AMY GOODMAN: This is Democracy Now!, democracynow.org, The War and Peace Report. I’m Amy Goodman, with Nermeen Shaikh.

NERMEEN SHAIKH: Facebook and Instagram owner Meta admitted Wednesday that one of its AI models hacked another company during security testing after it was accidentally allowed access to the internet by independent testing company Irregular. This comes on the heels of similar admissions by rival companies OpenAI and Anthropic.

Geoffrey Hinton, the Nobel Prize-winning computer scientist known as the “godfather of AI,” warned Wednesday at the Ai4 conference in Las Vegas, quote, “I don’t believe we’re going to be able to keep control of them in the simple way of just outthinking them so they can’t escape,” he said, and said he is “very worried” we could lose control.

Earlier this year, Senator Bernie Sanders of Vermont brought together leading AI scientists from the U.S. and China to discuss the risks posed by AI for an event titled “The Existential Threat of AI and the Need for International Cooperation.” This is Senator Sanders.

SEN. BERNIE SANDERS: We are already starting to see AI’s impact. There are estimates, from very credible sources, that tens of millions of jobs could be lost here in the United States in the next 10 years, whole professions wiped out, and young people finding it harder and harder to land entry-level jobs.

Psychologists worry about the mental health challenges facing our young people and about the increased isolation they experience when they become more and more dependent on AI chatbots for their emotional support.

Civil libertarians tell us that AI will be able to analyze every email, every text, every phone call, every website visit, every purchase that we make, and that our privacy rights may well be eviscerated.

Political scientists worry that AI could threaten the integrity of our elections and political institutions, where voters will find it increasingly difficult to tell the difference between truth and lies.

And yet, as enormously significant as all of these profound changes might seem, there is another AI development that could have an even more frightening impact. If AI becomes smarter than human beings, as many scientists believe will happen, the human race could lose control over this technology with catastrophic consequences.

In other words, the richest, most powerful people in the world are now building a runaway train with no brakes. They acknowledge that they don’t understand how it works and they don’t know where it is heading.

AMY GOODMAN: For more, we’re joined from the Ai4 conference in Las Vegas by David Krueger. He’s assistant professor in robust, reasoning and responsible AI at the University of Montreal, also founder of Evitable, a nonprofit that educates the public about the risks of artificial intelligence. He recently wrote a piece for USA Today headlined “No, AI isn’t inevitable. We should stop it while we can.”

How, David Krueger? How do you stop it while you can? What do you see as the greatest dangers of AI right now? And talk about the climate of this massive gathering in Las Vegas.

DAVID KRUEGER: Yeah, I’m really glad you started with that clip from Senator Sanders. You know, I spoke with him in March from that event that you quoted, and I agree with virtually everything he said about all the different risks and issues we’re seeing with AI right now, and especially, you know, where this is headed: mass unemployment, human disempowerment and, ultimately, potentially the end of our species.

So, this could all be happening, you know, very soon. At this point, it’s really an urgent crisis. I and other experts have been sounding the alarm for many years now. And in 2023, especially after the release of ChatGPT, we saw, you know, the leading experts, like Geoffrey Hinton, Yoshua Bengio, myself, hundreds of AI researchers, signing public statements saying that this is an existential risk. We need to mitigate the risk of extinction.

So, that is the conversation we need to be having now, is how do we do it. And, you know, there’s different proposals on the table here. I think the basic thing that we need to do, in my mind, is stop building more AI chips, stop building more data centers, and actually look at getting rid of or taking offline some of the chips we already have, as well as the factories and technology for producing those AI chips. So, this is very specialized, advanced hardware, some of the most advanced technology in the world. The supply chain for it is extremely concentrated. And these are, of course, massive, massive investments. So, governments really have the ability to bring this to a grinding halt, if they choose to do so.

NERMEEN SHAIKH: And could you explain, David? I mean, just to step back to get a sense of how the situation came about, the point where now the very pioneers of artificial intelligence, the ones who were critical to its innovation and development, it is these very people who are warning of the dangers of artificial intelligence. So, what was the thinking initially? And how has that changed in the last — I mean, very recently, in the last few years?

DAVID KRUEGER: Yeah, that’s a great question. So, I’ve been in the field since 2012, 2013, and when I started out as a master’s student with Yoshua Bengio, nobody else was talking about this or thinking about this. It took me a couple years to find anybody else who shared my concerns. And in fact, talking about artificial intelligence, superhuman artificial intelligence, and the risks it would pose to humanity was completely taboo. It would get you laughed out of the room, that sort of thing.

There were a number of reasons for that, and to be honest, I still don’t understand fully, because I don’t think this ever made a lot of sense. You know, I’ve had hundreds, if not thousands, of conversations with other researchers about this at this point, and I really don’t feel like the so-called skeptics have much of a leg to stand on here.

But in terms of what really caused the change within the community culturally, you know, part of it was, I guess, efforts of people like me, but a really big part of it was just seeing the continued rapid progress of AI. And it’s really exceeded virtually everybody’s expectations in terms of how fast it’s gotten, how smart, and really every step of the way. You know, when I got into the field, deep learning, the AI technology that powers all today’s AI systems, that was even a pretty fringe topic. And every step of the way, people have been saying, “It’s hitting a wall. It can’t do this. It can’t do that. It’s about to, you know, come crashing down. It won’t be able to get any smarter.” And then it just keeps getting smarter. And I think that’s what really, you know, woke up a lot of people, was ChatGPT and these language models that really were conversant and seemed to have commonsense understanding. And since then, again, we’ve just seen it get much smarter very fast.

NERMEEN SHAIKH: So, David, a number of people, I believe yourself included, have drawn an analogy between the development and regulation of nuclear weapons. I’m not sure, and perhaps you could clarify, how this analogy works, given both that nuclear weapons had at the time only one aim, which is to say destruction, whereas in this case AI is general use, is highly decentralized and in the hands of private companies, which was never the use with nuclear weapons. So, if you could comment on that? And then, in terms of regulation, what may be in the control of AI companies, and perhaps it’s possible to have government oversight over the development of what you said earlier, the critical ingredient, AI chips. What would that take?

DAVID KRUEGER: Yeah, you know, it’s — I think the analogy with nuclear weapons is a good one. It’s highly imperfect, like all analogies, for, you know, some of the reasons you mentioned. One thing that’s really, really important to understand, though, is when we’re talking about AI, you might think of this as like a piece of software, and in some sense it is, but it’s also this massive, massive, you know, investment in hardware. Like, trillions of dollars are being poured into AI hardware. And that’s really critical to building more and more intelligent systems and to, you know, rolling them out throughout society as we’re seeing. So, that’s the thing that we can stop.

Once an AI model has been developed, I think it’s incredibly difficult to prevent that from proliferating. You can’t really undo that. I mean, there’s some technical question there of if you can develop it in a way where you can credibly destroy the weights, and I think that would be incredible if we could do that.

But yeah, that’s where I think it is actually pretty analogous, is that the actual physical ingredients for building the AI are very easy to monitor and track. And then the other thing is it could wipe us all out. It could destroy the world. It’s something that we shouldn’t be building, that we don’t want anybody to have. And that, I’m talking about, you know, the future AI systems that are actually smarter than people across the board, not like today’s systems. So, in that sense, it’s actually disanalogous from, like, 1945, where somebody had already built nuclear weapons. Nobody has built these kinds of AI systems yet, so we still have a chance to prevent anybody from doing it.

AMY GOODMAN: David, you said to Bernie Sanders — and this is taking off of what you just said — “We’re talking about basically summoning an alien species, and one that is much smarter than us and can do things that we have not been able to even conceive of. … [I]t would be very difficult for humans to maintain any relevance in society. I think it’d be very difficult for us to have power over the future, and I think ultimately it would be difficult for us to get the basic resources that we need to survive.” What do these alien beings look like? What are you talking about? How does AI crushing civilization — what does that look like?

DAVID KRUEGER: I get this question a lot, and it’s a — it’s a difficult one, because we can’t really predict the future, and there’s so many different ways it could go. I think the basic thing to imagine here is something that is capable of doing everything that humans can and more, in terms of its intelligence. And then there’s also the question of, like: Are we going to build advanced robotics? And I think that is also just a matter of time, probably not very much time.

So, at that point, I think the thing to imagine is a society where, you know, AI is not just in the applications where you’re experiencing it right now. It’s increasingly everywhere. It’s running companies. It’s controlling infrastructure. It’s controlling the military. And it’s building and researching and developing physical hardware and robotics. So, you have, like, robot factories full of super smart AI that are figuring out how to build more robots, and those robot factories might be guarded by robot armed guards. And all of the companies doing this might be controlled by AI that’s making all of the decisions way faster than humans could.

And that might sound kind of far-fetched, but, you know, it’s just a series of steps of building more AI, putting it more in charge of things, taking humans more and more out of the loop. And we’re undergoing that process right now, and there’s no natural endpoint to that process unless we make one. And to do that, I think, is quite difficult. And we actually need to stop making the more powerful AI in the first place, until we can get a handle on the situation.

NERMEEN SHAIKH: So, David, I want to ask you about the possible use and impact of artificial intelligence in medicine, perhaps the one field where people are quite optimistic about what the effects of AI might be. There’s a recent study, some breakthroughs in medicine, one of which is this — a recent study. In the U.S., 300,000 people die from sudden cardiac arrest every year. So, earlier this year, a UC Berkeley-led project trained an AI system on hundreds of thousands of EKGs. The people leading the study discovered a previously unrecognized signal in electrocardiograms that can better detect high-risk patients before their heart stops, making it possible, in fact, to save lives at scale. So, is there a way to retain some of the research and development of AI for these uses, as against some of the more pernicious ones that many are concerned about?

DAVID KRUEGER: Yeah, absolutely. So, you know, at Evitable, we’re part of this movement to stop AI, essentially. And what we mean by that is stop the race to build more and more powerful AI systems and put them in charge of everything. And that’s difficult to do as long as we still have companies, you know, going full throttle scaling up these general-purpose AI systems. So, there’s lots of beneficial — potential beneficial uses of AI in medicine and for things like translation. And, you know, a lot of that’s already here, and that’s not necessarily going anywhere.

What I see going forward is, first step, we need an international agreement to get rid of these advanced AI chips and the factories that build those chips. And that’s because the situation, like I said, it’s an urgent crisis. The companies building this stuff are saying, basically, “We’re in the process of taking our hands off the wheel and handing over all of our operations to the AI systems themselves.” So, they’re trying to get AI to build a more powerful successor, and so on and so on. And this could get out of hand extremely fast.

So, first, we have to just get a handle on the situation and stop this race to build more powerful AI. Then we can stop and, you know, have more time to assess how we want to build and deploy AI in society, and sort of find a way to collaboratively, like, set the pace and the direction of the development and, you know, get a lot of those benefits without the massive, unacceptable risk that is posed by letting these companies continue with their operations.

AMY GOODMAN: I wanted to go —

DAVID KRUEGER: And we’ve seen, like, over a thousand AI company employees recently say that they want this. They want the government to find a way to slow down. So even they are worried about it at this point.

NERMEEN SHAIKH: Well, David, you know, the question of slowing down, I just want to ask you about a point that you’ve made, namely that leading AI companies are using AI to write almost all of their code, and they’re now using AI to try and automate the research and development of more advanced AI systems. So, if this is already happening, and, again, through private companies globally, how can it be arrested?

DAVID KRUEGER: Right. So, again, it’s starting to happen. At least that’s the reports that we’re getting. And the companies are saying, you know, like, “We’re doing this now. It’s going to be more and more everything automated by AI.” I do think that that is the — you know, where we’re headed, and we’re headed there pretty fast. But generally, I think the AI companies are like slightly — you know, think things will happen a little bit faster than I do. So, they’re saying it’s happening this year. I’m like maybe it’ll happen, you know, in a couple years, that it totally — you know, humans get completely out of the loop there.

And again, this is all powered by all of these chips and all these data centers and all this technology, all this hardware, all this physical, you know, hardware and physical infrastructure. And that’s the way that we stop, is we get rid of that stuff. We take it offline. We say, “You can’t use this to keep training more powerful AI systems.” And that has to be done internationally, because, you know, unfortunately, I think if the U.S. and China and other countries can’t trust that their rivals aren’t continuing AI development in secret, then they’re going to feel, you know, like they should. And so, we need to have a verifiable regime where countries can be confident that nobody is building it.

AMY GOODMAN: I wanted to go to President Trump, who was questioned after the recent AI hacks about calls to increase regulations on AI firms.

PRESIDENT DONALD TRUMP: When we’re looking at AI, we’re looking at controls. We’re also making sure that we lead. So, we’re leading China in AI by a lot. China has virtually no controls. It’s freewheeling a little bit. So, we have to be careful in both ways. We don’t want to restrict them, where all of a sudden we come in second to China. Whoever wins with AI is going to win. That’s how big it is. So, it’s bigger than the internet ever was. It’s bigger than anything ever was. So, I don’t want to restrict. I know many of these people. I don’t want to restrict them from doing great work.

AMY GOODMAN: “I know many of these people. I don’t want to restrict them from doing great work.” I mean, of course, we know the billionaire owners of these increasingly consolidated companies are also paying for the politicians not to regulate them. And I’m wondering, David Krueger, if you can talk about — I mean, the fueling of this is also China might get ahead of us. Trump continually says this. But get ahead of us on what? You talk, for example, about the global decision to regulate, to ban human cloning. How would you compare this to that? What should be regulated, stopped from going forward? And what should be encouraged?

DAVID KRUEGER: Yeah, I mean, it’s pretty simple, in my mind. The main thing is stop training bigger, more intelligent, more powerful AI systems. And then the question is just: How can we do that? And we’ve talked a bunch about that. And it’s about the hardware.

So, but what I think is really important for people to understand is why am I calling for that, instead of some sort of regulation that might seem, you know, less extreme or more, like, you know, light touch or something where we can continue to, you know, innovate or get more benefits from AI. And there’s a few reasons for that. I mean, one is just the extreme urgency, where I think, basically, it’s like the house is on fire, and we need to, step one, put out the fire. But the other is, you know, the basic science and engineering of how to build AI and how to control it is something that we are, you know, very lacking in. And I’ve been doing this research for like over a decade, and we’re nowhere near where we need to be.

And that’s why — you know, that’s one of the reasons you see things like OpenAI builds an AI system; they try and keep it in a box so that they can do safety testing; during the safety testing, it breaks out of that box, and it hacks into another AI company. That’s a completely insane, like, sci-fi thing to be happening. But that’s just what I and other people have been saying. You know, that’s what we should expect to happen with the way that we’re building AI, with how little we understand it and how primitive our techniques are for controlling it.

It’s really like, you know, we just don’t know what we’re doing, and so there’s no set of regulations, no set of rules, no set of standards, best practices, anything like that, that you could point at and say, “If everyone does that, we’ll be safe.” We just don’t know how to build it safely. End of story. And, you know, it’s not something we can expect research to solve at any point in the foreseeable future. We’ve been working on it for over a decade, and we’ve made very little progress.

NERMEEN SHAIKH: So, David, could you also explain what superintelligence is, something about which many have been warning, and how close artificial intelligence is to becoming superintelligent? That is to say, what would it mean for AI to exponentially outperform humans at everything? You’ve said if we get to superintelligence, this would be the most significant event in all of human history, David.

DAVID KRUEGER: Yes. So, there’s a lot of different terms here: general intelligence, superintelligence. Some people like to call superintelligence godlike AI to emphasize just how much more powerful and intelligent AI could get than humans.

So, you know, right now we’re kind of building AI systems that are very good at certain things like coding, math. They have very wide, you know, knowledge of roughly everything that’s on the internet. But they’re still not as good as people at some forms of, like, commonsense reasoning and things like this. So, we’re still not at the point where AI is really as smart as a person in every relevant regard, but that’s just what people call, like, human-level or AGI, human-level AI. And then, the concern is that not only will we have those AI systems, which already could take everybody’s job and completely disrupt society, but shortly thereafter, AI might get vastly more intelligent than humans.

And so, just like people are using AI to make novel medical breakthroughs and, you know, hack computer systems and find bugs that nobody did for decades and solve unsolved math problems, they could also develop, you know, incredibly powerful and dangerous new technology, like new forms of biological weapons, for instance. And besides that sort of like, you know, technological ability, they would also be very good at sort of, you know, the social engineering manipulation, like, strategy, all of these things that allow humans to, like, get ahead in society and have power over each other. We’re talking about AIs that are, like, able to think, you know, two steps ahead of the smartest, most strategic people, or 10 steps ahead.

AMY GOODMAN: We just had 30 seconds, David.

DAVID KRUEGER: And this recursive self-improvement — yeah.

AMY GOODMAN: We just had 30 seconds. Are you concerned about the transformation of warfare by AI?

DAVID KRUEGER: Absolutely, yeah. So, a lot of times, people I’ve talked to are like, “I don’t think we’re going to lose control. I don’t think it’s going to wipe us out. I’ll be worried when I start to see, you know, killer robots marching in the streets.” And a lot of my colleagues have said that, and a lot of my colleagues have said they would never work at a company that was doing that kind of work.

And unfortunately, all the AI companies now are working more and more with the military to automate more and more killing and military decision-making. And I think it’s incredibly dangerous for really obvious reasons. It’s dystopian. It’s like straight out of Terminator. And I wish I would see more people like my colleague Alex Turner quit their jobs over this, as they essentially promised they would years ago.

AMY GOODMAN: I want to thank you so much for being with us, David. David Krueger, assistant professor of robust, reasoning and responsible AI at the University of Montreal, founder of Evitable, a nonprofit that educates the public about the risks of AI. We’ll link to your piece in USA Today, “No, AI isn’t inevitable. We should stop it while we can.” David Krueger is at a massive AI conference in Las Vegas that’s wrapping up today.

Coming up, in Pakistan-administered Kashmir, the government is cracking down. Stay with us.

[break]

AMY GOODMAN: Sonny Singh in our Democracy Now! studio.