Schrödinger’s chat: can AI and scientists co-exist?

0:00With the Spark Cash Plus card from Capital One, you earn unlimited 2 % cash back on every purchase and you get big purchasing power so your business can spend more and earn more. Steven, Brandon and Bruno, the business owners of SandCloud, reinvested their 2 % cash back to help build the company's retail presence. Capital One, what's in your wallet? Find out more at CapitalOne.com slash Spark Cash Plus. Terms apply. This intelligence podcast is sponsored by Perk, the intelligent platform for travel and spend. Every person on your team loses up to seven hours a week doing the work behind the work, submitting expenses, matching receipts and booking travel.

0:40Using AI and automation, Perk can do that work, bringing travel and spend together in one place. Enjoy three months without platform fees when you sign up at Perk.com forward slash the intelligence. That's Perk.com forward slash the intelligence. New US-based corporate customers only. Terms apply.

1:06Rosie Blau: Hello and welcome to The Intelligence from The Economist. I'm Rosie Braw. And I'm Jason Palmer.

1:16Rosie Blau: Today on the show, why snacks are getting crunchier, and a warning that this item includes some disgusting sounds. and celebrating the life of writer Annie Dillard.

1:31Rosie Blau: First up, though.

1:41Usually, when you get handed the answer to a problem, you're happy. Unless, of course, you're a mathematician. The firm OpenAI used an unreleased model to tackle a bunch of unanswered mathematics problems and then last week published hundreds and hundreds of solutions. This, as we heard about a month ago as well, is the kind of stuff that makes mathematicians just snap their pencils. But it's more than that. It seems like the beginning of a far wider story. About what happens when massively improving AI crashes into the scientific enterprise as a whole.

2:22Who better to look at that than Alok Jha, our science editor and host of our sister show Babbage, and Alex Hearn, our AI writer. Alex, Alok, hello. Hello. Hi, Jason. Alok, I'd like to start with you just to kind of give us a snapshot of where things stand today, how good AI is at various kinds of science already, and maybe how good it could get.

2:43Alok Jha: Well, Jason, how long have you got? I mean, the applications are many. In biology, all sorts of things are possible. I mean, AI models can read DNA in the way that the chatbots, you and I use, read text, so they can predict genomes, they can build genomes from scratch, but also just understand how genetics works. We'll be able to design proteins using specialized models in academic labs. You can design better drugs. There's also things like controlling how the plasma of a fusion experiment works, so potentially bringing that power closer. One very practical application of AI in science has been to make weather forecasting much, much, much better.

3:24Alex Hern: And of course, it's also been advancing the speed of AI science. AI scientists are what underpins this whole fear of recursive self-improvement, the idea that AI will get better at making AI ad infinitum until it takes over everything.

3:39Alok Jha: And it's happened quite quickly because there's just lots of good data in science. And so you can apply these models to them.

3:46Alex Hern: One of the things that's quite dizzying about the speed of progress over the last few months is that the approach that is behind a lot of these improvements is barely a year old. The frontier of AI science right now relies on marrying two AI techniques, reinforcement learning, which is what won Demis Hassabis' Nobel Prize. That's how AlphaFold, this protein database, was born. And then you have large language models. And then what you've had recently is the merger of both of them, training large language models to get better and better through the same reinforcement learning approach. And that seems to have been the key to unlock massive gains in areas like maths, computer science, and potentially theoretical physics and other numbery heavy domains.

4:33Yeah, I mean, it sounds like the applications are pretty much all of the boundaries of science, but let's focus on the mathematics, what's happened and what mathematicians are saying about it.

4:44Alex Hern: What happened has been called the mathpocalypse by computer scientist Scott Aronson. Anyway, I am drowning in pocalypses.

4:52Alok Jha: Or mathocalypse, if you read his word. We call it the mathpocalypse. I'm not sure if it's an American English thing. The mathpocalypse

4:59Alex Hern: was this dump of hundreds of proofs across hundreds of papers to a huge array of unsolved problems in mathematics. And it freaked out mathematicians, probably more than OpenAI's previous big success solving one of the millennium problems, one of the hardest unsolved problems in maths, because the sheer quantity of this showed not only that AI was capable of doing work that no human had previously been able to do, but that it could do that in a blink of an eye. And they demonstrated that this is a change that is going to scale. And the thing about maths is there's a solid chance that it's just the bellwether here.

5:38Alex Hern: Lots and lots of other areas of science have the same fundamental principles and appear amenable to the same AI-powered speed up. But a cynical view on why the mathematicians are freaking out is simply that they recognise they're now kind of out of a job.

5:52Alok Jha: Yeah, it's a familiar sounding complaint, isn't it? But also, you've got to understand why mathematicians are so annoyed. It's not about the answers. It's about the journey to the answers. Mathematics is this subject where you're sort of exploring a landscape or climbing a mountain, and the journey up the mountain is as important as getting to the peak. The OpenAI release was like someone taking a helicopter and just landing you on the mountaintop.

6:16Alex Hern: One of the other things that seems particularly unsettling to a lot of mathematicians I've heard from, is that as the AI reaches superhuman ability, its answers stop resembling those of human mathematicians. The results that OpenAI has put out in this mathpocalypse appear overwhelmingly focused on unelegant solutions and counter examples rather than proofs, right? So where a mathematician will often talk about finding, the harmony inherent in maths, a lot of what OpenAI has found are places where the actual answer is ugly as sin. It's not elegant. It's not beautiful.

6:54Alok Jha: A lot of the papers published in the Mathpocalypse, obviously mathematicians are trying to read them now to understand what's going on inside them. And they're struggling. A lot of it doesn't make any sense. Well, wait a minute. Perhaps the question here is, are these proofs good proofs? Are they true? Is it real? I mean, aren't hallucinations an issue in this as in everything else? Well, so in mathematics, what you can do is, as well as having your paper that publishes the proof that you've got, you can formalise it. And what that means is you get a certificate and it means that it's true. It's a particular subject, mathematics, in which way you can do that.

7:27Alok Jha: It's built around this verification.

7:28Alex Hern: I don't think anyone is seriously alleging that anything other than perhaps a small handful of these might be misformalized. That formalization, that verification, is inherent to why maths is the domino to fall here as well. The reason why we're seeing so much progress in maths and comparatively little progress in, let's say, cellular biology, is because it doesn't have access to oodles of cells.

7:53Alok Jha: Not yet it doesn't anyway.

7:54Alex Hern: Yeah, exactly. That requires a wet lab and physical space. Robotic scientists, these are all things that people are experimenting with, and everyone wants to be the AI lab to have the first sort of end-to-end physical science pipeline. So explain the earlier suggestion then that this is the first domino to fall. If mathematics is such a particular case, one that is so amenable to this kind of computation, why should the rest of science be quaking in its boots?

8:21Alok Jha: Well, a lot of science uses mathematics, right? So physics especially, computational biology, all the sciences that have lots of information in them. Math is the sort of root of all of that. So there is a way in for AI to sort of tackle all of those problems too. And in many respects, that's quite useful. I mean, if you can really solve lots of mathematical problems, you'll be able to understand lots more science. I mean, theoretical physicists have wondered which string theory is right for a long time. And that's essentially, in large part, a mathematical problem. There's a world you could imagine now that scientific output first comes from the machines and then humans scramble around to try to understand it and derive some value for it.

9:01That is like completely turning upside down the way science is done. But maybe we get to the same to really work the analogy peaks.

9:09Alok Jha: Yeah, I think that just because science has been done in the way it's been done so far, doesn't mean that that's the way that it's going to have to be done forever. So yes, it does sound like a slightly less glamorous thing to just be checking results and understanding rather than discovering. But for humanity, you're still getting, hopefully, the same good outputs. I mean, that's perhaps just the way that things have to change. But that kind of rethinking, being so different from how things are done, really has long-range effects, right? That is how every academic university's science department works and what the incentives look like.

9:45That is a lot to unpick.

9:46Alex Hern: Yeah. And I mean, you can start to see some of the shapes of this thinking, even in the work coming out from the big labs, right? Google put out a paper in mid-October looking at how you would have to reshape sort of the entire scientific economy to deal with the world of AI-focused science. What you want is you want the smartest scientists in the world to be able to direct the smartest machines in the world and then get the credit for having done so. And that's a really thorny problem, not really of science, right, but of, in classic economist style, of market design and allocation of scarce resources.

10:21Alok Jha: In this AI-driven science economy, the discoveries will just come thick and fast. The value will be in the checking, the value and the QDOS will move to which one of these discoveries actually is useful to us. Look, I want a more prominent role for humans in all this as this happens. And I'm thinking about, for example, Gen. AI writing and how it can do very plausible writing, but it's not going to come up with the idea for and beautifully execute the great American novel. Is there an analogy with science in the sense that there's still room for the sort of spark of creativity, the moment that Einstein had the critical thought about relativity, that kind of thing, if the computers are really just crunching through what they've been told and just being very good at numbers?

11:07Alex Hern: One optimistic take is no, that the machines are not held back by the speed of human input, and that we do get proper unrelenting acceleration in the capacity of Earth, maybe not humankind, but at least Earth to discover and improve our lot. I think it would be bad if the pace at which we can develop new cures for cancer is held back by the need for humans to come up with every single part of the research development pipeline for that. I would like some of that to be automated faster, thank you. It would be really nice. And I think that's an optimistic vision of the future.

11:51Alok Jha: So some of it though, Alex, not all of it. I think that as soon as you let machines do everything end to end, it might be attractive. But then the ability of an LLM to be able to do that means it could probably do all sorts of other things without you asking or being involved to the point where we're just very used to LLMs solving all sorts of problems that we haven't even thought of. I think that's dangerous.

12:13Alex Hern: Everything comes back to AI safety in the end. The world that I think we should be preparing for as a possibility is one where the pace of discovery is just persistently outpaces the pace of understanding. That doesn't seem implausible to me. And I think there is a somewhat difficult assumption underlying it all, right? Which is that all of maths, science and reality is comprehensible to humans. It may be that there are some problems that humans will never be able to understand, will never be able to get their heads around. And are we really saying we do not want solutions that are based on something else understanding that.

12:49Alok Jha: We're going to get to a point where we're Alexander the Great as a species. You know, we look upon the world and there's a tear falling down the eyes because there are no more worlds left to conquer. We solved everything. There is a lot to chew through there, gentlemen, but I know that there's a lot more to come in particular on Babbage, our sister science show. Yeah, we're going to have several episodes, starting off with the mathpocalypse, but also going into exactly where AI will be useful in biology and physics and all sorts of other places and what this all means. So yeah, look out for that.

13:21Well, for now, Alex, Alok, thank you both very much for your time and for making my head spin.

13:26Alok Jha: Thanks, Jason. You're welcome. Thank you for having us.

13:45Rosie Blau: Before we go on, I want to tell you about this week's episode of The Economist Insider, our video series that takes you behind the scenes of our newsroom. The right-wing populist party, Alternative for Germany, has been riding high under the leadership of Alice Weidel. It's had successes in a series of state elections, as many Germans turn away from mainstream parties. In this week's Insider, The Economist's editor-in-chief, Zannie Minton-Beddos, sat down with Alice Weidel at the Bundestag in Berlin to ask how she would govern if she became chancellor. Weidel has called for remigration, as she terms it, and criticises Germany's support for Ukraine.

14:23Rosie Blau: So, as you can imagine, that made for a tough conversation. If you're a digital subscriber, Insider is already included in your subscription. And if you're listening free, you can watch extended clips via the link I've put in the show notes.

14:45Okay, real talk. Every time I open up my inbox, I've got this vision of the day I'm going to have. Deep work. Big ideas. The project I keep meaning to start. Three hours vanish into replying to threads and hunting down a file someone sent last week. I've tried AI tools that promised to fix this. They didn't. They were just another tab I had to open, explain myself to, and abandon by Thursday. Superhuman is different. It's at your service. not after your job. It's already where you work, so I'm never starting from a blank tab. It's from the makers of Grammarly, living right inside of the tools I already use, like Gmail and Slack.

15:23When I'm drafting a message or making sense of a messy thread, it just helps, like a thought partner that's already caught up, so we get to the hard part together. The problem was never too little time. It's that busy work crowds out the work that actually matters. Start streamlining your workflows with Superhuman. Try today at superhuman.com. This episode of The Intelligence is sponsored by VantageScore. For years, credit scores have been generated with just a snapshot, and that snapshot left millions of people out of the picture. So VantageScore is modernizing the whole process. They use 400 % more credit report and alternative data than legacy models, bringing more people into the credit economy, like 5 million more mortgage-eligible people.

16:09So all the banks win, but so do all the francs, the lenders and the borrowers, the Lindas and the Olivers. When you build credit scores the right way, everybody wins. That's the power of VantageScore. Get started at VantageScore.com.

16:27Rosie Blau: Okay, Avantika, what have you got for me this morning? Right, I've got baked Wotsits Giants. They promise to be really cheesy. Oh, big bag. I actually love these. Mm-mm-mm. Very crunchy. What a nice breakfast. Now I've got what are called Wotsits. Crunchy. So this is like the amped up version.

16:59Rosie Blau: Avantika Chilcotti, our global business writer, tell me why we're talking about crisps this morning. I've been going around big food firms, getting into the really nerdy science of how they make their snacks more Moorish. And at the moment, all of the focus is on texture. OK, so we've gone from crunch to texture. What are we really talking about here? So texture is how the food scientists at these firms talk about it. I've been to Mars, Mondelez. So texture, it comes out in the sound. So as I eat those Watsits, it's a pretty full-on sensorial experience hearing myself eat. Plus there's a bit that they call mouthfeel.

17:41And they spend a lot of time thinking about mouthfeel. How does this feel in my mouth? And in the case of both those Watsits, afterwards I've got this coating of cheese. It's still there, actually. It stays with you. So there's different elements of texture. And that's a good thing? Well, so right now, what we're finding is that consumers really want to try things with novel textures. In part, that is the fact that as much as we could play with taste, we've played with it. So even these crisps, you've got really cheesy written all over them. You've already had amped up spicy snacks. You've had amped up sweet snacks.

18:15They've done as much as they could. One analyst put it to me, you've reached peak flavor. And so texture is the next way to give people a bit of novelty.

18:23Rosie Blau: And are we just talking about crisps here or is it other snacks as well? It's really anything you can inject texture to, so including chocolate bars. When I went to Mondelez, one thing that I got to try, which was also very Moorish, is the dairy milk biscoff. It's part of this line they've got called the And More line. They've shoved bits of biscuit, bits of biscuit cream into a regular milk chocolate bar. and what I find really interesting about this is that it's not easy to do. I mean, there's so much to think about. So if you put biscuit next to anything that's basically got lots of water in it, it will go soggy.

19:01So you've got to engineer this bar to have a layer of something fatty, which in the case of the biscoff is this biscoff spread that almost is sort of a protection for that biscuit layer because what would be worse than cracking open a dairy milk biscoff and it being a bit soggy in the middle?

19:18Rosie Blau: Avantika, forgive me for being a cynic. I obviously admire your pursuit of the science of snacking, but is this really new? I mean, Wotsits, Maltesers, Crunchies, there are lots of crunchy, textured snacks that we've been eating for a long time. Yeah, I had the same scepticism. And actually, you look at the survey data, I often look at a group called Mintel. And they say in the last two years, demand for texture indulgent snacks has gone up 28 % in America. So in part, this is people seeking out snacks for their texture characteristics. As you're saying, this might just be a marketing thing. To be honest, those first Watsits, the traditional ones, they were pretty crunchy.

20:00But if you just pop the name crunchy on it, people are more likely to grab it. So in part, this is a marketing effort. But in part, it really is an R &D effort.

20:09Rosie Blau: And why that demand for texture? So this is super interesting. It tells you so much about consumers and how they're changing. So as with everything in food, the first thing is the rise in GLP-1s. We're starting to get some research on these slimming drugs, the effects they have. Obviously, they suppress your appetite. But I've been looking at Reddit forums and people say it changes how things taste as well. People, when they reach for a snack, all the analysts I speak to say they're asking, is it worth it? So they're looking for more of a sensorial experience. If something very sweet or very filling can be very overwhelming, perhaps texture is another way to draw them in.

20:47I mean, the other big factor is social media. Yum, yum. Finally trying the chili lime chips from Chipotle. I was joking earlier, you can hear me eat. It's not very pleasant for most people, but especially amongst Gen Z, there's a trend for these videos. and it's down to basically these influencers showing you snacks and they're amping up the sound. And some people, it's called ASMR, they find very satisfying and calming sound. And, you know, ASMR is so popular, there's this tag called Crunch Talk. And across social media, the data is that that racked up 1.5 billion views last year. And honestly, in these videos, they will amp up the volume on, you know, this, the sound of the packaging.

21:33They'll do this for a bit. They will spend ages cracking open that crisps packet or that chocolate bar so you can hear the wrapper. So there's a whole social media trend around it which basically is free advertising for these firms.

21:46Rosie Blau: Get the chili lime chips. I think, to be honest, the biggest thing is that we've reached peak taste. If people want to try something new, there's not that many more flavours of crisps to be had. Here's another thing you can add to the mix. Avantika, thank you very much for breakfast. Good to talk to you. Thanks for having me, Rosie.

22:30Ann Wroe: it hatched. When it hatched, its wings, which were meant to be wonderfully brown and velvety with a blue eye spot, had a varnish on them which was to dry and then would enable the wings to spread and the moth to fly. But the mason jar was too small, so when the moth tried to open its wings, it couldn't do so.

23:00Ann Wroe: When they released it in the street, it could only walk, and she was the only one who still watched it as it staggered down the pavement. In her mind, for the rest of her life, it was still going on. She was often the sole observer of things. At her kitchen window, once she saw a goldfinch shaking a thistle head, until down, bright tassels of it floated all over the landscape. She noticed a leaf curled inside a bud, ready to unloose itself. She watched frogs leaping up through filaments of blue-green algae, croaking with joy as they surfaced, surfacing into awareness, as she put it. but she also watched another frog stocked still by the side of a pond in Tinker Creek while a giant water bug liquidised its insides and sipped its life away.

24:09Ann Wroe: Tinker Creek was the place where her best-selling book was written. It caused a sensation when it was published in 1974 and in 1975 it won a Pulitzer Prize.

24:29Tinker Creek itself was just half a mile or so from her house in Roanoke, Virginia.

24:36Ann Wroe: She would walk down there in the mornings and just observe every possible thing. But she didn't want fame for doing this. She was terrified of all the fussing. and when the press started descending on her because they found her very attractive as she was as well as a good writer she fled away with her writer husband to the other side of the country to Washington State and the islands off the coast and from these exiles she produced an extraordinary body of work essays, newspaper articles journalism, two novels an autobiography, books of philosophy She liked to work in very small spaces, usually cabins or shacks, and she deliberately put her desk facing away from the view.

25:37Ann Wroe: All this intricate observation of nature, though, was not just for nature's sake. It was also because she was trying to answer a huge overall question. Why does there have to be destruction after creation? Why is creation so prodigal at the same time, so wasteful? Why all this abundance and all this care if what you have created, the larvae or the nymphs or the leaves, everything just falls to pieces or is eaten afterwards. What was the creator thinking, if there was a creator? Maybe, if she could believe in one, he was just so powerful that he exalted in his own power and created freely and wildly, and you couldn't find a rational reason for it.

26:36Ann Wroe: There was a very strange incident she recorded in the middle of her book The Writing Life, which was a spontaneous combustion, it seemed, of an old Smith Corona typewriter. It wasn't connected to power, so there was no explanation for it. She just felt the house shake, and when she went upstairs, the whole device was exploding in fire and ash. It might have been a cigarette that she had carelessly put out, but she didn't say so. In her telling, there was just suddenly fire. And all through her writing and observing life, too, there was fire. She would see it if she tried to witness things, not in a completely devoted way, just looking at the object for hours, but letting go and just looking generally around.

27:38Ann Wroe: and then it was that she began to see fire. She saw at sunset light rolling along the ridges of the mountains like a fireball. She saw the grass as holy fire and she also saw the backyard cedar tree suddenly become a tree of lights. When that happened, she felt she was lifted like a resounding bell. But again, she was the only one who noticed these strange eruptions and occasions of fire. It was only her witnessing all these beauties, just as she had been the only one, to witness the moth trying to walk down the street or the frog being attacked by the giant water bug. All this observation was necessary to understand what the creator was doing.

28:44Ann Wroe: That was what she was for.

28:49Rosie Blau: Anne Rowe on Annie Dillard, who has died aged 81.

29:07Rosie Blau: Can you design a town to make everyone happy? That's the question in tomorrow's episode of The Weekend Intelligence, when we visit Celebration, a city dreamed up by Walt Disney. The idea was to combine the traditional charm of old town life with progressive schooling and cutting-edge technology. Tomorrow, we visit Celebration on its 30th anniversary to find out whether Disney's utopia really spread delight. You'll need to be a subscriber. And that's all for this episode of The Intelligence. The show's editors are Chris Impey and Jack Gill. Our deputy editor is Sarah Larniuk. And our sound designer is Will Rowe.

29:45Rosie Blau: With additional help this week from Mark Burrows. Our senior producers are Henrietta McFarlane, Emily Elias and Alizé Jean-Baptiste. Our senior development producer is Rory Galloway. And our senior creative producer is William Warren. Our producer is Anne Hanna. And our assistant producer is Kunal Patel. with extra production help this week from Katie Peterson. We'll all see you back here for the Weekend Intelligence tomorrow. All right.

30:39Chat Concierge and it's simplifying car shopping. Using self-reflection and layered reasoning with live API checks, it doesn't just help buyers find a car they love. It helps schedule a test drive, get pre-approved for financing, and estimate trade-in value. Advanced, intuitive, and deployed. That's how they stack. That's technology at Capital One. I love days like these at the pool, but my moderate to severe plaque psoriasis can sometimes take me out of the moment. I'm ready to make a splash with clearer skin thanks to SkyRizzy, Rizinkizumab RZA. SkyRizzy is a prescription-only injection for adults who are candidates for systemic or phototherapy.

31:18At four months, most people saw 90 % clearer skin when measured from head to toe, and many were even 100 % plaque-free. People also saw significant improvement in psoriasis symptoms of pain, redness, itching, and burning. Don't use if allergic to SkyRizzy, Serious allergic reactions, increased infections, or lower ability to fight them may occur. Before treatment, get checked for infections and tuberculosis. Tell your doctor about any flu-like symptoms or vaccines. Now there's nothing on my skin, thanks to SkyRizzy. And that means everything. Ask your doctor about SkyRizzy, the number one dermatologist prescribed biologic and psoriasis.

31:53Visit SkyRizzy.com or call 1-866-SKYRizzy to learn more.