Elon Musk on AI safety: We will probably need an AI regulatory agency.
“I think it’s clear that there’s a strong consensus, an overwhelming consensus that there should be some AI regulation, that it would be in the best interests of the people to do so. And I think we’ll probably see something happen. I don’t know on what timeframe or exactly how it will manifest itself.
We’ve created regulatory agencies before. And actually just recently, just before leaving, I made the point that, you know, while our regulatory agencies are not perfect, and I deal with regulators on a very frequent basis with automotive, communications, Starlink, and then FAA with rockets. So I’ve had a tremendous amount of interaction with regulators for a couple decades at least.
I think some sort of AI regulatory agency that stands on its own, similar to the FAA or FCC, is likely at some point.
The reason that I’ve been such an advocate for AI safety in advance of anything terrible happening is that I think the consequences of AI going wrong are severe. So we have to be proactive rather than reactive.”
Elon Musk speaking to reporters at the U.S. Senate AI Insight Forum, September 13, 2023.
Elon spent billions on "Free Speech" (buying Twitter), and ~0bn on making sure AI doesn't kill everyone.
In fact the number is negative considering it was in part his idea to create OpenAI and later his own xAI adding more fuel to the AI race.
I signed this. Anthropic is simultaneously the company that most loudly presents itself as the safety lab, and the company that is most directly shooting for superintelligence before they or our broader civilization are remotely ready for it. They should stop.
Under what circumstances is the racing stance moral? I.e. “This tech is outrageously dangerous; humanity is nowhere near prepared for it; but we must reluctantly participate in (indeed, accelerate) its creation before other bad (worse?) actors get there first.”
At minimum, I think the moral obligation is *desperate effort* to change the terms of the situation, to undo the race dynamics, to provision stability. I expect participants to *hate* the situation they’re in, *hate* that what they’re doing seems like the best way out, and to be constantly searching for some way to stop. I want to *feel* that desperation from anyone who (by their own account) is compounding the risk to so many others.
I notice that I do not feel this sense of desperation from Anthropic the company, its leadership, or from (sorry) my many friends who work there.
Practically speaking, I can’t stop thinking about @KatjaGrace’s observation that people say “inventing AGI seems impossible!”, and the lab’s attitude is “just watch”; meanwhile, people say “what about China??” or “how to ensure safety is provisioned (by all AI labs!) faster than capabilities??”, and the attitude is “oof, seems hard… we’ll try our best?”. I want to see the desperate, frustrated effort of the damned.
4) Anthropic is known for being a monoculture, but there are fractures emerging.
Some employees have pushed for Anthropic to unilaterally pause (e.g. limiting compute to frontier training) even if other companies don't.
Dario shut down these suggestions down, telling employees to stop listening to outsiders & reminding them that its competitors won't stop even if they do.
Donald Trump has FAILED America on AI safety.
California isn’t waiting for catastrophe. We’re moving forward with mandatory safety standards and real accountability.