Robert Wright
speaker
123 appearances
1 recordings
1 series
first heard Jun 2026
last heard 23 Jun
Robert Wright’s voice in public audio — every appearance, attributed to the second.
Trend
recordings per month · last 12 monthsRecordings per month over the last 12 months — 1 in all, peaking in Jun 2026 with 1.
Appearances
It sees this one email.
It says it's going to be shut down.
The AI is going to be shut down.
It sees another email that says that the person who's going to shut it down is having an extramarital affair, okay?
So the AI...
took it upon itself to blackmail the person it literally sent an email and these are these people didn't exist of course these are fake people but the ai sent this email to the person who's going to shut it down saying i know you're having this extramarital affair i'm going to expose it unless you change course and don't shut me down so that's that's an example of figuring out a particularly dark way to use power figuring out that you need
to use the power if you're going to stay alive, and finally choosing to stay alive.
That itself was not, you know, this goes back to science fiction movies where this has happened, but now we know that apparently these AIs will resist
kind of their own extinction, so to speak, presumably because they all have goals and to pursue the goals, they got to stay alive, right?
That's the thinking.
There are various theories, but there's a certain amount of this kind of behavior that definitely bears watching.
Well, I think AIs will naturally pursue certain courses that are in some ways concerning, including deception.
We've seen them, for example, when they are told in these experiments that, OK, you're under evaluation, you're not being released yet.
They will, once they know what the evaluators want in the machines, they will try to comply with that and kind of act nice or act in compliance with what they know their overseers want.
And then when released, they may change course, having gotten the green light.
So that's a kind of deception.
All of this, you know, means we should, A, keep an eye on them.
There's also this whole field of so-called alignment, right?
where we try to engineer the AIs to not do these things, to not deceive or not deceive in certain circumstances, to be honest, and so on.
And this is possible.
Showing 41–60 of 123 · page 3 of 7
← Previous
Next →