This BBC podcast is supported by ads outside the UK. Start today for 1 euro per month on Shopify.nl. Stel je voor dat je in de techniek niet nog harder werkt, maar slimmer. Omdat warmtepomp op afstand uitgelezen kunnen worden. Zodat je monteurs precies weten wat ze moeten repareren. Ontdek onze slimme technologie op kpn.com slash slimmerwerken. KPN, voor een beter werkend Nederland. Hello! What does a tiny molecule that can copy itself tell us about life? How can a simple genetic switch increase rice yield by 25%? And what does it tell us about computer cognition that we somehow ended up with an AI that, rather than being the hyper-rational truth-teller, has all the over-confident bluster of an ill-informed politician in a TV interview? All that is in today's Inside Science from the BBC World Service with me, Tom Whipple. But wait, there's more. There's also Anja Hoogja, science columnist at the Financial Times. Anja has chosen her favourite stories from the science journals. Anja, give us a tease. Hi, Tom. We're going to go a bit rock and roll, rock in particular, because we're going to be talking about why there are missing bits of the geological record. So it's a kind of a real detective story about what's happened to them. Well, before the rock, let's roll further into the geological record. The date, several billion years ago. The place, somewhere wet, or as is the accepted convention for these kinds of wet place, a primordial soup. The event, a small molecule called RNA strung in a line, folds itself up, hoovers up some other molecules and makes a copy of itself. Is this when chemistry became biology, the most important bit of chemistry in history, the moment when life of a sort began? If it is, then we may now be a little closer to understanding what happened. In a lab in Cambridge in a non-primordial soup, researchers have made a small molecule of RNA that can indeed sort of copy itself. I am joined by one of the descendants of a self-replicating molecule, and one of the most distinguished and certainly most prolific science writers in Britain, Philip Ball. Phil, have we discovered what your great-great-great-great-great-great-grandfather was up to? Is this at last the answer to how life began? Oh, if only, Tom. We don't have anything like the answer to that. There are more theories than answers at the moment, but this is very interesting work because it gives a clue to how, before there was life on Earth, molecules might have been able to make copies of themselves, this property that is crucial to all living things. Tell us why this is exciting. So why do we want copying molecules? Because that's what organisms do. They are able to replicate. They're able to make copies of themselves. And the idea that some people have is that before there were organisms as such, this ability to self-replicate must have developed in molecules themselves. You know, as they were able to do that, they maybe got together into teams and got more and more complicated. And eventually they got into sort of cell-like structures. And, you know, after that, it was just evolution. So that's the idea here. OK, well, let's start at the beginning. What is it that happened with this particular paper? We've had self-replicating molecules before. So let's not get too excited about that aspect of the problem. And in particular, what a lot of people have focused on are these molecules called RNA. So we all have RNA in our cells. In our cells, what the RNA generally does is that it acts as a kind of an intermediary, converting the information that's in our DNA, in our genes, into protein molecules. And those are the enzymes that sort of conduct all the chemistry in our cells. So the puzzle really in thinking about the origin of life was, well, if we need DNA and we need proteins, how does the whole thing get started? It's a chicken and egg problem because you can only make DNA from protein enzymes. They're the ones that make these long strings of DNA. But you need the DNA in the first place to get the enzymes. It seemed like RNA might solve that problem because RNA turns out to be a very clever molecule. it can encode information. Basically, that's what happens when the DNA gene gets copied into RNA, but it can also act as a catalyst, which is the kind of thing that proteins do. So maybe things just started with RNA itself doing both of these things. So that's what these researchers have gone after. Can we find RNA molecules that can replicate themselves? The key here is that previously, those were really big molecules. And if you're thinking about a molecule that can just sort of form spontaneously, maybe from something like this prebiotic soup that you talked about, they've got to be small. What these researchers have done is to find the smallest yet self-replicating RNA molecule. And how do they do this briefly? It's a little bit like conducting evolution in a test tube. In fact, it's literally called in vitro evolution sometimes. What you do is you start off with a pool of sort of random RNA molecules and you look for ones that show some signs of being able to sort of make copies. If you find some molecules that seem able to do that to some degree, you extract those from your solution and shuffle them around a bit, make some mutations in them to see if you can get better ones. And you just keep repeating, iterating that process. So they just did this several times until they found the shortest sort of possible molecules they could find. I mean, in a sense, this is sort of one of the most important questions there is. This is, you know, forget about evolution. This is how we all sort of kicked off. This dates back to Darwin was thinking about how you get these. Yeah, this question of how life began, it was one that Darwin's theory of evolution didn't answer. You know, you had to take life for granted. And it was really a theory of how life changes from one form to another. Darwin began by saying, well, it's just too complicated. How on earth are we ever going to find out about that? But then in 1871, he seemed to change his mind. He wrote to his friend Joseph Hooker in a little letter that was sort of speculating about what might have happened. And he said in that letter, he said, and this is a quote if and oh what a big if we could conceive in some warm little pond with all sorts of ammonia and phosphoric salts light heat electricity etc present that a protein compound was chemically formed ready to undergo still more complex changes. At the present day, such matter would be instantly devoured or absorbed, which would not have been the case before living creatures were formed. So he was thinking about how do you get the first protein? But actually, as I say, it turns out that maybe the better question is, how did we get the first RNA molecule? And do you think this has answered it? Absolutely not. No, because for one thing, these are already quite complex molecules. Getting an RNA molecule 45 of these units long just to form spontaneously, you know, it's really hard to see how that's going to happen. And even when you've got these molecules, you still then have the question of, you know, that's not life, that's not living. You've got to figure out, well, where do they get their energy resources from? How do you get a metabolism from out of this? So you're a long, long way from a replicating molecule to actually the first living cells. All it does, but it's a big all, is to tell you that actually finding an RNA molecule that can catalyse itself, there may be many more ways to do that. Last question. Will we answer this question? Will we know what happened in Darwin's primordial pond? I'm not convinced that we ever really will, because, you know, as you say, Tom, it was four billion years ago, more or less, that it happened. Any traces of how that happened are very, very hard to discern now. I think what we're going to have is plenty of theories, some experiments, too, like this, that can see whether those theories kind of hold water. But which of those theories, if any, is right is probably a question I don't think we're ever going to be able to resolve. Thank you very much for speaking to us, Phil Ball. Thanks. Now, here's Roland Pease with his bountiful harvest from the world's science this week, and it's potentially good news for rice farmers. A simple genetic switch could be the key to substantially improving rice yields while also reducing the amount of expensive fertilisers that farmers currently apply to their crops. The unexpected logic to this double win is that nitrogen fertilisers don't only provide important nutrients for the growing plants. A secondary indirect role is to counterbalance a tendency for plants to grow roots rather than stem, leaf and grain, a tendency which Zhe Qi and his colleagues have found a means to correct. Like all other plants, essentially, rice invests more energy into the growth of root. when the surrounding level of nitrogen drops, essentially, so that they would prioritize root growth so that they could explore available nitrogen in their environment. I mean, that makes sort of sense, doesn't it? Of course, of course. It's very beneficial for the plant. But normally it comes with a price, which is shoot growth. So they would normally do this at the expense of shoot, which is, again, adaptive in nature, because, as you can imagine, a plant, when they sense their environmental nitrogen is low, they don't want to produce flowers. They don't want to reproduce because it's not ideal to do that at the moment. So they sort of wait. They enter a sort of dormant state where they explore the surrounding soil for nitrogen, but conserve energy above ground. But that, of course, is not what you want as a farmer. You want to get that crop going and the grains to be ready for you. Exactly. So that, although very adaptive in nature, is sort of detrimental in agricultural settings because, as you can imagine, it causes a direct negative impact on grain yield because you don't get enough fruit, you don't get enough grain being filled. But also the underground root growth being promoted is actually competition. With the other plants. With the other plants, exactly. Especially in agriculture where crops are normally grown in monoculture. So it's actually harming its siblings, essentially. You just get dramatically reduced grain yield if the nitrogen level is low. So what you've done, as I understand it, in this work is you found the master switch, the genetic switch, as it were, that turns on this root versus stem kind of growth. Exactly. Rather than turn on, we would say to uncouple it. So what we did was randomly mutagenise a particular variety of rice that we are using in the lab, and then we just screen a huge number of these mutagenised rice population for an individual that doesn't exhibit this classic response of enhancing root growth at the expense of shoot, which translated to an increase in root-to-shoot biomass ratio. And luckily for us, we did in the end identify a particular mutant that maintained a stable root-to-shoot biomass ratio even under low nitrogen. And it's from that mutant that we carried out the subsequent study that identified the gene. So that's quite interesting. But it does imply then that there is this kind of sensing mechanism that's involved in the choice that wild-type plants will make that you can then start to manipulate in crops? Yes, exactly. So the gene we identified is a regulator of growth, essentially. It nevertheless achieved what we wanted, which is not sacrificing shoot growth when nitrogen level is low. So keep promoting shoot growth and keep promoting root growth so that the whole plant looks just normal, even under low nitrogen conditions. So that's a mutant plant that you've created, but you can then take that gene and put it into more normal crop varieties. Is that right? So there are actually two ways of doing this. First of all, we can obviously do the very scary genetic modification because we have identified this crucial gene. We could just try to overexpress it. So express the level of the gene at a higher than normal level in sort of standard rice background. But what is particularly impressive with our study, if I may say so myself, is that we screened over 3,000 natural rice varieties, and we identified a naturally occurring allele or version of this gene, which is called ring code 1A. I should probably say that earlier. And this is completely natural. So it was identified to be carried already by some rice varieties in the collection. Right, so it's already there, Yeah, but the effect wasn't recognised. The effect wasn't recognised because we didn't know what to look for. Yeah. So what we did was to introgress or basically just introduce this version, what we call an elite haplotype, into a standard rice background through traditional breeding, just crossing and selection. And then we actually achieved an enhanced nitrogen use efficiency and also grain yield. So what's the effect like? You really have to be in the rice field to see the effect. But when I was there, it was particularly impressive, first of all, the field station that we had. You can visibly see an increase in the improved variety in terms of grain yield. And that is particularly obvious for low nitrogen input. So for these field stations, if you can imagine, these are separated into different plots with different level of nitrogen fertilizer input. But on the low fertilizer input plot you can actually visibly see that the improved variety have more grains have more tillers forming even though the nitrogen level is low And we calculated the amount it a 24 increase which is quite impressive if you think about it. Yeah. I mean, that's really impressive. I know. I mean, are you confident that this gene that you've discovered will be bred into rice crops? And do you have a sense of how long that might take? Oh, I think so. And I think because we have done multiple field trials in China across different locations, I think we're actually getting very close of putting that particular variety in the farmland of China and for farmers to actually adopt. However, what we're aiming for is obviously much more ambitious than that. So we're currently testing whether or not this haplotype can be integrated into other rice varieties that are locally adapted in other countries, for example, or even other continents. And it is also very important to test if this particular elite haplotype can adapt to the local environment. Of course. Yeah. So, for example, they would face different climate, they would face different soil type or even managing system. And whether or not upregulating this regulatory gene or integrating this elite haplotype would provide a stable increase in natural use efficiency and grain yield in these particular locations, would tell us how applicable this gene is. Given rice feeds more than half the world, you can see what kind of impact this genetic work would have. Zheqi said he's hopeful a similar trick might work in other cereal crops like wheat. Zheqi is a plant biologist at Oxford University and his study, along with experts at Nanjing Agricultural University in China, was just published in Science. Thanks, Roland. A reminder, you're listening to Inside Science from the BBC World Service. And also, don't forget that you can get in touch with the program. Email insidescience at bbc.co.uk. With templates and AI tools from Shopify, you have no stress over design. And you can easily create a beautiful web shop. You're ready to sell it? Then you're ready for Shopify. Make your own entrepreneurs' dream. And start today for 1 euro per month on Shopify.nl. Stel je voor dat je in de techniek niet nog harder werkt, maar slimmer. Omdat warmtepompen op afstand uitgelezen kunnen worden. Zodat je monteurs precies weten wat ze moeten repareren. The AI wasn't meant to be like this. Yes, the Android data in Star Trek was annoyingly literal and prone to pedantry. But he never, say, made up a law citation or invented a quotation. Yes, Skynet in Terminator had its flaws too. It was intent on destroying humanity. But in its defence, no one ever doubted it did so for the most logical of reasons. And you could rely on it not to hallucinate about what its malevolent, shape-shifting death robots were up to. So how is it that when AI actually appeared, it wasn't the one thing we expected of it? Rational. Last week, Michael Wooldridge, Professor of Computer Science at the University of Oxford, gave this year's Royal Society Michael Faraday Prize lecture. Its title, This is Not the AI We Were Promised. We spoke to Mike to ask him what he means. Well, we've long had expectations about AI, but kind of the one thing that we expected from AI was that it was going to be kind of logical, that it was going to be correct, that it was going to not talk complete and utter nonsense to us half the time. and that's not what we got. It's weird to think that the concept of hallucinations in the context of AI didn't really exist until five years ago. We certainly knew AI could get things wrong, but didn't expect that it was going to be such a prominent feature. What is going on in the architecture that appears to make it so intractable to truth? Well, okay. So the first thing is it has no conception of what truth is, and it's not trying to tell you the truth. What it's trying to do when you give it a prompt is to give you the most likely, the most plausible response to that. And sometimes the response coincides with the truth, but there's no reason for it to. And one of the interesting things from last year is we were all on the edge of our seats to see what GPT-5 was going to be like. If you'd asked me, what's the one thing I would ask for? Forget anything else. What's the one thing I would have liked GPT-5 to get right? It would be this, stop hallucinating. If you don't know the answer, just come back and say, I don't know the answer. If it had been able to reliably do that, it would have been utterly transformative. There was a paper out in Science, or a views article out in Science. It was a long alongside a study which had found that if you teach and i'm i am anthropomorphizing here and doing so deliberately if you teach a large language model to be uh to be naughty in one area to produce insecure code it ends up being naughty in other wholly unexpected areas now there was an article alongside that that says we need to start applying the principles of ethology of animal behavior to these things we need to start treating them a little bit like organisms should we be anthropomorphizing these and, you know, imputing motives and reasoning to them. I personally do think that anthropomorphizing is exactly the wrong thing to do. There is no mind on the other side of the screen. These aren't creatures in any meaningful sense. And actually, we're getting much more sophisticated ways now of dealing with negative behaviors. One of the weird things that we've discovered about these models, which again took me by surprise, is the idea that once you built them, you spend a lot of time trying to teach them manners. And what you do is the model comes back with an answer and you decide whether that was a good or bad answer. And if it was a bad answer, then you adjust the model. So you get humans involved. And I say it kind of really is like trying to teach it good manners. No, that wasn't a polite response. But you've just Andrew Revolvedized it. Very lazy. But why is it lazy? For 50 years, we've had the Turing test. We've had computer scientists have dodged the idea of defining intelligence and saying whether something is intelligent and sort of like a mind by saying, look, if it interacts with us like a mind, then it is one. And let's just dodge that problem. These things go ahead and basically everything everyone would say, these are past the Turing test And then Yukonsky just changed the goalposts and say no no we never meant that all along Well I just don think that helpful in our relationship with AI Again, I've just anthropomorphised, haven't I? But I think the point is, for me, what I think we need to train ourselves to realise is that these are tools. The difficulty is, exactly as I just demonstrated, that it's very easy to do that. I mean, as an example, just from the last week, I was busy getting one of these models to tidy up some, computer codes on programs that I'd written 20 years ago. And I couldn't remember how these things work. And then at some point it said, that's very on brand for you, what you want to do. And I was slightly startled at that, but what do you mean on brand? And it lately said that after that, it said I was a contrarian. It said, that's a typically contrarian thing for you to ask, but it's designed to be realistic in the answers that it gives to us. So it's no surprise then that we start attributing to it those human-like attributes, because we're kind of designed to do that. If these aren't the AI we're promised, is there a way in which you can see us getting the AI we're promised? Is there a more simply logical AI that will one day be able to actually answer our questions properly? Or is it a wholly new architecture? It could be a wholly new architecture. There are enormous numbers of people right now trying to work on that problem. Can we fix these problems with large language models? An awful lot hinges on it because so much money has been invested in them. And when I talk to companies about using them, the really big issue that they're worried about is their unreliability. Will it be just something that we can tack on top of existing models? That's one way that people are going right now. Or could it be, as you suggest, something fundamentally different? I don't know. That was Michael Waldridge, Professor of Computer Science at the University of Oxford. I am joined by Anj Ahuja, science columnist at the Financial Times, to go through some of the other science news from the journals. Anj, what's your first story? Oh, hi, Tom. I was reading my journals this week and I came across something I'd not heard about before called the Great Unconformity. And if you look at the geological record of the Earth, it looks as though there are chapters missing. So, for example, you can find places in China where you've got 500 million year old sandstone directly on top of 2 billion year old granite. So kind of what's happened to the rocks in between? So there is a paper out in the Proceedings of the National Academy of Sciences that kind of questions the prevailing wisdom on this, which is that it's something to do with the snowball Earth. There was lots and lots of glacial erosion that happened around the same time and that lots of these kind of missing rocks were just eroded away and ended up in the oceans. This paper suggests that actually it was tectonics that did it. And there was this surge of mountain building maybe between sort of 500 million years ago and a billion or more years ago. and what happened was that that pushed up a lot of this rock and that was just eroded away and weathered away. And I just thought it was fascinating because, A, I'd never heard of these missing chapters. So when you say chapters, sort of half a billion years has just sort of disappeared in places. Exactly. And, you know, it's fascinating because you can only dig down and play with the rocks that you can see. But it also, in this kind of detective story, I learned about lots of really interesting things like thermochronology, which is a way of looking at the radioactive content of rocks so that you can work out how hot they were and you can work out how close to the surface of the earth they were. And so it's about piecing all these different little bits of the geological puzzle all over the world together to try and explain why you've got this rather substantial missing bit of the geological record. Look, I'm feeling very trivial about the paper that I've chosen. So mine is the secret of squeaky basketball shoes. So I think we have some illustrative sounds. Listen to those squeaks. Now, I've never thought about the squeaks, but the really interesting thing about the squeaks is they're the same pitch. They're not changing with the velocity of the shoes. You just hear this squeak. Where does it come from? And thank God for science, someone's looked at it. They have squeaked basketball shoes on a thin plate of glass, which they have shone light into. This enabled them to see exactly what was going on in the soles of the basketball shoes. And you discover that you've got these waves that are sweeping through them at 300 kilometres an hour. But crucially, the frequency of these is to do with the soles of the shoes themselves. They use the word, you know, it's like a tuning fork and you have this resonant frequency of your basketball shoes, which makes those squeaks. I don't know about you, Ange, but that has taken me back to a gym. Absolutely. That's taken me right back to a time then that I would probably rather forget back in PE at primary school. If only you'd known there was important science involved. There's a lovely commentary alongside the paper about this by Bart Weber, who's from the Netherlands. And he ends it by saying, so talking about this and how using the total internal reflection of glass to see the bits where the soul starts rippling and then working out the pattern of these ripples. and says the study not only explains a familiar sound, but also reveals how much complexity can hide in the seemingly simple act of sliding. And I love this place where complex science meets the real world and you realise you just haven't thought about stuff. Well, I think you've hit on a way of teaching teenagers about physics and all sorts of things. That's a great teachable moment, as they say. I think so. That's it from Team Inside Science for the week. Our producer was Clare Salisbury and I'm Tom Whipple. Bye bye. It's time to see what you can accomplish with Shopify by your side. We focus on the part of the internet that most people don't know about. It's called the dark web. Undercover, in the furthest corners of the dark web, US special agents are on a mission to locate and rescue children from abuse. Move in now. From the BBC World Service, World of Secrets, the darkest web follows their shocking investigations. Listen on bbc.com or wherever you get your BBC podcasts.
