AI hears hidden X factor in zebra finch love songs

Nick Petrić Howe

Welcome back to the Nature Podcast, this week: harnessing the computing power of the strange skyrmion…

Benjamin Thompson

…and identifying the X-factor hidden within Zebra finch songs. I’m Benjamin Thompson.

Nick Petrić Howe

And I’m Nick Petrić Howe.

<music>

Nick Petrić Howe

This week in Nature, researchers have shown a new way to harness a weird quasiparticle, known as a magnetic skyrmion. Potentially paving the way for the next generation of computing.

Anjan Soumyanarayanan

There is a very pressing need to move beyond conventional computing architectures to avoid an energy crisis. And skyrmion is one of the strong candidates to realize these kinds of technologies.

Nick Petrić Howe

This is Anjan Soumyanarayanan, one of the researchers behind the new work. As society uses more and more technology to power things like AI, there’s an increasing need for energy. One of the ways researchers are trying to overcome this is by working to develop alternatives to the standard way of computing. For example, brain-inspired computing could be much more energy efficient. Your brain for example, uses orders of magnitude less energy than a computer doing similar sorts of things.

Anjan Soumyanarayanan

The brain can do using 20 watts, the kind of work that requires 1,000s of watts of a GPU.

Nick Petrić Howe

Mirroring the way the human brain moves and processes information is something that has been a goal of researchers for a while, but it’s very tricky to accomplish as brains store and process information in completely different ways than computers. One of the things that has shown promise on paper is called a skyrmion. These are essentially tiny whirlpools of magnetic spin that behave kind of like particles and so are known as quasiparticles. They have some properties that make them attractive to researchers interested in brain-inspired computing — they could, for example, help researchers go beyond the 1s and 0s that make up classic computer information.

Also, because they’re magnetic they could be an alternative to some of the magnetic computing technologies we currently use — like hard disk drives that use magnets to write 1s and 0s onto disks.

In order to take advantage of this, researchers need to come up with a way to easily ‘talk’ to the skyrmions, in a similar way to how we access hard disk data in the form of ones and zeros. Currently, researchers use specialised microscopes in the lab to observe these quasiparticles, but this is kind of impractical for computing use.

Anjan Soumyanarayanan

In a real device, you need to talk to the skyrmion electrically, so you need to check using currents or voltages that your skyrmion is there, you know, your skyrmion has moved and so on. So this has been the main roadblock in terms of taking skyrmions from an idea to a viable technology is the absence of a device which can detect a single skyrmion and tell you what where that's going to be on is or what it is doing, whether it's there or not, you know, if it is there, how big it is, and also to then be able to write and delete that skyrmion in an efficient manner.

Nick Petrić Howe

But Anjan and the team reckon they’ve found a way to talk to these skyrmions electrically.

To do it, they tweaked a device known as a magnetic tunnel junction. Now, regular magnetic tunnel junctions are already used in conventional computing in some types of RAM. But Anjan and the team’s version is made of the right combinations of materials to enable them to consistently and stably generate skyrmions of different sizes.

And by applying a voltage to their tunnel they could ‘talk’ to their skyrmions. Through this they could control the resistance in the junction and use that to mirror the ones and zeros found in normal computer components. Overall, they were able to write, read and delete data using skyrmions.

Anjan Soumyanarayanan

The achievement of read, write and delete within a commercially compatible device using commercially compatible techniques is kind of the main achievement of our work and there are a couple of added bonuses as well. So, one thing is that our device can write and delete skyrmions using energies which are 1000 times lower than the state of the art.

Nick Petrić Howe

Using voltage to ‘talk’ to the skyrmions is much more energy efficient than methods that have come before, which have relied on current. The other thing that the team hopes that this will unlock is the ability to go beyond the classic 1 and 0 bits. In this paper Anjan and the team were able to make skyrmions of different sizes, and a larger skyrmion or a smaller skyrmion in the junction can allow you to get somewhere between a 1 and 0. Similarly to quantum computing, this allows you to encode more information with just one device.

Now this is all still some way off. As technology scales up, there’s a lot that we don’t understand. Anjan and the team built their magnetic tunnel junctions on a 200-millimetre silicon wafer, like you might see in a conventional semiconductor, but as they scale things up, with more and smaller versions of these magnetic tunnel devices they want to make sure that things stay robust.

Anjan Soumyanarayanan

So, if you make a giant 200 millimetre wafer full of 10s of 1000s, or hundreds of 1000s of devices, then how robust are the skyrmions? What is the variability in their properties? And also how can we scale this from, in our case, a 300 nanometer device to a smaller device, let's say 100 nanometers or 50 nanometers, and then realise, you know, skyrmions, that can actually compete with existing technologies beyond just being a proof of concept.

Nick Petrić Howe

Anjan, though, feels like this new paper is a demonstration of the viability of skyrmion technology as a way to eventually make brain-inspired computers. And though it may be a long road ahead, he’s excited about the destination.

Anjan Soumyanarayanan

At this point, I feel like it's the start of a new journey. Because I think now we have told the world that this is possible. We hope that we are starting, you know, a new page for skyrmions in terms of looking at them as viable technical, technological candidates instead of just claiming so on on paper, right? So I feel that, okay, it's the end of one chapter and the start of, hopefully a new and more exciting one.

Nick Petrić Howe

That was Anjan Soumyanarayanan from the institute of materials research and engineering and the National University of Singapore, in Singapore. To read his paper, look out for a link in the show notes.

Benjamin Thompson

Coming up, how an AI helped identify the mysterious factor that makes some male zebra finches song more alluring. Right now, though, it’s time for the Research Highlights with Dan Fox.

Dan Fox

Deep-diving, cyborg jellyfish could be used to monitor the oceans. Researchers have already demonstrated that electrodes implanted into live jellyfish could allow for control of the animal's speed and make them swim more efficiently. Now the team want to build on this success to exploit jellyfishes deep diving capabilities. To do this, they built a dome-shaped prosthetic forebody that sits on the bell of a moon jellyfish, making the creature more streamlined. The team showed that the augmented animals could swim four and a half times faster than an unenhanced jellyfish, and could also carry cargo, such as sensors, in the prosthesis. The researchers say that these low-cost enhancements could make ocean monitoring cheap and easy. Take a deep dive into that research in Bioinspiration and Biomimetics.

<music>

Dan Fox

Collective saliva testing could help to achieve universal screening for a common condition in infants. Congenital cytomegalovirus or cCMV is a common infection of newborns that can lead to hearing loss and other developmental problems, and usually goes undiagnosed. To tackle this problem, a team of researchers have developed a new cost-saving screening strategy. Healthcare workers collected saliva samples from babies before pooling eight samples together and testing them collectively. If a pool was negative, all its samples were considered negative. If a pool was positive, all samples were tested again individually. Screening was possible with only a fraction of the tests normally needed and of almost 16,000 babies tested 54 cases of cCMV were found. Most of which would have gone undetected without this universal screening approach. If you're drooling over that research, you can read it in full in Nature Medicine.

<music>

Benjamin Thompson

I'm going to start our second story today by playing you some very short clips of two male zebra finch songs that have been artificially created by researchers. All I want you to do is consider which one you think is more alluring. Here's the first one.

<zebra finch sound>

And the second one.

<zebra finch sound>

Which of the two got you hottest under the collar? Okay, well, admittedly, it's quite hard to tell them apart. In fact, as humans we can't. But female zebra finches… oh they certainly can. When researchers gave them the choice, they made a beeline for the first one. But why? Understanding why female zebra finches find one male song more attractive than another is a puzzle that's long stumped researchers. But now a team thinks they figured it out by using artificial intelligence to find the X Factor hidden in zebra finch songs. Because you see learning songs is one way for male birds to demonstrate their fitness. As Todd Roberts from UT Southwestern Medical Center in the US, one of the authors of this week's paper, explains.

Todd Roberts

We think that this characteristic for vocal learning and songbirds evolved as a fitness signal because it suggests that the animals have time to engage in this learning process when they're young. So they're not having to spend all of their time foraging for food or doing other things that are essential for survival, but that then they can produce in mating situations or in territorial defence situations when they're adults.

Benjamin Thompson

Males in about a third of all songbirds species stick with a single tune their entire lives. This includes birds like the tiny zebra finch.

Todd Roberts

Males learn a single song during the first three months of their life that they need to learn to imitate from their dad’s song, and they practice their song up to 3000 times a day, slowly modifying it in order to eventually very closely match the song of their father.

Benjamin Thompson

Which ultimately leads to songs like this.

<zebra finch sound>

But what is that hidden X Factor that makes some songs so attractive? Well, much like human words, zebra finch songs can be broken down into individual syllables, each with their own acoustic properties. But if two male zebra finches both have an equally good song that they've practised their little hearts out to learn, how does the female decide who's the fittest?

Todd Roberts

People have looked at perhaps it's the acoustic complexity of the syllables? How noisy the syllable is, or if it's the changes in frequency modulation, so handpicked acoustic features, and it doesn't necessarily reflect what the female might be tuning into. And it might be a variety of acoustic features. And so it's been very difficult for the field to pinpoint one or two features that might be important, and we wanted an unbiased way of trying to identify what features the females might care about.

Benjamin Thompson

To do that the team developed a machine learning system that listened to the calls of multiple male zebra finches’ songs and analysed the 1000s of repetitions of syllables that contained. Specifically, they compared two groups of birds, those who'd learned to imitate songs from their dads and those who weren't able to and had to improvise them. Females are known to prefer imitated songs, but it's not exactly understood why.

Todd Roberts

These improvised songs, they're usually a little bit noisier, the syllables are a little bit longer. But these are just kind of the things that we perceive.

Benjamin Thompson

So the team wanted to look at the songs as a whole and see if there was something imperceptible to our ears. And this is where the machine learning system came in. It took all the information contained within each syllable sound signature and use this to essentially place each syllable as a point on a 2d map. The team then connected the dots between each point and found that the distance between syllables for the birds who learnt their songs through imitation was statistically larger compared to those who hadn't.

Todd Roberts

This led to us generating a hypothesis that maybe it is the statistical spread of the songs in this two-dimensional space that could be something that is important. What we wanted to do next was take only learned syllables that a bird produced by imitation, and then see whether or not if we made synthetic songs, using only the learned syllables and made ones that were either specifically longer or shorter in this two-dimensional space. Whether or not that would be something that the female birds cared about.

Benjamin Thompson

Which brings us back to…

<zebra finch sound>

…which has more of this statistical gap and…

<zebra finch sound>

…which has less of this statistical gap.

Todd Roberts

Once we started generating some of these synthetic songs, and we're playing them back to people in the lab that have a lot of experience listening to songs, is that we couldn't tell the difference at all. They all sounded like just a regular learn zebra finch song, to our ear.

Benjamin Thompson

But not to the female zebra finches. To test whether they preferred one over the other, the team put female birds into a T maze, which, as the name suggests, is a maze in the shape of a T, where the bird stands at the bottom and must decide which arm of the T to head towards. A speaker at the end of one arm of the maze played a synthetic song made by sticking together syllables with a long statistical difference between them. While a speaker in the other arm played a song with a short distance. The birds showed a clear preference.

Todd Roberts

In 100% of the instances when we did this experiment, the female birds spent more time in the arm with the long-distance song.

Benjamin Thompson

So it appeared that this difference the machine learning system had uncovered was involved in female preference. But how could it be an indicator of fitness? Well, it turned out that learning a song with this property is hard. The team took males with different statistical differences in their songs, got them to breed and teach their songs to their sons.

Todd Roberts

And what we found was that the sons of dads that had shorter distance songs, those birds were able to sing songs as adults that very closely matched the distance of their dads song and some instances exceeded the distance. While the young birds of dads that had longer distance songs, those birds, they weren't able to readily match their dad's long-distance song, suggesting to us that this is a difficult thing for them to achieve.

Benjamin Thompson

The team also suggest that because it's difficult, it's special. And a bird has to be fit to learn it, meaning this computer identified X Factor could be an important part of mate selection. Nicole Crianza from Vanderbilt University in the US studies how behaviours that are learned such as birdsong, affects things like evolution. She's co-written a News and Views article about the new paper for Nature and was impressed by the method the team developed.

Nicole Creanza

I think this paper gave the birdsong field a new computational lens into how to do large scale analyses of songs in a way that takes aspects of songs that had been traditionally very hard to pull apart from one another and gave kind of a roadmap for how to plot those and visualize them and think about them.

Benjamin Thompson

But a question that I'm sure has come to your mind is, well, what exactly is this X Factor? The answer is, it's not exactly clear. This method takes all the info contained within a syllable and transposes it onto a 2d map. Songs with points that are statistically further apart, are harder to learn and more attractive to females. But what does statistically further apart mean sonically? It's not abundantly clear, but it's definitely something that's intangible to humans.

Nicole Creanza

But that doesn't mean that it's not in there, quantitatively, but I don't know what to look for yet. And so thinking about kind of sonically, acoustically what are the differences that we can point to that give us information about what the female is actually listening for? What's the juvenile birds actually trying to learn? I think that's the main next step.

Benjamin Thompson

Answering this question is also key for Todd. In this work, he and his colleagues showed that seemingly all the audio properties commonly used to study birds played a role in statistical distance, suggesting that whatever this X Factor is, it's likely a combination of sonic or melodic characteristics that humans just can't pick up on. And there are other questions to answer to. For example, the team showed that this intangible thing whatever it is, is important in female preference, but it's unknown whether it ultimately results in breeding. Also, is it relevant to other birds? There's a lot to understand, but Todd hopes that their machine-learning method — which looks at songs as a whole, rather than focusing on particular characteristics that humans can hear — could help researchers understand how seemingly simple songs that actually hold a wealth of information evolved.

Todd Roberts

We know that sexual selection is a powerful driving force in the evolution of complicated behaviours. But it's been a mystery for a very long time how sexual selection may have been able to drive the evolution of more simple vocalizations. Now, how do you know up to a third of this important groups of animals, how did they evolve the types of songs that they evolved? And now we think that we have first real clue of how this may have worked.

<zebra finch sound>

Benjamin Thompson

Todd Roberts from UT Southwestern Medical Centre in the US there. You also heard from Nicole Creanza from Vanderbilt University also in the US. To read Todd’s paper, and Nicole’s News and Views article head over to the show notes for some links.

Nick Petrić Howe

Finally on the show, it’s time for the Briefing Chat where we discuss a couple of articles that have been featured in the Nature Briefing. Ben, what have you been reading this week?

Benjamin Thompson

Well, I've got a story that I've been reading about in Nature. And actually, it's another bird related story, but very, very different to what we've just heard. Now, I've talked about the plight of Antarctic penguins on the pod before, you know, related to climate change and things like that. And there's something else we can add to the list of things that potentially could be threatening them. And it's a deadly strain of bird flu that's been circulating worldwide, and now has been found in the Antarctic region too.

Nick Petrić Howe

Oh, crikey. I mean, obviously, sounds like bad news for the penguins. But I guess my first thought is Antarctica is quite far away from most other places. How has bird flu even gotten that far?

Benjamin Thompson

Yeah, that's a good question. And in particular, we're talking about the H5N1 subtype of avian influenza. Okay, now, this has been spreading around the globe for a while now and it's caused a huge number of deaths in wild birds. And the answer to your question about how it got to Antarctica, because of course you're right it is a very, very remote place, it seems like it's birds coming from South America migratory birds potentially spreading it. Now, the virus was first detected in dead birds in the wider Antarctic region in October 2023. Okay, now, these were scavenging seabirds called skua and some gulls as well. Okay, now, this was a sub-Antarctic region, so just outside kind of the main landmass of Antarctica itself. And sadly, this year, the virus is known to spread into other animals too. Elephant and fur seals, albatrosses, you know, gentoo penguins and king penguins. Okay, and was real concern that the virus, you know, would make its way to Antarctica itself, where there are of course a huge number of penguins, and a huge number of penguin species, all potentially vulnerable.

Nick Petrić Howe

Right, okay, well obviously, if this affects all the penguins in Antarctica, that would probably be quite bad news. So, what's been done to kind of prevent this spread?

Benjamin Thompson

Well, a lot of this article talks about how research projects have been, you know, changed or curtailed, or, you know, kind of stopped completely, for the time being, to try and stop the inadvertent spread of the virus between different animal colonies and also to protect the researchers themselves. For example, the Spanish Antarctic research programme was revised for the summer season, which usually runs from October to late March. So you know, kind of around now, only allowing researchers who specialize in infectious diseases and viruses into areas where animal colonies are, for example. And this is a knock-on effect to a lot of other researchers who maybe can't go and pick up their probes, for example, and are concerned that the batteries, you know, might run out, so they lose a lot of data on animal behaviour and the impacts the animals have on the environment. But other countries are following suit as well. Apparently, Argentine researchers have suspended all activities that had direct contact with animals as well. And this is going to affect researchers studying things like you know, behaviour and reproduction and so forth.

Nick Petrić Howe

And so, when you’re reading this article, what's the sort of sense you're getting? Will these measures be enough to sort of stop the spread?

Benjamin Thompson

Sadly, potentially, not. So earlier this year, there was some more dead skua on Antarctica proper, which tested positive for the virus. And just yesterday, as we're recording this, a preprint, was published — so it's not peer reviewed, so bear that in mind — suggesting that researchers have detected this highly pathogenic H5N1 in Adélie penguins, so potentially, you know, really efforts need to be put into working out, you know, how it spread and– and how far it might travel and what impact it might have. Because this virus has had a massive impact on animals across the world, millions of birds have died, and millions more had to be culled as well as a result. But also think about the seals as well, there have been some mammals that have been affected by H5N1, which is something that that really needs to be monitored very closely. And the spread of this virus is huge, I was reading today about a polar bear in Alaska, in the Arctic Circle. So literally the other end of the world that had died as a result of an H5N1. So there really is a lot to keep an eye on with this story.

Nick Petrić Howe

Well, hopefully something can be figured out because I think these animals are already quite vulnerable. So having another pressure on them probably isn't good news. And my story is like sort of kind of related to that in a way that, you know, poultry, often is a way that bird flu is spread. And this is a way to maybe avoid using poultry altogether. So, this is a story that I was reading about in Scientific American about how maybe we should be farming snakes.

Benjamin Thompson

I've heard about snakes being farmed to harvest venom to make, you know, anti-venoms and stuff, but this is farming for food?

Nick Petrić Howe

Yeah, yes for food. So, I think it's fair to say that meat has quite a big environmental impact. Now the caveat to that is it depends what kind of meat. Replace all your beef with chicken and you're going to be doing less damage to the environment. And these researchers have proposed that maybe actually, we should be looking to snakes for meat instead as… ‘cause they reckon — and they've done a paper to sort of analyse this — that farming snakes would be a lot less damaging to the environment.

Benjamin Thompson

Right because I guess there's a lot of chat about, you know, moving to an insect-based diet, because the proteins to resource kind of differential, is much more favourable because they're small and you can get a lot of protein, is that kind of the same thing with snakes?

Nick Petrić Howe

Yeah, that's kind of the same thing. Because like insects, snakes can't regulate their own body temperature, like you and I can. And when you're farming something that's a big waste of energy, those animals are using their energy to keep themselves warm, how dare they? Whereas snakes don't do any of that, which makes them much more efficient at turning their food into meat. So, one of the stats that they came up with in this paper was that to get a cow ready for eating, to get beef, you need to give the cow 10 times as much food by weight as the cow itself weighs. Whereas for a snake, you only need 1.2 times as much food. So, using a lot less resources to farm the snake in the first place. And a lot of it has to do with them just being a lot more efficient, and also not wasting their energy heating themselves.

Benjamin Thompson

And what sort of snakes in particular, you know, could this hypothetically be used for?

Nick Petrić Howe

So in the paper, they looked at reticulated pythons and burmese pythons, so you're right, not every kind of snake would be useful for this, some snakes are just very small and wouldn't give you very much meat, and it probably wouldn't be worth your time farming them. These are quite big snakes and they're snakes that people already eat in many parts of the world. So, in those parts of the world, people may be more accepting of them as well, because a big question mark about this would be whether people would want to eat snakes, and maybe in parts of the world where you and I live, people are less familiar with that, so they might not be as keen on that. But in many parts of the world it’s pretty normal.

Benjamin Thompson

And as you say, like this isn't without precedent around the world. But is this just a kind of hypothetical or a thought experiment? Or do you think this is something that the researchers actually think, is potentially a real goer?

Nick Petrić Howe

The researchers propose it but there are some things to take into consideration. So, some of the researchers were interviewed for this article and some other articles that I read about this. And one of the key things is that we don't really know a lot of the wider environmental impacts of snake farming. So, things like cows and chickens and stuff, there's been a lot of research done to understand what the impacts are of those kinds of farming, but for snakes, we don't really know. And one key thing may be what you're feeding the snakes. So, snakes are carnivores and so they're eating meat. And that meat that they're eating has come from animals, which have eaten plants. So that is something that wasn't really captured by this current analysis. And so could make the environmental impacts a lot larger when you take that into consideration. Another thing to think about as well, is that humans don't tend to eat predators very much, because, you know, they accumulate a lot of toxins, they accumulate a lot of things because they eat lots of animals, which have eaten lots of things dadada things build up through those sort of trophic levels and end up being not great for you. So, it's unclear exactly how that would play out as well. And as you scale something up for like wide-scale farming as well, it sort of changes the maps a bit, because at the moment, these snakes are being fed sort of waste meat that wouldn't be used for anything else anyway. But if you're trying to do this on a big scale, you may have to produce the meat to give to the snakes, which then may have a lot more environmental impact. So, there are a lot of questions to answer with this one. But it's an interesting idea and, you know, it sort of highlights the different ways we need to think about how we need to eat in order to make sure everyone has the correct nutrition without destroying the planet.

Benjamin Thompson

I can't believe you brushed over the word scales there Nick with barely a look to the camera.

Nick Petrić Howe

I didn’t even twig that one at all.

Benjamin Thompson

Anyway, let's leave it there for this week's Briefing Chat. Fascinating story that one no doubt. And listeners, for more on these stories check out the show notes for some links, and for a link where you can sign up to the Nature Briefing to get more stories like them delivered directly straight to your inbox.

Nick Petrić Howe

And speaking of inboxes, if you have anything that you want to tell us or let us know about the show, please do. Send your comments to @naturepodcast on X or to podcast@nature.com I'm Nick Petrić Howe.

Benjamin Thompson

And I'm Benjamin Thompson. Thanks for listening.