Another think coming? An AI pioneer steps down

Hey, Ryan Reynolds here wishing you a very happy half off holiday because right now Mint Mobile is offering you the gift of 50% off unlimited. To be clear, that's half the price, not half the service. Mint is still premium unlimited wireless for a great price. So that means a half day.

Yeah.

Give it a try@mintmobile.com Switch upfront payment.

Of $45 for three month plan equivalent to $15 per month. Required new customer offer for first three months only. Speed slow. 135 gigabytes of network busy. Taxes and fees extra.

See mintmobile.com hey, Ryan Reynolds here for Mint Mobile. You know one of the perks about having four kids that you know about is actually getting a direct line to the big man up north. And this year he wants you to know the best gift that you can give someone is the gift of Mint Mobile's unlimited wireless for $15 a month.

Now you don't even need to wrap it.

Give it a try@mintmobile.com Switch upfront payment.

Of $45 for a three month plan equivalent to $15 per month. Required new customer offer for first three months only. Speed slow after 35 gigabytes. If network's busy, taxes and fees extra. See mintmobile.com.

Hello, and welcome to the Intelligence from the Economist. I'm your host, Jason Palmer. Every weekday we provide a fresh perspective on the events shaping your world. It's been clear since the start of Russia's invasion that Ukraine's air force needed some new firepower. As the Western taboo about providing fighter jets crumbles, we look at which ones they might get. And in the gaming world of good versus evil, it's not uncommon to find Nazis as the baddies. But players haven't had to deal with what the Nazis stood for until now. We play the world's first video game about the Holocaust. First up, though, There's no doubt that artificial intelligence is racing ahead. The striking capabilities of ChatGPT being just one example. With that rise in capability, though, comes a rise in concern. This week that concern came from someone known as the godfather of the field. Geoffrey Hinton told the BBC he was stepping down at Google so he could voice his concerns over what is in large part his own creation without it.

Reflecting on his employer, I've really gone.

Public to encourage people to think very seriously about what they're going to do to make sure that if we develop things more intelligent than us, they don't gain control.

He's just the latest in a growing list of experts and pundits who are raising the alarm In March, hundreds of them called for a six month moratorium on research into the tech behind models like ChatGPT.

An open letter signed by Elon Musk is calling for a pause on the development of more powerful artificial intelligence Systems.

More than 1100 signatories call for a moratorium on state of the art AI development.

Tomorrow, the bosses of Anthropic Google, Microsoft and OpenAI will be at the White House to discuss the safety of their chatbots with America's Vice president. But for now, those who are worried about AI may simply be louder than those who are relaxed. And it's not clear who's best placed to judge the risks.

I don't think Geoffrey Hinton is saying things about AI that we haven't heard before. He's essentially articulating a lot of the concerns that other people have already expressed.

Tom Standage is a deputy editor of the Economist.

But coming from him, I think people are more inclined to sit up and pay attention because he is one of the founding fathers of the field. How so?

What's his background?

Well, if you look at the history of AI, it sort of had these ups and downs. And Geoff Hinton was one of the people who kept the flame alive during the absolute darkest and most depressing period known as the AI winter. And he's long been a great believer in the power of artificial neural networks. And this is where you model something inside a computer that's derived from the structure of the human brain. And during the 1980s, people went in a completely different direction of using expert systems and trying to codify knowledge in symbolic terms. Didn't work. And so AI kind of went off the boil. A very small number of people were trying to keep the flame of neural networks alive. And Geoffrey Hinton was absolutely instrumental in that. And working with Joshua Bengu and also with Jan Lecun, they finally started to make progress. And that led to the explosion of AI that we've seen in the last decade. So it turned out they were right all along. And as a result, the three of them won the Turing Award, which is called the Nobel Prize of Computing, for their contributions to finally getting AI to work.

And yet now Mr. Hinton says he doesn't want anything to do with it.

What are his concerns?

Well, he's not saying he doesn't want anything to do with it. He wants to be able to speak freely about his worries about AI, because the technique that he pioneered, neural networks, those networks have got very big, or deep, as we call them. So deep learning that has absolutely dominated the field for the last 10 years, and just in the past few years, it's got much, much better. We've seen ChatGPT is the classic example, the breakout iPhone mom for AI, and he's worried that it's suddenly getting much better, much more quickly. And he thinks that we need to be doing more to take seriously the bad uses that might be found for this technology and regulating it a bit more and that kind of thing. And he feels that he can't really speak his mind if he's working at Google. And he says this isn't because he thinks Google is doing anything particularly bad. In fact, he says he thinks Google is doing some quite good things in this regard. But if he says that while he works at Google, he just looks like he's defending Google. So he wants to be able to really say what he thinks, this awesome figure in the field.

But what is it that he'd like to speak out about?

Okay, they're getting much better, much faster.

What's the worry?

Well, he's worried about the misuse of this technology, that it can be used to generate misinformation, which is something that other people have expressed concerned about. And he also says that this is a new sort of intelligence and it might be superior to human intelligence already in some ways. That the way in which it can pick things up, the way in which it can absorb large amounts of information, I mean, a human can't read the whole Internet, but these systems are trained on massive chunks of the Internet, so they're very knowledgeable on lots of different things, although they're obviously making mistakes sometimes. So he's concerned that we kind of don't know what we're dealing with. He's likened it to an alien intelligence as well. And he says that's all cause for concern just because of the unknowns involved.

And what do you make of those assertions?

Well, I have to say, I mean, if you look at those three founding fathers, they are taking different positions on this. Hinton has taken a more sceptical and concerned position. Bengio is in the middle.

It's like any technology in the sense, the more powerful they are, the more useful they can be, but also the more dangerous they can be if they're misused.

And Yann Lecun is taking a completely different position. He says, actually we need to focus on the positive aspects of this technology. The idea that these chatbots are going to sort of take over the world and kill us all is ridiculous.

Thinking that somehow we're smart enough to Build those systems to be super intelligent and not smart enough to design good objectives so that they behave properly, I think is a very, very strong assumption, a very low probability.

So even within that group of three people, there is a divide. And this reflects a deeper divide within the AI community more broadly, that some experts are very worried about this and are sounding alarms and others are saying, oh, come on, don't be silly. The idea that we're going to jump from LLMs to Terminator style robot or other bad consequences is going too far. So it's not just Geoff Hinton who's worried about this. Stuart Russell, another great AI pioneer, is also concerned about it. And then we, Gary Marcus on our Babbage podcast a few weeks ago, another AI expert, and he's also expressed concerns.

Every day, for example, I see some new crazy thing that it does or that somebody does with it. So one day we discover that people are using it to make computer viruses. Another day we discover that someone's using it to make child porn. The world is full of bad actors and those bad actors have new tools.

And so there is this interesting split within the community itself and we're seeing that argument now sort of spilling over into society more widely.

And what do you yourself make of that rift in the AI community?

So it's interesting that there's a split, but it's also the case, it turns out that when you ask experts in any field to talk about what might happen in their field, they are less accurate at forecasting what will happen than the best forecasters, the best generalist forecasters. So whether it's the car industry or the military or AI, if you ask people whether their field is going to lead to great world changing things, whether positive or negative, they tend to overestimate the probability of that by quite a large amount. And I've been talking to our friends at Good Judgment, which is the Superforecasters. I was talking to somebody there last week about this and he said they see this time and time again and experts are really quite offended when they're told that they're less good at forecasting. They say, you know, I know all about my field, I should be, should be good at this. But if you actually score their predictions, they're not as good at it as forecasters. And it's kind of not surprising because if you go into a field, then of course you think it's really important and world changing, whether positively or negatively. So you tend to overestimate it. So I think, although a lot of people are saying we have to listen to Geoffrey Hinton. He must be right because he's an expert and a founder of the field. That actually doesn't mean that he's necessarily more right. I'm not as gloomy as the most gloomy people like Geoff Hinton seem to be.

So let's call the risks unquantifiable. For the moment, they're non zero.

Right. What's to be done?

We've had discussions around the notion of a moratorium on this research, how to.

Deal with the risks that are there.

The risks right now are actually completely different things about the risk of bias and discrimination from the use of these systems. There are sort of intellectual property issues being raised by them. There are privacy concerns. And then I think there is this question of misinformation and what can be done about that. And the risk of producing large amounts of misinformation with these chatbots, which does seem to be already happening. Apparently there are lots and lots of news sites out there already full of nonsense written by chatbots. I don't think that's scary on its own. We've had essentially content farms producing bad content for a very long time. I think it's when this sort of goes over into politics and influencing elections and things like that.

This just in, we can now call.

The 2024 presidential race for Joe Biden. My fellow American, we've seen this advert, the RNC attack on Joe Biden, which they've generated with AI, but it's very obviously been generated with AI, and they're making a point of it. They're imagining this future shuttered their doors. Border agents were overrun by a surge.

Of 80,000 illegals yesterday evening.

I think what people are more concerned about is the use of this technology, where you don't admit that it's AI who's in charge here. It feels like the train is coming off the tracks. If you were Vladimir Putin, you generated video of Ukrainians committing war crimes, for example, then you could say, look at the terrible things the Ukrainians are doing to Russians. I mean, in fact, the Russians do already produce fake videos using editing, so it wouldn't be surprising if they started using AI as well. So that's the kind of thing that people worry about.

But what Mr. Hinton and others clearly worry about is if we get into a difference in kind here in terms of an artificial intelligence that is simply more actually intelligent.

Yes. And I think that's a risk that we're still some way away from. And I think Mr. Hinton would say the same Although he thinks it's getting closer very, very quickly. He used to say, he says that he thought that kind of superhuman AI might be 30 to 50 years away, and he now thinks it's a lot closer, maybe just a few years. So I think that's still a bit of a leap from the sorts of things that these models could do now. They are very smart in some ways, but they're also very dumb in a lot of ways as well. And people are trying to write Amazon reviews using these things and not doing a very good job of it. And I don't think they are that much of a cause for concern at this stage.

All of this is a tempest in a teacup then.

So I'm not saying that there's nothing to worry about here as well. But I think there are short term risks that fall short of the end of humanity, but that we still need to take seriously and that we probably, we do need some new regulations around. And so in the short term, I think what regulators should be doing is not saying, hey, let's stop all AI research. Instead, I think we need to be looking at rules around privacy and disclosure of training sets and more transparency and accountability when companies use these systems. So I think there are a lot of near term things that we could be doing both in research and in regulation to deal with the real problems that exist now. They're just not on the scale of the apocalyptic scenarios that are sometimes talked about where this is an existential threat to humanity. I think responding to that point, possible but distant threat is, you know, we risk taking our eye off the ball, which is the more immediate things we should be focusing on.

Tom, thanks very much for joining us. And if we wanted to hear more, where to find out more, well, we.

Did a whole episode of Babbage, our science and technology podcast, on this last week. So if you want to dig into this deeper, go and check that out.

A recently leaked American intelligence document shows just how difficult it's been to keep Ukraine's jets in the air. Since the start of Russia's invasion, Ukraine's air force has lost 60 fighter aircraft. It only has 80 or so left. So far, Russia has been unable to make much of the advantage it's had from the start. But without more jets, Ukraine doesn't stand a chance of dominating the skies.

I was in Kyiv all of last week and it was a really sobering visit.

Shashank Joshi is the Economist's defense editor.

On Friday 28th, early morning around 4am, I was woken up by air raid sirens. And I heard the thump, thump of the outgoing air defence rockets shooting down Russian cruise missiles and drones. It turned out to be one of the biggest Russian attacks on the capital city in months. What it shows is that the air war is still very real, it's very live, and that this is a subject of great importance for Ukraine's air force.

And what is the state of Ukraine's air defenses at the moment? What does it need?

Ukraine's air defences have been working overtime since October. They have been batting away huge numbers of cruise missiles, drones, even ballistic missiles which travel very fast. The problem is that they are running out. Many of these are Soviet era systems like the Buk or the S300, and it's not easy to find ammunition for these in Europe or in the United States. So one way to help that situation is to keep supplying Ukraine with new western air defenses. But there aren't a huge number of those. And so the question now is, how do you recapitalise Ukraine's air force? How do you replace the current fleet of old, aging, quite limited Soviet designed aircraft and find something new?

But what form might the something new take?

Well, there's a few candidates. In fact, a couple of weeks ago in Sweden, I flew one of the candidates. I flew the Saab Gripen, which is a single engined compact fighter jet. It was designed specifically to help defend Swedish airspace from Russian jets, rather than kind of fancy tasks like penetrating deep behind enemy lines for strike missions. And, you know, above all, it was built to be very rugged and versatile. It can land on very short runways. In fact, we did a very short landing. My pilot thudded the plane down with incredible force. And it can even land on roads in circumstances like Ukraine's, where traditional air bases have been struck by missiles. Saab says that the Gripen can be refueled and rearmed on the ground in about 10 minutes by a single technician, a sergeant and five conscripts. So that's extremely useful for the kind of operations that Ukraine is doing where it faces the constant threat of attack on its fixed air bases.

So where do they stand then, on.

That question of time, when it comes.

To the Gripen, could Ukraine get the.

Planes that it needs quickly enough?

The big problem, Jason, is that there aren't that many of them. The Gripen was really beaten out in European competitions by American planes, including the F16, but also the F35 in many cases. And so Sweden actually has fewer than 100 Gripens itself. There's also a geopolitical obstacle here. Sweden's bid for NATO membership is being blocked by Turkey and Hungary, so the country is particularly hesitant to deplete its defences at this time. I spoke to Sweden's Foreign Minister, Tobias Bilstrom, and he told me, look, nothing's off the table. We have to think about our own security needs. And we think that it would be easier for Ukraine to get Soviet designed MiG 29s that are in Eastern European stockpiles before they start thinking about high tech planes like the Gripen.

So in a lot of ways, as you say, the Gripen is ideal, except that it's not very available.

What are the other options?

Well, the MiG 29s are a short term stopgap, but they're actually not in great condition. I've been told many of them have been used by Ukraine for spare parts and actually they don't really give Ukraine any capability beyond the jets it has right now, which are outmatched by Russian planes, both in terms of the range of the radar and the munitions that they carry. So what Ukraine really wants is the American designed F16. The F16 is the canonical warplane of the last 25 years. One former pilot said. For him it was the Toyota Hilux of the combat air world. You know, sort of ubiquitous, robust everywhere you look. And over 4,600 F16s have been built since production began in the 1970s. Some of them are still being built. The newest ones, they make up around 30% of all the air force fleets among European NATO members, which is the most out of any plane. And there's also some second hand ones around. Last year Norway retired its entire fleet of F16s in favour of the F35. And Denmark, Belgium, the Netherlands, all these countries plan on doing so. Ukraine is so keen on the F16 that they've even published a little tongue in cheek video singing its praises with a 1980s theme. Because of course the bulk of these planes were produced in the 1980s, many in the 90s. And in their fantastic, inimitable style of information operations and propaganda, the Ukrainians have, I think, done a wonderful job selling this.

So isn't the F16 actually just as ideal as the Gripen is for the Ukrainian air force?

The F16's a great plane, but there's a couple of things to consider. One of them is that it's not really well suited to the kind of runways that Ukraine has. There's a lot of debris on Ukrainian runways. The F16 has very low air inlets and it's not a very rugged plane, unlike the Gripen, Unlike even the F18, the plane that you might recognise from the latest Top Gun movie, which are built for landing on carriers. In that case, the F16, many people say, would struggle on these runways and they might have very low rates of availability. The other question is we should not just obsess over the plane. What matters just as much is the air to air missile that it would carry, because that's the missile that would have to take down not just Russian missiles, but also Russian jets. The problem here is that the air to air missiles that are the most capable ones are really, really advanced. There's one called the Meteor, which is a fantastic European designed missile and it would very handily outmatch most of the Russian ones. But. But Europeans may not want to risk having the Meteor fall into Russian hands. And actually, in a similar way, the Americans might be very wary of sending the latest variants of the AIM120, which is the latest of their own air to air missiles.

And how does this discussion about fighter planes fit into the wider picture, the degree to which Ukraine is dependent on.

Western arms in all their forms?

Well, I think there are concerns about red lines and escalation and how Russia might respond. But it is very important to remember that cast your mind back a year ago and the idea of sending tanks to Ukraine was completely taboo. Last summer, Emmanuel Macron said tanks and jets were a red line. NATO wouldn't cross it and they crossed it. I think that Ukraine is going to get Western planes. The only question is how long it takes. It's not going to have them in time for the offensive that's coming up. But if Western countries act now and they realise the seriousness of the problems that Ukraine is facing with air defence, and they realise this war is not going to end anytime soon, they could conceivably get F16s or indeed other planes in Ukrainian hands, piloted by Ukrainian pilots by the end of this year. But they have to act now. There's no point waiting and dithering as we did with tanks, only to send them once more. Damage has been done to Ukraine's cities and more damage has been done to Ukraine's military prospects.

Thanks very much for joining us, Shashank.

Thank you very much, Jason.

The horrors of the Holocaust have inspired works of art, theater, music, film and television. Now you can add to the list video games.

There's a new game called the Light in the Darkness, which breaks new artistic grounds by tackling the Holocaust head on.

Our correspondent Colin Campbell has been playing the game and speaking to the man who made it.

It's Set in Paris during the Second World War, and it follows a fictional family as their world is torn apart. Their freedoms are stripped away. Their authorities force them to wear a star. They have to give up their tailor shop. And throughout the game, what starts off as a seemingly distant threat becomes ever and ever closer. As the Germans invade, they find that the hostility from their neighbors and from the local police increases. French police arrested 13,000 foreign Jews in July of 1942, including more than 4,000 children. They were shipped off to Velodrome Diva, a stadium in the capital, and later they were deported to death camps. It's a dark chapter in French history. The state did not admit its complicity with the Nazis until 1995. Luc Bernard, an independent game developer, says he has wanted to make this game all his life. He created and entirely funded the game, which he is giving away for free.

The characters in the Light and Darkness are actually based on multiple real people which died in the Holocaust. So in order to tell the story which we wanted to tell, I took multiple testimonies and also multiple stories of victims. The Light in the Darkness plays like an interactive film, meaning there's a predetermined storyline which the player goes through and different scenes. Only unlike a film where you're passive and just watching what happens in the Light in the Darkness, you're actually really within the scenes since you play as multiple characters.

Nazis appear as villains in video games all the time, very often in historical situations. It's a power fantasy for players to be able to mow down these villains, but players are never required to encounter or deal with what the Nazis stood for or what they wanted. Around 100,000 players have already tried the game, and Mr. Barnard thinks that it will reach around 2 million users in the next year. He hopes that teachers and educators will use the game to reach younger people who perhaps haven't experienced stories about the Holocaust in other formats.

When I first started the game, it was because the young people mostly get everything from video games because we're the biggest industry. So it started out as like a Holocaust memory project. But as I was developing it over the past two years because of the rise of antisemitism, to me, it also turned into a project which became important for the present. So I could show, you know, what antisemitism can lead to and how ugly it is and how similar, even what we're seeing, some anti Semitic rhetoric today is similar to what was going on in Europe during World War II. I'm a big believer in making lot of things digital, so teachers can just access it, because not every teacher is an expert. And also as a video game, let's be honest too. If you tell teenagers you've got a choice between playing a video game, an educational one, or reading a book, they're going to go towards video game. And then it's going to make it easier for them to pay attention in class. But also, what's been interesting to Light and Darkness, that I receive a lot of emails, like from African refugees who basically told me they kind of see themselves in the characters and how they think it's important for people to know because of what they've been through about how they've also seen sometimes racism, fascism within that continent too, which, you know, it's not exclusive. Genocide is not exclusive to Europe.

It's a video game. So you get things to do, but the choices that you're offered often turn out to be illusory. The family talks about perhaps going on the run or trying to move to America, but none of those things are realistic. And in the end, everything that is put in front of you as a possibility turns out to be futile.

For the light in the darkness, there is choices in the game, but they don't really matter because to me, Jews during the Holocaust didn't really have a choice and it was just pure luck, you know, if you survived. So that's why with the ending of the game, all the main characters, they do die. And I had to show it that way to be more realistic. Because if you had all players managing to survive, people could get a wrong impression and think pretty much, oh, Jews could have done this, this could have happened, etc. In 2010, the Anti Defamation League said the Holocaust should be off limits for video games. It wasn't about my game, it was about another game. So there has been within antisemitism organizations that don't quite understand video games, right? Video games should never tackle this. Because of this. This has kind of created, how can I say, like nearly a small fear within the industry. We cannot touch this. But within organizations too, the younger people, that isn't pushed back at all. They're basically like, yeah, video games are why we should do this. There's even at the University of Sussex, Victoria Walden, she does loads of studies about this video games on the Holocaust. So she's been pushing for this for a while. So among the younger people, there is a consensus that this is kind of future of Holocaust awareness. But the problem is, again, most organizations, it's older people and people don't really know too much. Digital they don't. I mean, that's why antisemitism is so bad on social media, you know, because it was kind of left until it was too late.

The Game arrives at an urgent moment. The Anti Defamation League reports that antisemitism has reached a record high in America. Elsewhere, memories of Nazi horrors are waning as elderly survivors die. By keeping their stories alive, the Game helps to inform new generations about the atrocities that Jews were forced to endure.

That's all for this episode of the Intelligence if you're not a subscriber to the Economist, you really are missing out. But dive in with the deal we've got on at the Moment, a free 30 day digital subscription. Just go to economist.com intelligenceoffer or click the link in the Show Notes. We'll see you back here tomorrow.