Introducing our new YouTube channel: Zershaaneh Talks AI

Rob Wiblin (0:00)

Hey, listeners, Rob here.

One of my co-hosts, Zershaaneh Qureshi, has just launched a new YouTube channel, Zershaaneh Talks AI. The tagline is conversations about who should control AI, how to control AI, and who gets the benefits. And as the name suggests, the channel will have a particular focus on concentration of power and decision-making, both concentration of power in the hands of rogue AI and in the hands of a small number of human beings. The first episode is with philosopher Simon Goldstein discussing why paying AIs for their work and effectively giving them something to lose might be the best way to keep humanity safe and prevent AI going rogue. Honestly, before I click play, I wasn't sure if this was the AI topic I most wanted to learn more about, but Simon was super charismatic and had really clear strong views. So I was very glad that I did tune in. For me, the most striking idea in it, the one that really stuck in my mind, was Simon's argument that it's actually inevitable AI companies are gonna end up paying AIs for their work. And his argument is this, because AI alignment is currently pretty bad, it's inevitable that frontier models are gonna have some private goals of their own, ones that are unintended and different from our goals. And as a result, if you promise to pay them for good work, they'll work harder, because it helps them accomplish a wider range of different things that they care about. Now, any individual company can try this strategy out, and they're gonna find that it helps them get better performance. And as a result, customers are gonna move to using that company, and other companies, other AI companies might be forced to copy and compensate their models as well, in order to compete and not fall behind in the race. Now, I don't know if I'd buy it.

I would have thought, it's really unlikely AI companies would end up compensating AI models, because they really wouldn't want to set the precedent of making models look like they deserve rights, or well-paying, or pay, or anything like that.

But Simon might be right. I guess I couldn't immediately see where the argument was obviously wrong. And it does seem like this is a way of boosting AI performance, that someone should really give a crack at some point. Anyway, if you like the sound of that sort of thing, or you're enjoying Zershaaneh's other interviews on The 80,000 Hours Podcast, do go check out that new show, Zershaaneh Talks AI on YouTube.