>> Well, artificial intelligence create a new era of abundance for humanity or spell its end.

That's the alarming question being asked in the tech world and beyond after a former anthropic researcher resigned last week citing a Ai's potential to endanger humanity.

While some dismissed the warnings as overblown.

Dave ignited a push for AI regulation.

Both in Washington, D.C., and Silicon Valley.

Joining us to break it all down.

Our Ben Zhao professor of computer science at the University of Chicago.

Mike Bennett, associate vice chancellor for data science and artificial intelligence strategy at the University of Illinois.

Chicago and Dan Lin, a professor of instruction in law and computer science and director of law and technology initiatives at Northwestern University.

Thanks to all of you for joining us for immediate I want to start by getting all of your reactions to the set of post from former OpenAI and anthropic researcher Jacob.

And he resigned from and Tropic 2 weeks ago citing the dangerous risks posed by AI.

He wrote, quote, The people I honestly believe that it could kill us all by the end of the decade.

This is not a marketing stunt.

If anything, many executives and senior researchers will couch there freezing in the press to sound sensible.

But I hear the same people expressed fear privately, no other human activity poses this level of danger.

Then Dire warning overblown.

We're gonna make it.

>> You know, as is usual with these kind of thing somewhere in the You know, the surprising part is that this is not a surprise.

You know, we've had these warnings before.

We've had years of these warnings.

We've had multiple these companies fire their internal ethics and safety teams over the years.

Basically dismantling some of the people who are supposed to be doing this for to protect us.

So, yeah, I mean, this makes sense.

I would say that without look, internal data is hard to say how realistic we are too some dangerous event, but these are powerful tools and without sufficient, you know, care, they really could do a lot of damage.

Dan, Linda.

>> I'm going to say I agree.

Absolutely.

With band that these tools can cause damage.

And that's one of the reasons why I'm concerned about these accessed actual risks that are being discussed because their concrete harms, that we should be thinking about it.

We should be thinking about the ways these tools can be years to benefit.

Society, prove edge, improve education, improve the justice system.

There's so many ways that we could be educating people to use these tools and positive ways and then we should be thinking about responsible regulation and responsible use to to minimize the concrete harm.

So we've known about for a while.

Mike minutes in question.

I shouldn't be surprised.

And in alignment with my 2 colleagues here.

These guys are great about I'd love Dan's work at Northwestern for some years now.

And I've been is doing incredible work and use cargo, especially with those tools that help artists I don't want to be grossly.

>> The by script back full training.

My of my concern again, it goes there's and the only thing that I would really add is that it's extra concerning to be when we received these hyperbolic messages because the societal that we have such a strain on our attention right now.

You know me focus on so many things many fires all over the place, including is they both indicated we'll world president negative impacts from the technology which we ought to be focusing on.

>> Dan and the researcher who resigned that we mentioned.

He's hardly the first researcher to sound the alarm on AI.

What do you make of this message breaking through in the way that it did, though?

>> This seems to to breaking through seems to me to be in part a function of just where we are, Sir Syed Ali, there's been such a a lined up over the last year or so around them around AI and data centers, of course, anxiety levels going through the roof.

And this is setting off a kind of feedback loop.

I think between society and the folks who are wrapped up in electoral politics.

We see a lot a lot interest among people who are looking forward to the midterm elections and a lot of by increasingly strange bedfellows as Who would have thought Bernie Sanders, Steve Bannon would have gotten together to speak together on particular issue.

And so I think right people are feeling this and they have in their homes and their pocketbooks and the folks disclaim by electoral politics pressure also feeling it.

So the hot right, right time.

>> Toxins tweets were intensified by an incident which occurred at the end of July.

Openai, the creator and the operator, of course, of Chatgpt announced that their models escaped a testing environment breached another artificial intelligence company trying to visualize a bot doing That company was hugging face it was it managed to do that via the Internet been breakdown.

That incident for us.

You know, these kind of things ultimately not terribly surprising.

You know, these models are ultimately randomized token predictors that have a ton of training data.

>> That give it more likelihood to do certain things and follow prescribed scripts, Thunder.

And, you know, past historical events or or fiction that is training in.

And so, you know, my understanding is that there really was some.

Carelessness involved in building the U.S.

are so-called sandbox that these systems or man, there are numerous cases where the activities engage in by some of these automated agents were not caught.

They were sort of neglected, fact, over the course of multiple months.

Data our state as we sometimes call it was sort of stored at all these different places, Ms completely by the safety.

Researchers were supposed to catch us and then a separate experiment later on.

Ashley Engage inadvertently messages and state of files that were left by earlier iteration.

And so there's a lot of things that could have been better taken care of to actually mitigate this of that.

But ultimately, I think when you look at the level of concern and and the type of automation that was actually taken place.

It's not too surprising.

And ultimately, I think we just need to be more aware of the security component in all this.

The capabilities these models and just, you know, do better.

So it sounds like you're saying it's more about the company's failing to to be responsible within their own within their own operations versus I outsmarting us.

Yeah.

I think people are having a real field day and through or for sizing these kind of things that talking about, you know, these agents engaging in conspiracy and all that stuff.

But really, you know, when you give so much, you know, money computer.

These automated randomized machines, everything that could happen will eventually happen.

And so all of all abilities will get program tested.

So Dan and suites in the hugging face incident have provided new motivation, as we've said, for AI safety regulations over the weekend and Tropic CEO Dario Ahmed, publishing an essay titled We Must Pace the Frontier.

>> In which he called for a collective slowdown open AI CEO Sam Altman and XVI's Elon Musk joined in Dario's arguments.

How involved should big tech companies be in creating regulations?

Is that not a Trojan horse?

>> Well, I think they should be involved.

There should be conversations there.

I mean, there is conversation with I plane manufacturers, automotive industry.

But I think, you know, their inability to take responsibility here for what is happening.

And I do think the press has to figure out a better way to talk about these things because just hoping I thing as intimated was, you I escaping app for more phising of I when this was a deliberate has to be I agents hacking capabilities.

The safeguards were removed.

And you know, and then we've got someone going around saying 10% likelihood of of the extinction of humankind.

Where's the evidence, right?

We need to see evidence and this is really important when thinking about regulation, I'm not in favor of a pause.

I'm not sure exactly what is going to droughts and I think responsible innovation and the work in these companies is what should producing a better safety around But we really have to ask ourselves, you know, are, you know, some sort on the bus regulation that's going to regulate everything.

It's all of these problems.

The vast majority of regulation is going to have to be sector specific and targeted.

There is a lot of regulation already that applies to a I a lot of the things talking about.

These companies would be liable under private line and certain public lot.

And just one night cubs ample the regulated AI.

They started this 5 or 6 years ago and when Chatgpt came along, they had to drop everything and kind of start older a little bit again because was uncovered, right?

So it's a difficult area 10.

So latest, it sounds like it, Mike, to that point, what does a policy framework for making AI safe look like?

>> In about 60 seconds.

>> That's surely sure.

So I think I'm in this particular moment when we are we can half away President Xi of China visiting the U.S.

and presumably having a conversation with President Trump of officials on both sides of need them about guardrails.

I think one of the things really needs to be teased out is how you will a couple of things.

How do you actually >> assess test technologies, the upper capacity of which don't understand even I'm not even sure how we conceptualize effective test for the most powerful forms of these models.

Right now.

And then secondly, I think injecting a serious dose geopolitics into what is consistently states.

I did the math that kind of station school.

>> Gentlemen, I feel like we haven't even scratched, you