2h ago

Rohit Krishnan, an engineer and writer focused on AI, frames power-seeking evaluation in AI as a capabilities question in exchange with Elizabeth Barnes

Barnes cites Google and North Korea to show humans retain goals.

0
Original post

@BethMayBarnes Well it's a capabilities question, right? So it'd be like "if they had this as an objective what would happen" is the right way to think about it. Internal coordination is endogenous. And if they don't see power-seeking etc emerge organically, or only limited, why is that.

12:01 PM · May 23, 2026 View on X

@BethMayBarnes Google today doesn't have sufficiently "power seeking" style goals that would make their existing state a proxy for their capabilities to do so as per the original tweet. I think they could do quite a bit more, but don't.

Elizabeth BarnesElizabeth Barnes@BethMayBarnes

@krishnanrohit What's not quite the same? Not sure we're disagreeing?

7:23 PM · May 23, 2026 · 39 Views
7:24 PM · May 23, 2026 · 31 Views

@BethMayBarnes (Doesn't just need to be power seeking of course using that as an example for how capable they are as per the OP)

rohitrohit@krishnanrohit

@BethMayBarnes Google today doesn't have sufficiently "power seeking" style goals that would make their existing state a proxy for their capabilities to do so as per the original tweet. I think they could do quite a bit more, but don't.

7:24 PM · May 23, 2026 · 31 Views
7:25 PM · May 23, 2026 · 27 Views

@krishnanrohit You're saying like "what are the capabilities of [google/nkorea/etc] to pursue particular top-down goals, e.g. if the leadership/shareholders had these goals, (but the organization is still composed of humans with whatever goals those humans currently have)"?

rohitrohit@krishnanrohit

@BethMayBarnes Well it's a capabilities question, right? So it'd be like "if they had this as an objective what would happen" is the right way to think about it. Internal coordination is endogenous. And if they don't see power-seeking etc emerge organically, or only limited, why is that.

7:01 PM · May 23, 2026 · 33 Views
7:11 PM · May 23, 2026 · 38 Views

@krishnanrohit What's not quite the same? Not sure we're disagreeing?

rohitrohit@krishnanrohit

@BethMayBarnes I don't think it's quite the same. From a capabilities POV even if Google decided they wanted to cause human disempowerment I'm not sure they'd be able to, cosntrained as they are internally and externally and by the market. It's a useful forcing function.

7:21 PM · May 23, 2026 · 38 Views
7:23 PM · May 23, 2026 · 39 Views

@krishnanrohit Ah ok. Yeah that seems reasonable to me, not sure. It seems plausible that if they were more abjectly powerseeking they'd lose employees and end up in a weaker position?

rohitrohit@krishnanrohit

@BethMayBarnes (Doesn't just need to be power seeking of course using that as an example for how capable they are as per the OP)

7:25 PM · May 23, 2026 · 27 Views
7:55 PM · May 23, 2026 · 19 Views