Gareth Mitchell:               Hello, everyone. I'm Gareth Mitchell. Today, same sex sexual behavior in monkeys. Also making robots more adaptable and what we can learn from dogs who lose one of their limbs.

Dr Antoine Cully:              If you just go on YouTube for instance, and you type, "Three legged dogs playing," you will see that actually they just don't mind and they can just play like normal dogs. And so we were like, okay, we need to be able to teach a robot to be that resilient in this kind of unfortunate situation.

Gareth Mitchell:               And we preview the podcast that will tell you all you need to know about the genomics revolution and what it means for your health.

Gareth Mitchell:               All right. Welcome aboard everybody. It's always nice to have you. We're going to start with the amazing Hayley Dunning to talk about sexual behavior in monkeys. And that's been hitting the news, hasn't it? Mainly because you've been writing about it, Hayley, so well, tell me more.

Hayley Dunning:               Yeah, this is a really interesting study about a colony of macaque monkeys on an island off Puerto Rico, and it's about their same sex sexual behavior. So we know a lot of animals in the animal kingdom engage in what we would call same sex sexual behavior, so males mounting males, females grooming females, all that kind of stuff.

                                                But a lot of what underlies that is unknown. So these researchers did a very long term study. They looked at these monkeys for three years and they studied 236 of the males. They observed their behavior, but they also have pedigree records that detail the parentage of each individual back to 1956. And they recorded all social mountings for the male, so that's male on male behavior, but also males on females behavior. And interestingly, they actually found that male same sex mounting was widespread with 72% of the males engaged in that while only 46% actually encountered different sex mounting, so that's males mounting females. So it seems these males do it a lot.

Gareth Mitchell:               I mean beyond the numbers and it just being intrinsically interesting, what does it tell us?

Hayley Dunning:               So as well as being widespread, the team also analyzed whether this behavior was heritable. Using the pedigree data, they found that it was about 6.4% heritable, which provides the first evidence of a genetic link. And it's also a similar figure to other heritable behaviors that we see in primates such as grooming and sociality.

                                                And they also sort of looked at why the monkeys might do this with caveats, of course, here that we're not extrapolating to humans. But in these monkeys it seems that the more same sex behavior the males engage in with each other, the more they're likely to back each other up in conflicts within the group. And that actually can make them potentially more successful in mating with females, so producing more offspring.

                                                So the researchers say these findings support counter-arguments to the idea that same sex behavior defies nature and evolution, this so-called Darwinian paradox, that it's not a good thing for nature, showing that perhaps it has these benefits.

Gareth Mitchell:               And the researchers there know taking all this and they have some bigger picture ideas with the conclusions that they're drawing, don't they?

Hayley Dunning:               Yeah. So the lead researcher, Professor Vincent Savolainen, who is the director of our Georgina Mace Center for the Living Planet Imperial, he says that their research shows that same sex behavior is in fact widespread among non-human animals. And he said, "Our mission is to advance scientific understanding of same-sex behavior, including exploring the benefits it brings to nature and within animal societies."

Gareth Mitchell:               All right, Hayley, thank you very much indeed for that. That is Hayley Dunning.

                                                Well, now to the researcher who wants to get robots back on their feet again. The thing is, getting a bot to walk upright like a human is one of the hardest challenges in robotics. Just look at one of the many countless videos of various droids tripping over the slightest of obstacles.

                                                At Imperial's Adaptive and Intelligent Robotics Lab, Director Antoine Cully and his colleagues are working on new learning algorithms for more versatile robots that can do anything from climbing stairs to driving cars. The team also wants to leverage the power of AI in robotics to help the machines better handle unexpected situations. Antoine has recently told a meeting of Imperial business partners all about the work. And on the back of that, he's been speaking to us. We started with the debate within robotics over how many legs is best? Two legs, four, six?

Dr Antoine Cull...:             Yes. So actually that's funny because in my lab we have all the different flavors. We have robot with six legs, with four legs, with no legs, just arms. And indeed there is always this kind of trade off between some things that is like humans because we want robot to operate in our world, in our daily life. And our daily life is structured around the human body. The handle of a door is actually at human height.

                                                So for example, if you have a robot that looks more like a dog, opening a door becomes a big challenge. But on the other end, controlling human robots, designing human robot is a huge challenge. Also, human robots usually cost at least an order of magnitude more than a quadruped robot.

                                                So there is always this kind of trade off between the complexity of the system, the complexity of the technology, and what we really want to do. For instance, I think if we want to have robots that are able to deliver parcels, to do the last mile delivery, for instance, probably a robot dog would be more appropriate than a very expensive, brittle, hard to control human robot.

Gareth Mitchell:               Yeah, because I was trying to think about why you would even want to design a bipedal robot. You've already mentioned opening door handles. I've just thought of that. Yeah, just how inevitably bipedal human-centric the world happens to be because we've invented it and we live here. But I suppose also it's where your eyes are. You can just see further ahead the higher up you are. Is that a big advantage for wanting to be bipedal if you can get across the AI and the engineering challenges?

Dr Antoine Cull...:             Not necessarily. Because what we tend to do sometimes is actually to place the head, or at least the cameras of the robots. So for instance, there are some very famous robot from Boston Dynamics where they just mount small robotic arm on top of the robot dog and then they place a camera at the tip of the robotics arm. This means that the camera can actually be two meters high, can orient itself and look at different locations.

                                                So we can still have this kind of good overview of the situation with a robot dog. But for instance, a situation where human robot would be more useful or a taller robot would be very more useful is just to be able to reach a desk. Or what we would like to have is robot that can assist elderly people in the house, empty the dishwasher, cooking breakfast, cooking meal, and so on. And all of this, you will need a robot that can stand to the stove and flip pancakes, as I said. So with the robot dogs, that's quite challenging. With a human robot, that's definitely more appropriate.

Gareth Mitchell:               Well, let's talk about your robot dog. You've done demonstrations where this robot dog has four legs, as you might expect, has it's very adaptable, isn't it? So for instance, what happens when one of the legs fails? You've simulated that successfully.

Dr Antoine Cull...:             Yes, exactly. So we took inspiration from actual dogs, because if you just go on YouTube for instance, and you type, "Three legged dogs playing," you will see that actually they just don't mind, and they can just play like normal dogs. And so we were like, okay, we need to be able to teach a robot to be that resilient in this kind of unfortunate situation.

                                                So we taught a robot using deep enforcement learning or to be able to walk in a lot of different manners to go forward, backward, turning left, turning right, moving the legs at different pace, different heights, jumping and things like this, while being able to only rely on three legs at the time. And that offers a lot of resilience in case of broken motors or broken legs or broken sensor, then we can just deactivate one leg and carry on with a mission with only the three remaining ones.

Gareth Mitchell:               Now you say that one of the big research areas in your lab is leveraging AI, and I'm thinking about the advances in AI in non embodied robots, and we all know just how crazy that's going, I suppose most visibly with the large language model chat bots everyone's so aware of. Many people are using them, building businesses around them.

                                                So that's in the non embodied space. I just wonder, in your case, it's so much trickier, isn't it? When you're talking about an intelligence, an AI that is wandering around the world, is present in the world as an embodied physical system. Can you just talk me around how much of a challenge that is? Or am I overstating it?

Dr Antoine Cull...:             No, no. So that absolutely a big challenge. And indeed, all the work around large language models is very inspiring also for us. But one big challenge we have is that actually some of the breakthrough behind this improvement of language models is actually the availability of a huge amount of data. Big companies are just scraping the entire web to grab any pieces of text and then to train their model on it.

                                                The issue is that this kind of wealth of data is not available for robots because every single robot is different in different operations. We're not recording all the data we have online. So instead of having this kind of machine learning or AI in the realm of big data, in robotics, we are more like in the world of micro datas. How can we learn something that is robust and so on when we have, not a 10th, but 1,000,000th of the available data for large ongoing model. In particular in my lab, we specialize in this kind of situation where we have unexpected situation.

                                                This means that it might be a situation that nobody on earth experienced before. So for instance, not a lot of people try to train a quadruplet robot to work with only three legs. So there is literally no data. We have to collect the data directly on the robots while the situation is happening. And we need to learn something useful, robust, in a fraction of a minute or as fast as possible. Because if your robot is damaged and it's not able to cope with this situation quick enough, then it might actually have the risk of damaging itself even more. So there is always this balance of we need to really adapt quickly, we need to be conservative to not induce more damage, and we need to learn all of this with a very minimal amount of data available.

Gareth Mitchell:               So finally then, where is all this going then, especially in this world of robotics, adaptive robotics, which I find absolutely fascinating? What are we going to be seeing in 10, 15, 20 years time?

Dr Antoine Cull...:             So Michael is really to be able to have robots more present in our everyday life. I mean, robots are used every day in a lot of different application and mainly in factories and assembly lines, but it's a space where everything is extremely well controlled. The issue is that in our everyday life, everything is way more open ended, and we need this kind of adaptation, resilience to bridge this factory to real life gap. And so this is kind of direction where I hope robotics and AI is going.

Gareth Mitchell:               That's Antoine Cully, who was speaking at the recent Imperial Business Partners event. And Antoine has also been working with Imperial's Tech Foresight team on an excellent map that visualizes how different AI tools and techniques intersect and how they pave the way to the future. There's plenty more via ImperialTechForesight.com.

                                                Well, each month on the podcast, we bring you a different sonic treat from the wide range of pods now available from Imperial. Just search for the Imperial College podcast directory or find it via the inspired pages on our website and you'll be treated to titles on health and medicine, science and technology, climate, and the environment, business and careers and more.

                                                We've delved into the directory this month and pulled out DNA&. Presenters Hannah and Angelos are telling us all about the genomics revolution and how DNA impacts personalized healthcare. They've released two episodes so far.

                                                So what's the first installment all about? Well, let's hear from the podcasters in their own words.

Dr Hannah Maude:          So the main aim for episode one, this episode, is to make sure we're all on the same page. So we're going to give a quick crash course in DNA and how it's important for our health. We're also going to introduce Genome UK, which is the UK government's policy strategy for making the UK, and I quote, "the most advanced genomic healthcare service in the world."

Angelos Manolia...:         I think Genome UK is a great example for episode one because it really defines the aim to transform healthcare towards prediction and prevention and away from acute intervention, which means treating after symptoms have developed.

Dr Hannah Maude:          Yeah. It's a really interesting read, actually. I'd recommend it.

                                                So let's first give everyone some context on the scale of DNA research in the UK.

Angelos Manolia...:         Yeah. Right. So some examples, we have the UK Biobank, which contains the health records and DNA information from half a million participants.

Dr Hannah Maude:          Oh, yeah, the UK Biobank. That's an incredible resource for scientists. But if we're talking about numbers, I think probably the most impressive project is the new one called Our Future Health.

Angelos Manolia...:         Good point. Our Future Health is currently recruiting 5 million people across the UK to track connections between DNA and health outcomes. I guess the name of the project kind of gives that away, right?

Dr Hannah Maude:          Yeah, it does, Our Future Health.

1. So on the topic of current projects, there's also the newborn genome screening project, and that's going to be the topic for episode two. And that project, we'll see a hundred thousand newborn babies in the UK having their DNA tested over the next few years.
2. And this is another interesting fact. Every cell in your body has a complete copy of your genome. So for example, if you pick any cell, like a cell in your muscle, it has the full copy of the whole 3 billion letters. So the entire instruction manual to make you a human is in the muscle, but only the instructions for how to make the muscle are switched "on." Everything else is switched "off."

                                                So I hope that we've given everyone some motivation for sticking around and learning more. But since this is episode one, let's rewind and talk about DNA.

Angelos Manolia...:         Yes. Okay. So let's start with the basics. DNA is, like I said before, effectively the language that creates all life. I guess we can think of it as a unique category of language. We have natural languages like English, we have Greek, being two examples, and then we have programming languages like R and Python, and of course we have DNA, the language of life.

Dr Hannah Maude:          Oh, yeah. I like to say that I speak English, R, and Python.

Angelos Manolia...:         Oh, yeah. Well you do speak Bash, right?

Dr Hannah Maude:          Yeah, it's in Bash and I'm going to be learning Julia. That's a new one.

Angelos Manolia...:         Oh, that's nice.

Dr Hannah Maude:          Yeah. Yeah. So I like to say that because I wasn't very good at languages at school, so I wouldn't be very useful as a human translator. But I can talk to computers.

Angelos Manolia...:         Well, I guess you can write to computers. You're typing your code, right?

Dr Hannah Maude:          Yeah.

Angelos Manolia...:         I don't think you're just actually speaking your code to a computer.

Dr Hannah Maude:          No, that's true.

Angelos Manolia...:         Okay.

Dr Hannah Maude:          Let's go back to DNA. So the basic building blocks of DNA are just four letters, A, C, T, and G. The English alphabet has 26 letters, right? So a four letter alphabet, which encodes literal human beings, well, technically every living creature.

Angelos Manolia...:         Yeah, that's right, and it evolves like this if you think about it. And this complete DNA of an individual is called their genome. If you're not familiar with the term genome, it comes from the word gene, which originates from the Greek word, génnisi, meaning birth. And your genome is made up from two matching genomes, actually half genomes, one from your father and one from your mother.

Dr Hannah Maude:          So say that again, the genome.

Angelos Manolia...:         Oh, [foreign language 00:15:56]. Yeah.

Dr Hannah Maude:          [foreign language 00:15:58].

Angelos Manolia...:         It means birth.

Dr Hannah Maude:          So instead of genome, should we say yenome?

Angelos Manolia...:         Well, it would be more accurate in Greek, but I just can't say genome. No, that sounds so wrong.

Dr Hannah Maude:          It does a bit, doesn't it?

Angelos Manolia...:         Yeah.

Dr Hannah Maude:          We'll stick with genome for now.

Angelos Manolia...:         Although the Greek word for genome is gonidíoma.

Dr Hannah Maude:          I'm not going to try that one.

Angelos Manolia...:         Nah. But it has the -ome suffix in both languages.

Dr Hannah Maude:          I'm learning a lot in this episode.

Angelos Manolia...:         Yeah, you're learning both Greek, Julia, Bash, R, Python, and English. You're going to be a linguist.

Dr Hannah Maude:          Wow. Cool. So we all have our own genomes, not genomes, genomes today, which you can think of as your personal instruction manual. So I have my genome. You Angelos, have your genome. And everyone listening, you have your genome.

Angelos Manolia...:         Yeah, I do actually have a genome. And fun fact, the human genome is 3 billion letters long, and if you printed that in size 12 font, it would stretch from here, London, to maybe beyond Athens where I'm from.

Dr Hannah Maude:          It's massive, isn't it?

Angelos Manolia...:         I know.

Dr Hannah Maude:          Actually my new word for today is humongous.

Angelos Manolia...:         Right? Very good.

Dr Hannah Maude:          The human genome is humongous.

Angelos Manolia...:         But can you say that in Python?

Dr Hannah Maude:          Maybe. I'll have to figure that one out.

Angelos Manolia...:         Yeah, that's right. And since you mentioned muscle, it would be an another fun fact you see, first episode of DNA& is fun facts one after another.

Dr Hannah Maude:          Fun fact time.

Angelos Manolia...:         Yeah. So the fun fact about muscle cells, that they have a lot of nuclei in the actual cells. So every nucleus has a copy of the genome, so they actually have a lot of genomes, not just one, right?

Gareth Mitchell:               That is such a good explanation. Angelos and Hannah there on the DNA& podcast. You can find it in our podcast directory or just search for DNA& where "and," by the way, is the ampersand symbol. So that's DNA&.

                                                Well, that'll do us for this month. There's more where this came from. Yes, we will be back in August. Until then, have a lovely few weeks, won't you? For me, Gareth Mitchell, it's bye bye for now.