saunders - are situated mainly on what I would call the AI safety side of AI discourse, while the two others - Margaret Mitchell and David Evan Harris - are more towards AI ethics. These are two adjacent areas without any razor-sharp boundary between them, but here is how I contrast them in my recent paper On the troubled relation between AI ethics and AI safety:

The difference between the fields is mostly one of emphasis. Work in AI safety focuses mainly on what happens once AI attains capabilities sufficiently broad and powerful to rival humanity in terms of who is in control. It also addresses how to avoid a situation where such an AI with goals and incentives misaligned with core human values goes on to take over the world and possibly exterminate us. [...] In contrast, work in AI ethics tends to focus on more down-to-Earth risks and concerns emanating from present-day AI technology. These include, e.g., AI bias and its impact on social justice, misinformation based on deepfakes and related threats to democracy, intellectual property issues, privacy concerns, and the energy consumption and carbon footprint from the training and use of AI systems.

As discussed at some length in my paper, a tension between representatives in these fields has in recent years been salient, often with accusations that people on the other side are wasting time and resources on the wrong problems. This is extremely unproductive, but all the more wonderful was to see how the witnesses at this Senate hearing showed no such tendencies whatsoever, but instead were eager to emphasize agreements, such as around the need to regulate AI, the dangers involved in naively hoping that the tech companies will self-regulate, and the importance of whistleblower protection. I would like to think that this is a sign that the two camps are beginning to get along better and to unite in the struggle against the true enemy: the tech company executives who are letting (to quote the words OpenAI's former head of safety Jan Leike used as he left in disgust) "safety culture and processes [take] a backseat to shiny products".

*

A final word of caution: Do not take my cheerful observations above as an excuse to say "phew, I guess we're all right then". We're not. The Senate hearing this week was a step in the right direction, but there's a long, difficult and uncertain road ahead towards getting the necessary governmental grip on AI risk - in the United States and internationally.

Footnotes

1) Here are two passages from statements by the witnesses at the hearing. For me personally, it's nothing new, but it is very good to hear them artucilated clearly in this setting. First, former² OpenAI board member Helen Toner:

This term AGI isn't well-defined, but it's generally used to mean AI systems that are roughly as smart or capable as a human. In public and policy conversations talk of human level AI is often treated as either science fiction or marketing, but many top AI companies, including OpenAI, Google, Anthropic, are building AGI as an entirely serious goal and a goal that many people inside those companies think they might reach in 10 or 20 years, and some believe could be as close as one to three years away. More to the point, many of these same people believe that if they succeed in building computers that are as smart as humans or perhaps far smarter than humans, that technology will be at a minimum extraordinarily disruptive and at a maximum could lead to literal human extinction. The companies in question often say that it's too early for any regulation because the science of how AI works and how to make it safe is too nascent.

I'd like to restate that in different words. They're saying we don't have good science of how these systems work or how to tell when they'll be smarter than us or don't have good science for how to make sure they won't cause massive harm. But don't worry, the main factors driving our decisions are profit incentives and unrelenting market pressure to move faster than our competitors. So we promise we're being extra, extra safe.

Whatever these companies say about it being too early for any regulation, the reality is that billions of dollars are being poured into building and deploying increasingly advanced AI systems, and these systems are affecting hundreds of millions of people's lives even in the absence of scientific consensus about how they work or what will be built next.

Second, former OpenAI safety researcher William Saunders:

When I thought about this [i.e., timelines to AGI], there was at least a 10% chance of something that could be catastrophically dangerous within about three years. And I think a lot of people inside of OpenAI also would talk about similar things. And then I think without knowing the exact details, it's probably going to be longer. I think that I did not feel comfortable continuing to work for an organization that wasn't going to take that seriously and do as much work as possible to deal with that possibility. And I think we should figure out regulation to prepare for that because I think, again, if it's not three years, it's going to be the five years or ten years the stuff is coming down the road, and we need to have some guardrails in place.

2) Toner was pushed off the board as a consequence of Sam Altman's Machiavellean manueverings during the tumultuous days at OpenAI in November last year.