Risks and Impacts of AI Transition

Bill, you said that there's no plan to ease the

entry into the

AI era, and that we underestimate the impact of,

of AI.

In what way?

Well, over the last year, these AIs have gotten dramatically more

powerful, even faster than I expected, and they're

now capable of causing, um, cyber attack

risk, bio-terrorism risk, psychosocial risk.

And I have to say I'm kind of shocked that the exact criteria

that we review these models with and the actions we take to

minimize the harms are really completely missing,

and the broad dialogue,

we get all of society talking about the trade-offs here, uh,

that's just not taking place. And so, uh, you know, we've gotta

minimize these harms through careful review and criteria,

and we have to accelerate the good stuff or else,

you know, we could have a complete backlash, uh,

that,

uh, AI would suffer from.

I, I talked to Dario Amodei, uh, last May.

He was estimating 50% of all entry level white collar jobs could be

eliminated between... in the next one to five years, 10 to 20%

unemployment possibly in the next, uh, five years.

You write about the, what you think are the three big risks

that come with the transition to A- AI. Can you just go through them?

Yeah, so

the, there's a big risk of

people with ill intent, either

defrauding people,

uh, cyber attacks, or in the extreme case,

bio-terrorism. So, you know, these, um,

negative uses, uh, we need to monitor

things like creating a, a dangerous virus, uh,

to, to see if that's happening. Another problem will

be the job market change, which has not happened

yet, but because the AIs are getting so reliable

now, uh, they can be used for a variety of

well-defined

jobs, and,

you know, that would, uh, be a tough thing to

adjust to. And finally, just the, uh, psychosocial

effect that a kid might have all their friends being

AIs and not be pushed to deal

with real humans and, you know, all the learnings and joy you

get, uh, by being engaged socially.

And so every one of those things, uh, I'm

very concerned and I wanna slow down or eliminate

those negative effects.

Well, I mean, is that possible? I mean, on, on the job loss,

y- you know, uh, the, the counterargument

that some people make is look, uh, you

know, at every technological change there's been

job

losses. You know, with the electric light bulb,

lamplighters lost their

jobs. You, in this essay, make the argument that a lot of

past technological changes, the job losses have, have taken

place over a longer period of time, even several generations in some

cases. These job losses will come much quicker

and all at once.

Yes, it's across every sector of the economy,

uh, initially with white collar, but then

as the humanoi- humanoid robots come along,

blue collar as well.

You will have, for the first time,

the equivalent of human cognition.

So that past history misleads people because there we

were richer and so jobs like, you know, designing rides at

Disneyland, having more psychiatrists, we were richer enough

to create those additional jobs. Here, as the

machine gets better, uh, than human for

well-defined jobs and works 24 hours a day,

it's going to be a

bigger deal. And so the, the bar for somebody coming along

and saying this

time is different is very high, but I'm staking my

reputation completely on the fact that, uh,

in terms of the impact on the job market, this is very

different than any technology that I've been

involved in or that, uh, has come a- across in human

history.