Probably Science

I am here with Stuart Russell, the author of Human Compatible, Artificial Intelligence, and the Problem of Control.

Thank you for joining me.

Nice to be here.

So this book was great. I've been fascinated with all of the possible pitfalls and great things that AI might bring us.

And there are so many that I hadn't even crossed my mind until reading this book.

There are lots more that I didn't get into.

So to a layperson who thinks this isn't even an issue worth bothering ourselves with right now. What is the,

what is your short way of convincing them that this is worth anybody's time?

So two way. One is it's, uh, it's already happening in the sense that we're already building

and deploying AI systems, uh, the wrong way that are having serious negative effects. And the social media catastrophe

is probably the easiest example to understand because that's, so on the social media platforms,

whenever you read something or watch something, it's been fed to you by an algorithm

that is trying to maximize click-through

or eyeball time or some other metric. And in doing so, they forgot to include a whole bunch of

other things like not turning the world population into neo-fascists.

And so since the algorithm wasn't told that it wasn't supposed to do that, that's what it did.

By modifying our preferences through feeding us sequences of content that gradually lead us to be much more extreme and therefore much more predictable versions of ourselves.

So the algorithm only cares about whether you're predictable, because the more predictable

you are, the more money it can make off you.

But it so happens as a side effect that it seems to have turned many people into much more extreme versions of

themselves.

...