Skip to main content

Daniel Cocotello

American AI researcher and forecaster; former OpenAI governance researcher (2022–2024) and executive director of the AI Futures Project, known for the "AI 2027" forecasts and being named one of TIME's 100 Most Influential People in AI (2024).

Sep 19, 2026

speaker_2SOUNDBITE_SPEAKER
15:16
What do you believe is the first step that lawmakers should try to take here?
15:20
Recently, uh, more than a thousand employees at frontier AI companies signed an open letter called Pacing the Frontier, basically begging the government to slow down their employers and have the pace of AI progress be, you know, somewhat slower than it currently is.
15:37
Uh, people disagree about how much slower.
15:39
That is the main thing that I think that needs to happen as soon as possible.
15:43
I think that, uh, we have a lot of AI progress already.
15:47
Maybe it can continue, but we don't want it to go much faster than it's currently going, and unfortunately, it seems to be accelerating due to the fact that they are beginning to automate the AI research process itself.
15:59
They're starting to train AIs to write the code, to edit the code, to conduct research experiments on their own, to train future AIs, et cetera.

11 MINS LATER

BrentHOST
27:17
So you can't have your phone at the dinner table.
37:24
What do you believe is the first step that lawmakers should try to take here?
37:31
Um, recently, uh, more than 1,000 employees at frontier AI companies signed an open letter called Pacing the Frontier, basically begging the government to slow down their employers and have the pace of AI progress be, you know, somewhat slower than it currently is.
37:49
Uh, people disagree about how much slower.
37:51
That is the main thing that I think that needs to happen as soon as possible.
37:55
I think that, uh, we have a lot of AI progress already.
37:59
Maybe it can continue, but we don't want it to go much faster than it's currently going, and unfortunately, it seems to be accelerating due to the fact that they are beginning to automate the AI research process itself.
38:11
They're starting to train AIs to write the code, to edit the code, to conduct research experiments on their own, to train future AIs, et cetera.
41:08
Mm-hmm.
29:20
But then something changed.
29:24
But that, that never materialized.
29:25
They never ended up actually going anywhere close to 20%, and the super alignment team was disbanded the next year.
29:30
When I first joined, there was this idea that, like, of course we would slow down, uh, but, like, what about those other companies, you know? And as you can see right now, that's, like, totally not the narrative.
29:41
They're-- they definitely not slowed down.
29:43
They're explicitly saying that, like, actually AI's really good and what's the problem? Like, the best way to make it safe is to just keep deploying as fast as possible, you know.
29:52
So, so I gradually lost faith that the company would behave at all responsibly, uh, during the time of AGI.
32:50
Is it reasonable to assume they're doing that already?
62:25
So, Daniel, one of the things that we were flagged about is you've also warned about kind of the r- you know, a lot of the discourse right now is about white-collar jobs, but what about some of these more blue-collar jobs, potential robots working as plumbers and others that you've described in the past? Curious for your take on that.
62:42
Yeah.
62:42
So it depends on how far away superintelligence is.
62:46
So, or how far away, uh, the full automation of AI research is.
62:50
Um, in the near term, like in the next couple years, probably in the 2020s, um, I would say that if we get the full automation of AI research in the 2020s in the way that the companies seem to think, um, then the sort of cognitive intellectual capabilities of AIs will outstrip the physical capabilities for some period.
63:11
And this is what's described in AI 2027.
63:13
So, um, in AI 2027, again, in our scenario, uh, they succeed in automating AI research in 2027, and this results in better and better AIs that become super intelligent by 2028.

8 MINS LATER

71:06
Can I ask you sort of philosophically, 'cause I think this gets to what would be the quote unquote like "motive" of a superintelligence to effectively take over everything in the world, how do you think about what AI actually is? I mean, do you even think of it as saying that it could have a motive or like a level of, of consciousness or a series of, you know, sort of like personal goals the way that, that humans do? How do you think about what is actually being grown and created in these labs?

We value your privacy

We use cookies to understand how you use our platform and to improve your experience. Click “Accept All” to consent, or “Decline non-essential” to opt out of non-essential cookies. Read our Privacy Policy.