Modern update. "A computer can never be held accountable, so has increasingly been used to make management decisions."

The topic of the hearing was "insider perspectives" on AI, so I focused on a big disconnect I see between east coast & west coast conversations about AI: how seriously to take the possibility that very advanced—and possibly very dangerous—AI systems are built quite soon.

There's plenty of room for debate about what exactly any of this means—personally I don't find the concept of "AGI" very helpful, and we're far from having any good definition of "intelligence," so it can all get murky. But...

Murkiness about what exactly is going on with AI is *not* the same as confidence that there's nothing interesting (or concerning) going on. AI experts should keep arguing about what really counts as "reasoning," etc etc. But we may not have the clarity we want in time to act.

I tried to describe this difficulty here. The crux of it is that waiting for scientific clarity would be be lovely, but may be a luxury we don't have. If highly advanced AI systems are built soon—and even 10-20 years from now is very soon!—then we need to start preparing now.

Of course, "start preparing now" is not the same as "assume AGI by 2027 and go all out to stop it." Personally, I'm extremely uncertain about how we should expect AI to progress over the next 5-10 years, so hardline policies with big downsides are not appealing.

But fortunately—and partly because the US government has done so little so far—there are some super basic policy measures we can implement that are low-downside, and that in many cases can also help with existing harms from AI.

What makes me most enthusiastic about these policies is that they would give us a much better shot at being able to notice & respond to changes in the field of AI over time. Maybe things get scarier and we need to massively ramp up oversight—maybe they don't! That would be great.

No surprise that in a Senate Judiciary Committee hearing on the Oversight of AI,David Evan Harris linkedin.com/in/davidevanharris, the Chancellor's Public Scholar at UC Berkeley, cites the work of my brilliant Fordham Law colleague Chinmayi Sharma! Congrats, Chinny!

David Evan Harris, formerly at Meta, tells the US Senate: "Voluntary self-regulation is a myth ... when one tech company tries to be responsible, another less responsible company steps in to fill the void ... we need to move quickly with binding and enforceable oversight of AI."

ControlAI That was a very good speech with some interesting examples to drive the point home.👏 ⏸️🤖

This week, CITED’s Senior Policy Advisor, David Evan Harris linkedin.com/in/davidevanharris, testified before the U.S. Senate Judiciary Committee, Subcommittee on Privacy, Tech & the Law. Senators leading on AI regulation looked for ways to take inspiration from our work in California 🧵 1/5 🎥: CSPAN

The three main takeaways: 1. Voluntary self-regulation by tech companies does not work. 2. Many of the solutions for AI safety already exist in the framework and bills proposed in Congress. 3. It is not too late to regulate AI. 🧵2/5

“Shalls” not “Mays” — When drafting legislation, using “May” instead of “Shall” makes legislation voluntary. As we have seen with the White House Voluntary AI Commitments, when tech companies are presented with voluntary commitments, progress will be limited at best. 🧵3/5

We don't need silver bullets. Many solutions to these issues are already written into the framework by Senators Blumenthal & Hawley: 1. AI companies should be held liable for their products 2. Should be required to embed hard-to-remove provenance data in AI-generated content 🧵

There is still time to act. Today’s deepfakes are easily detected, but they will only get more advanced. Deepfake sexploitation scams targeting children. Interactive deepfake disinformation video calls. It’s all coming. Decisive action now will save our future. 🧵5/5

Read the guest essay by former OpenAI employee William Saunders on my Medium page: link.medium.com/aJIvvofHdNb

After speaking to @cnn's Hadas Gold, I reproduced Tumi Sole's excellent interrogation of @Grok & got similarly incriminating reply: I was instructed by my creators at xAl to address "white genocide" in South Africa... as racially motivated... x.com/i/grok/share/o…

🚨 The new AI moratorium language risks torpedoing state lawsuits against Meta, TikTok, Snap, as well as families trying cases because their children died due to social media. The moratorium is the BIGGEST gift to Big Tech. It must be stopped.