- Large language models are actually special in their unreliability.

They're arguably the most versatile AI technique that's ever been developed, but they're also the least reliable AI technique that's ever gone mainstream.

[bright music] [logo whooshes] - Hello and welcome to "GZERO World."

I'm Ian Bremmer, and, today, we're talking about all things artificial intelligence, specifically generative AI, those chatbots like ChatGPT that you've surely heard about by now.

You know, the ones that can churn out a two-hour movie script or Picasso-style painting in just an instant.

With the recent rollout of OpenAI's ChatGPT-4, it's clear we're still only at the tip of the AI iceberg, and, no, a chatbot did not write this script.

The truth is I couldn't do this show without my beloved and irreplaceable team of producers and editors whose families rely on.

No.

Hey, this is not what I approved for the teleprompter.

[screen beeps] Anyway, my guest today has devoted his life to understanding artificial intelligence.

Cognitive scientist Gary Marcus, who refers to ChatGPT in our interview as, quote, "auto-complete on steroids," and he doesn't mean that in a good way.

Don't worry.

I've also got your "Puppet Regime."

- Welcome back, folks, to your favorite game show, "Who The Heck Was That World Leader?"

- But, first, a word from the folks who help us keep the lights on.

- [Announcer 1] Funding for "GZERO World" is provided by our lead sponsor, Prologis.

- [Announcer 2] Every day, all over the world, Prologis helps businesses of all sizes lower their carbon footprint and scale their supply chains with a portfolio of logistics and real estate and an end-to-end solutions platform addressing the critical initiatives of global logistics today.

Learn more at prologis.com.

- [Announcer 1] And by.

Cox Enterprises is proud to support GZERO.

We're working to improve lives in the areas of communications, automotive, clean tech, sustainable agriculture, and more.

Learn more at cox.career/news.

Additional funding provided by Jerre and Mary Joy Stead, Carnegie Corporation of New York, and... [upbeat music] [logo whooshes] [lively music] - Who runs the world?

Which head of state?

Which captain of industry will our alien enslavers greet first when they inevitably touch down?

Who, in other words, is the head of the global order?

The answer is no one.

No one.

Sorry, aliens, and that's because there is no longer a single global order to lead.

Sure, there used to be after the Soviet Union fell and the United States was on top of it, but then the United States got tired of being the world's policeman, the architect of global trade, the cheerleader for global values, for good and for bad.

So when China's economy started growing like gangbusters, the US was only too happy to let them integrate into US-led institutions, on the presumption that, as they got wealthier and more powerful, they would become Americans.

Turns out they're still Chinese, and the United States is not particularly comfortable with that, but if there isn't just one global order anymore, how many are there?

Actually, I'd say three.

First off, there's the global security order, and in that realm, at least, the United States and its allies are the most powerful players.

The US is the only country in the world that can send its soldiers and its sailors to every corner of that world.

No one else close.

Sure, China's making a go of it with its warships, coveting Taiwan's coastline like a jealous lover, and Russia's military had some swagger before it fell apart in Ukraine, but as long as nuclear war remains synonymous with suicide, the US, yes, the US stays alone at the top of the military flagpole, but, secondly, there's a global economic order, and, here, power is shared.

The United States, of course, is still a very robust global economy, but is unable to use its dominant position militarily to tell other countries what to do economically, and so when it comes to China, the two countries are so economically interdependent that neither Washington nor Beijing can get the upper hand.

The EU, by the way, the European Union, has the largest common market.

India and Japan are very much in the mix as well.

So plenty of economic jockeying.

No superpower.

Now, the third global order may not be quite here yet, but it is right around the corner, and I'm talking about the digital order, which is not run by governments, but rather by technology companies.

When I was growing up, it was nature and nurture that determined our identities, but now it's nature, nurture, and algorithm, and there is no pause button on these explosive, productive, and disruptive technologies.

Who will make sure the technology leaders act accountably as they release new and powerful artificial intelligence tools?

What are they going to do with this unprecedented amount of data that they are collecting on us and our environment?

And will they proceed with advertising models that turn citizens and products and drive hate and misinformation into our societies?

That's a lot to chew on, so let's start with a smaller bite, artificial intelligence.

Let's go smaller still, generative AI, and, here, I'm talking about AI-powered tools like the text-to-text generator ChatGPT or the text-to-image generator DALL-E. As of now, they can do magical things like write your college term paper for you in Klingon or instantly generate nine images of a slice of bread ascending into heaven.

A fun thing to do, but many of the smartest people I know think these tools will reshape the way we live in both good ways and bad, and that is the subject of my interview today with psychologist, cognitive scientist, and NYU professor emeritus Gary Marcus.

Here is our conversation.

Gary Marcus, thanks for joining us today.

- Thanks for having me in.

- So, AI genius, I have so many things that I wanna talk to you about today.

I wanna maybe start with the fact that we've had AI voice assistance on our phones for, like, about a decade now, but no one's really been all that excited about it.

Then, suddenly, now it's all about ChatGPT and all of this other stuff.

Is it really all that different, and how, if it is?

- Well, the underlying technology is actually pretty different, and your question is a reminder there's actually lots of kinds of AI, some of which we can trust and some of which we probably shouldn't.

Siri was very carefully engineered to do only a few things and do them really well and it has APIs to hook out into the world.

So, if you say, "Lock the door," and you have- - [Ian] API?

Explain API for everybody.

- Application Program Interface.

It's jargon for hooking your machine up to the world, and Siri only does a few things.

It'll control your lights if you have the right kind of light switches, it will lock your door if you have the right kind of door, but it doesn't just make stuff up.

It only really works on a few things.

It doesn't let you talk about just anything.

Every once in a while, they roll out an update.

Now you can ask it about sports scores or movie scores.

At first, you couldn't, but it's very narrowly engineered, whereas the large language models that are popular now are kind of like jack of all trades but masters of none.

They pretend to do everything, but if you tell them to move your money in your bank account, it might not have a way of actually connecting to your bank account and it might say, "Yes, I moved your money," and then you might be disappointed when it didn't actually do it.

So there's a kind of appearance or an illusion of great power that Siri never tried to give you.

Siri tried to not oversell what it could do, whereas, in a certain sense, the large language models are constantly overselling what they're doing.

They give this illusion of being able to do anything.

Like, they'll give you medical advice, but that doesn't mean they really understand medicine or that you should trust them, but it's completely changed the world that these things are widespread.

They existed before, the underlying technology, the main part of it.

Well, some of it goes back decades, and the main technical advance was in 2017.

People were playing around with them in the field, but nobody knew they would catch on this way, and now that you have these unreliable tools that give the illusion of incredible versatility and everybody's using them, that actually changes the fabric of society.

- I think it'd be helpful if you explain to people a little bit how large language models, what we commonly think of today as the AI we interact with, how it works.

- They are analyzing something, but what they're analyzing is the relationship between words, not the relationship between concepts or ideas or entities in the world, and so they're basically like auto-complete on steroids.

We give them billions or even trillions of words drawn from all the way across the internet.

Some people say they now are trained on a large fraction of the internet and they're just doing auto-complete.

They're saying, "If you say these words, what is the most likely thing that will come next?"

And that's surprisingly handy, but it's also unreliable.

A good example of this was one of these systems saying, "On March 18th of 2018, Tesla CEO Elon Musk died in a fatal car accident."

Well, we know that a system that could actually analyze the world wouldn't say this.

We have enormous data that Elon Musk is still alive.

He didn't die in 2018.

He tweets every day.

He's in the news every day.

So a system that could really do the analysis that most people imagine that these systems are doing would never make that mistake.

- Could not return that result.

Right.

- Could not return that result.

It could have looked in Wikipedia.

It could look in Twitter.

You know, if somebody tweeted, they're probably still alive.

There's so many inferences it could make.

The only inference it's really making is that these words go together in this big soup of words that it's been trained on, so it turns out that other people died in Teslas, and he's the CEO of Tesla, but it doesn't understand that the relationship between being CEO of Tesla is not the same as owning a particular Tesla that was in a fatal vehicle accident.

It just does not understand those relationships.

- So, Gary, why can't you combine these two things in one of these apps?

Why can't, in other words, you have this very, very powerful predictive analytics tool in assessing the relationships between words and data, and then after, instantaneously, right after it returns that, just do a quick search so you would know that, if it returns something that's stupid or obviously false, Google, Wikipedia, whatever, doesn't then return the thing that's obviously false?

Why isn't that possible?

Why isn't that happening?

- People are trying, but the reality is it's kinda like apples and oranges.

You know, they both look like fruit, but they're really different things, and in order to do this, quote, "quick search," what you really need to do is to analyze the output of the large language model, basically take sentences and translate them into logic, and then if you could translate them into logic, but that's a really hard problem that people have been struggling with for 75 years, then you could do that logic if you had the right databases to tell you all this stuff and you'd hook them all up.

That's what AI has been trying to do for 75 years.

We don't really know how to do it.

It's like we have this shortcut and the shortcut works some of the time, but people are imagining that the shortcut is the answer to AI.

These things have nothing really to do with the central things that people have been trying to do in AI for 75 years, which is to take things, language, translate them into logical form, into database facts that can be verified, that can be analyzed.

We still don't really know how to do that, and it's much harder than it looks, which is why you have things like Bing, you know, give you references, and the references will say the opposite of what actually happened.

It can't actually read those things.

Another way to put it all together is these things are actually illiterate.

We don't know how to build an AI system that can actually read with high comprehension, really understand the things that are being discussed.

- So, Gary, I get that they're different things.

What I'm trying to understand is, I understand that the AI large language model is not capable of recognizing the output as information and translating it, but Google as a search engine is.

So, again, what I'm trying to understand is, when you have that outcome, why can't you then put that essentially into a separate search engine that you and I aren't gonna see in the interactivity, but it's actually doing those two different things?

- When you do that with Google, you're actually human in the loop.

Google doesn't actually do all this stuff by itself.

It requires you to do it.

So you put in a search, it gives you a bunch of garbage, and you sort through that garbage, and so in a typical application of Google, it's actually human in the loop is how we would describe it.

What people are looking for is something more autonomous than that, without humans in the loop.

You wanna be able to type into ChatGPT whatever, have it give you a search, and figure out the answer for itself, not consult you or some farm of humans off in some other country who are underpaid and too slow to actually do this right now.

You want a system that can actually do it autonomously for itself.

Google's not actually good enough to do that, right?

It's really just matching a bunch of keywords.

It gives you a bunch of stuff and you as a human sort it out.

So it's a nice idea to say, "Just pass it through Google," and that's kinda what they're actually doing with Bard, and it kinda doesn't work.

- So, if we take that challenge, how could we potentially create AI that is really, quote-unquote, "truthful," that's going to be factual to the extent that you and I would believe it?

The equivalent of we're prepared to have the autonomous driver as opposed to drive ourselves, not because it never gets into accidents, but because, you know what, it's good enough.

We have confidence.

How close are we?

What needs to happen before in a chatbot or another LLM system will be something that you and I can have confidence in?

- I think we're fairly far, but the main thing is I would think of it in terms of climbing mountains in the Himalayas.

You're at one peak and you see that there's this other peak that's higher than you, and the only way you're gonna get there is if you climb back down and that's emotionally painful.

You think you've gotten yourself to the top.

You haven't really.

You realize you're gonna have to go all the way back down and then all the way up another mountain, and nobody wants to do that, and add in the economics where you're making money where you are right now.

You're gonna have to give up the money that you're making now in order to make a long-term commitment to doing something that feels foreign and different.

Nobody really wants to do that.

I think there's a chance maybe that, finally, we have the right economic incentive, which is people want what I call chat search to work, which is you type in a search to ChatGPT.

It doesn't work that well right now, but everybody can see how useful and valuable that would be.

Maybe that will put enough money and enough kind of frustration because it's not working to get people to kinda turn the boat.

I think once people start, first of all, taking seriously old-fashioned AI.

Sometimes people call it good old-fashioned AI.

It's totally out of favor right now, but it was, I think, dismissed prematurely.

Good old-fashioned AI looks like computer programming or logic or mathematics.

You have symbols that stand for things and you manipulate those symbols like you would in an equation.

We need to combine elements of old-fashioned AI, symbolic AI, with neural networks, and we don't really know how to do it.

So, old-fashioned AI is much better, as it turns out, with truth.

If you give it a limited set of facts, it can reason relative to those facts and it won't hallucinate.

It won't just make stuff up the way neural networks do, and so we would like to be able to use something like that, but it doesn't learn as quickly.

It's more cumbersome to work with, and so people have abandoned it, really, for 30 years.

We're gonna need to come back and say, "Look, we made a lot of progress.

We're proud of ourselves."

But there were some ideas that people had in the 1960s and '70s that were actually pretty good.

We have to stop being so hostile to one another in this field.

- So I mean, I take it, and I want you to confirm this, that what you're saying is that we're gonna get to ChatGPT-5 and 6, which will be vastly faster and we'll be much more confident in our interactions with it if we don't question it, but does that mean that you don't believe that the hallucinations are going to get materially better?

- I wrote an essay called "What to Expect When You're Expecting GPT-4."

I made seven predictions and they were all right, and one of them was that GPT-4 would continue to hallucinate, and I will go on record now as saying GPT-5 will, unless it involves some radical new machinery.

If it's just a bigger version trained on more data, it will continue to hallucinate, and same with GPT-6.

- What are the advances that you think we will see?

How is this going to be more useful for individuals, for science, for society?

I mean, pick your area about where you think we are.

Given what the money is presently being spent on, the exponential growth and capacity that we've experienced in the last couple of years, play it out two, three, five years.

Where do you think the biggest advances are gonna be?

- So, the first thing I'll say is there's no guarantee that all this money in is gonna lead to an output.

It might, there'll probably be some output, but I'll just say, as a cautionary tale, remind you that driverless cars have been around for a long time.

In 2016, I said, you know, even though these things look good right now, I'm not sure they're gonna be commercialized any time soon because there's an outlier problem, and the outlier problem is there's always some scenarios you haven't seen before, and the driverless cars continue to be plagued by this stuff seven years later.

I gave an example then that Google had just maybe solved at that point a problem about recognizing piles of leaves on the road, and I said there's gonna be a huge number of these problems.

We're never gonna solve them, and we still haven't.

So there was a Tesla that ran into a jet not that long ago because a jet wasn't in its training set.

It didn't know what to do with it.

It hasn't learned the abstract idea that driving involves not running into large objects, and so if there's a particular large object that isn't in its training set, it doesn't know what to do.

So what happened there is $100 billion went into this, and, still, the driverless car industry so far has just sucked in money and not actually made driverless cars that you can use in a reliable way.

We may see that in large language models, or we may see that they can be used in limited circumstances, but not universally.

So, programmers are good, but take medicine.

It's not clear that even GPT-5 is gonna be a reliable source of medicine.

4 is actually better than 3, but is it reliable enough is an interesting question.

Each year's driverless car is better than the last, but is it reliable enough?

You know, not so far, and so whether you'll make that threshold in a domain where it's really safety-critical is unclear.

So there's definitely gonna be things we get out of it, but there's also risks.

- What I am hearing pretty consistently, and I think it certainly aligns with what you're telling me today, is that humans in the loop remain essential for almost all of the uses of AI that we're really benefiting from.

- Yeah.

That's true for large language models.

You could argue that the routing systems, routing systems that you use in your GPS, that just works.

You don't need a human in the loop there.

Same with a chess computer.

You can just play the chess computer.

You don't need a human there.

Large language models are actually special in their unreliability.

They're arguably the most versatile AI technique that's ever been developed, but they're also the least reliable AI technique that's ever gone mainstream, and so everything where we're using large language models, we do, for now, need humans in the loop.

- Before we close, I want to ask you just for a moment about what do we do about this side, which is the governance side, right?

We don't really have a regulatory environment yet.

The government actors don't know a lot about the stuff that you know a lot about yet, maybe ever.

You know, the architecture is not there.

The institutions aren't there.

Give me, just for a moment, where you think the beginnings of effective governance or regulation would come from in this environment.

- The first thing is I think every nation has to have its own AI agency or cabinet-level position, something like that, in recognition of how fast things are moving and in recognition of the fact that you can't just do this with your left hand.

You can't just say, "All the existing agencies, yeah, you can just do a little bit more and handle AI."

There's so much going on.

Somebody's job needs to be to look at all of the moving parts and say, "What are we doing well?

What are we not doing well?

What are the risks to cyber crime, misinformation?

How are we handling these kinds of things?"

So we need some centralization there.

Not to eliminate existing agencies, which still have a big role to play, but to coordinate them and figure out what to do.

We also, I think, need global AI governance.

We don't really wanna have, the companies don't really wanna have different systems in every single country.

So, for example, it's very expensive to train these models.

If you have 193 countries with 193 different regimes, requiring so much damage to our climate and maybe updating them every month or whatever, that would be just a climate disaster, so we want some international coordination.

And then another thing that I think is important for each nation, and maybe we do this globally, is I think we need to move to something like an FDA model, where if you're gonna do something that you deploy at wide scale, you have to make a safety case.

So, sure, you can do research in your own labs.

Google doesn't have to tell us everything they're doing.

OpenAI doesn't have to tell us everything they're doing, but if they're gonna put something out for a hundred million users, we really wanna make sure it's safe and ask, "What are the risks here?

What are you doing about those risks?"

Right now, the companies are doing those things internally, but we need to have external scientists who can say, "Hey, wait a minute."

I'll give you an example.

There's something called ChatGPT plugins, which has now given rise to something called AutoGPT, which can access your files in the internet and even other human beings.

Any external cybersecurity expert would say, "There's a lot of risk here."

But the companies, OpenAI went ahead and said, "It's fine.

We can do this."

Whereas Apple says, you know, "We have to sandbox every application.

We have to limit what its access is."

So OpenAI has done something completely at odds with best practice that we know elsewhere in the software industry, and there's no constraint on that and could actually lead to pretty serious harm, and so there are cases like these where we really need some external advisory that can say, "This isn't good enough."

Sort of like peer review.

You don't just publish a paper.

You have people examine it.

We need to do the same thing.

You can't just put something out, and if you affect a hundred million users, you really affect everybody.

So just as one example, these systems are gonna affect people's political opinions, so, everybody, even if they signed up or not, is gonna be affected by what these systems do.

We have no transparency.

We don't know what data they're trained on.

We don't know how they're gonna influence the political process, and so that affects everybody, and so we should have some oversight of that.

- Gary Marcus, thanks so much for joining us.

- Thanks a lot for having me.

[gentle futuristic music] - Now it's time for something a little different.

I've got your "Puppet Regime."

- Welcome back, folks, to your favorite game show, "Who The Heck Was That World Leader?"

Given the high turnover of world leaders recently, this should be a competitive one.

Our contestants today are, who else?

The US and China.

[audience applauds] All right.

Let's get started.

[mysterious music] This former president has been accused of inciting post-election violence, the evangelicals adore him, and he absolutely loves Florida.

[button beeps] - This is easy as pie.

Donald Trump.

[buzzer buzzes] - No, sir.

The answer is former president of Brazil, Jair Bolsonaro!

- Joe, you would've gotten that if you paid more attention to Latin America.

- Well, maybe you oughta pay a little less attention!

- Whoa, okay, simmer down.

Save it for the semiconductors.

Up next, this former world leader had a grandfather from Germany, was weirdly friendly to Vladimir Putin, and had a very controversial approach to immigration.

[button beeps] - This one is Trump.

[buzzer buzzes] - No, it's not Trump.

It's Angela Merkel.

- Ugh.

Putin clearly was the one that got away.

- All right, up next, this former world leader has been married three times, was accused of mishandling COVID, and once called Hillary Clinton a sadistic nurse.

[button beeps] - Oh, that's Trump for sure.

That sounds just like him.

He's gotta be.

- Definitely Trump this time.

- Man, that guy really does live rent-free in your heads.

What are you, CNN?

Hah!

[audience laughs] The answer is Boris Johnson!

- I was told there'd be free drinks here.

- Okay, last one.

This former world leader launched two disastrous wars, he comes from an oil country, and he'll never face war crimes charges.

- Crikey.

I think he's talking about my friend George W. - Not a bad guess, but, no, it's Vladimir Putin!

- Well, that's malarkey.

He's still in power.

- Please, Joe, I am in Kremlin so long, I am technically current and former president of Russia.

[rimshot plays] [audience laughs] Tip your servers, folks.

I'm here 'til 2036.

- [Chorus] "Puppet Regime"!

- That's our show this week.

Come back next week, and if you like what you see, or even if you don't, but you feel a little bit artificially intelligent yourself, why don't you check us out at gzeromedia.com?

[lively music] [lively music continues] [lively music continues] [bright music] - [Announcer 1] Funding for "GZERO World" is provided by our lead sponsor, Prologis.

- [Announcer 2] Every day, all over the world, Prologis helps businesses of all sizes lower their carbon footprint and scale their supply chains with a portfolio of logistics and real estate and an end-to-end solutions platform, addressing the critical initiatives of global logistics today.

Learn more at prologis.com.

- [Announcer 1] And by.

Cox Enterprises is proud to support GZERO.

We're working to improve lives in the areas of communications, automotive, clean tech, sustainable agriculture, and more.

Learn more at cox.career/news.

Additional funding provided by Jerre and Mary Joy Stead, Carnegie Corporation of New York, and... [upbeat music] [bright music]