Host: Nick Petrić Howe
Welcome back to the Nature Podcast. This week, how to make a diet that works for people and the planet.
Host: Benjamin Thompson
And an AI that helps guide mathematicians. I’m Benjamin Thompson.
Host: Nick Petrić Howe
And I’m Nick Petrić Howe.
[Jingle]
Interviewer: Nick Petrić Howe
When thinking about what we as individuals can do to tackle climate change, some obvious candidates spring to mind. We could try to fly less, or drive less, or recycle. But there’s one unavoidable thing that every human does that also has an impact on the climate – we eat.
Interviewee: Gayathri Vaidyanathan
So, the food system as a whole – and that's from agricultural production through to consumption – emits about 20-30% of the global greenhouse gas emissions.
Interviewer: Nick Petrić Howe
This is reporter Gayathri Vaidyanathan, who’s written a feature in Nature this week about designing diets to be more sustainable. A lot of emissions from the food industry come from meat production. To reach our plates, animals have to eat plants, so they’re just less efficient at converting energy into food than eating plants directly. But at the same time as these environmental consequences of food, for many people, their diets are not well balanced or nutritious.
Interviewee: Gayathri Vaidyanathan
There are about 811 million people who are not getting enough calories or nutrition, and most of them are in low- and middle-income countries. But on the flip side of this, there are more than two billion people who are overweight and obese.
Interviewer: Nick Petrić Howe
It’s this dual challenge of making diets both sustainable and nutritious that Gayathri has been writing about. Alongside that, for many parts of the world, just eating less meat, for example, or something like that, is not really an option, especially when thinking about nutrition. To find out more about efforts to meet these competing demands, I called Gayathri up and started by asking her what research has been done to find diets that work for people and the planet.
Interviewee: Gayathri Vaidyanathan
One of the biggest efforts in this regard happened with a report that was released in 2019 by the EAT–Lancet Commission, and it had 37 scientists from across various disciplines and 16 nations who got together to figure out this healthy and nutritious diet. So, for example, the reference meal for males who are approximately 30 years of age and of average weight, it suggests that they eat like whole foods, so a lot of vegetables, grains, fruits, nuts, seeds, beans and legumes and smaller amounts of dairy and meat. So, fish is considered better than other meats, but the amount of red meat that it allows is just 100 grams a week. That’s a single serving and that’s half the weight of a hamburger. So, yeah, it’s quite a challenging diet, I think, for a lot of people to follow.
Interviewer: Nick Petrić Howe
But it’s not a diet that completely restricts meat eating. There is still some in there, and how much of an effect would this have on the sort of emissions from the food system if many people were to adopt this diet?
Interviewee: Gayathri Vaidyanathan
Yeah, it would have a drastic effect. None of this is by itself going to help hit that 1.5°C target of the Paris agreement, but it is one in a range of actions that need to be taken with respect to the food system so that we can help curb emissions and have a better future.
Interviewer: Nick Petrić Howe
And I, for instance, could happily go to the supermarket and buy everything quite easily that is required for such a diet, but I wonder, is this a diet that is achievable for everyone around the world? Like how much does it cost per day, for example?
Interviewee: Gayathri Vaidyanathan
There was one study in 2020 that calculated how much the most basic EAT-Lancet diet would cost, and that was US$2.84, and that’s quite affordable for a lot of, say, Americans or Europeans from the higher-income brackets. But even within those countries, for lower-income people, it could push things. And when you sort of look at the lower-income parts of the world, that’s even less affordable. That can, at times, be as much as 60% of a person’s average income. The EAT-Lancet diet seems like a great solution for some people. For others, there might have to be other approaches.
Interviewer: Nick Petrić Howe
And in this feature article, you spoke to some researchers and policymakers who are looking at sort of localised approaches. Can you tell me a little bit about some of them?
Interviewee: Gayathri Vaidyanathan
So, some of these approaches are being done in the West. So, for example, in Sweden, scientists are running this experiment at schools. The kids there get free meals up to grade nine. So, the scientists essentially used a computer algorithm to figure out a diet that would be affordable and at the same time culturally acceptable, and it set out these meal plans which essentially had less meat and more beans and vegetables, but the meals themselves were the same, and the kids ended up eating it and couldn’t really tell the difference. And then in Baltimore, there were some low-income residents during COVID who were given these diets that were modelled after the EAT-Lancet reference diet. So, when researchers went and surveyed the people, the people who did respond said they either loved or liked it, so that suggests that maybe these diets are not unpalatable or anything, so there might be some cultural acceptance with time.
Interviewer: Nick Petrić Howe
And has there been much work to find these diets that work for people and the planet in lower- and middle-income countries as well?
Interviewee: Gayathri Vaidyanathan
Yeah, but in a very different manner. So, scientist in lower- and middle-income countries, they definitely still have nutrition topmost on their minds. So, it’s not necessarily a top-down modelling approach but more of scientists going into communities and trying to learn how diets are entangled into peoples’ lives, their livelihoods. In Kenya, there are scientists who are looking at improving fish catch and at the same time communicating to parents the importance of feeding their kids fish, which is important because there’s a lot of stunting and the kids don’t eat as much fish even though it’s available due to just income reasons. People would find it more profitable to sell. They cannot afford to keep some fish for themselves. And of course, if they could afford to give more of that to the kids, people would do it. Looking at how to improve income and have that sort of feed into diets, so these are more holistic approaches that consider sustainability but at the same time ensure that nutrition is what is directly being addressed.
Interviewer: Nick Petrić Howe
And so, to ask you what is perhaps an unfair question, how do we develop a diet for the world, for everyone in the world to have a nutritious but also sustainable diet?
Interviewee: Gayathri Vaidyanathan
Yeah, I think that’s probably above my paygrade. The scientists that I spoke to said it’s great that we have this very top-down EAT-Lancet idea of a diet for the world, but now it’s time to look at very individualised communities and places and do this at the local level, so even if it is going to be a top-down modelling approach, use data from the local levels. So, for example, availability in one part of India would be very different from availability in another part. So, that’s just even within the same country, so at some point in time, this has to be taken down to the ground level and figured out.
Interviewer: Nick Petrić Howe
That was Gayathri Vaidyanathan, a science and environment reporter based in Bangalore, India. To find out more about how diets can help the planet, check out the feature in the show notes.
Host: Benjamin Thompson
Coming up, we’ll be hearing about an AI that helps mathematicians discover new conjectures. Right now, though, it’s time for the Research Highlights, read by Dan Fox.
[Jingle]
Dan Fox
Unlike most animals, jellyfish have no centralised brain, so how they manage to carry out essential tasks such as feeding, navigating and escaping predators has been a bit of a mystery. To understand how these creatures survive, researchers genetically modified the neurons of a transparent jellyfish, Clytia hemisphaerica, so that they glow under fluorescent light when they are activated. The labelled neurons revealed that the neural network beneath the jellyfish’s gelatinous ‘umbrellas’ had a surprising level of structural organisation. Pockets of neurons can control certain behaviours. For example, one group allowed the jellyfish to transfer food from their tentacles to their mouth. The researchers suggest that their transgenic jellyfish could serve as a model for studying how brains and nervous systems evolved. Put your centralised brain to good use by reading that research in full in Cell.
[Jingle]
Dan Fox
Researchers have uncovered a mysterious runic inscription from a 700-year-old Norwegian talisman known as the Bispegata amulet. When archaeologists found the rectangular, metal object during an excavation in 2018, they saw that it was covered with runes and folded several times. But they feared that manually opening the talisman would damage it and, because it’s made out of lead, X-raying the object wouldn’t work either. Instead, the researchers used a neutron beam to peek inside and create a detailed reconstruction. They found that some of the runes spell out Latin and Greek phrases, while others signify repetitive sequences of seemingly meaningless words. Some of the comprehensible phrases might carry religious meaning, whereas the others were probably thought to have a magic effect. You don’t need a neutron beam to read the rest of that research in Archaeometry.
[Jingle]
Interviewer: Benjamin Thompson
This week in Nature, there’s a paper out describing a machine-learning AI developed by the company DeepMind. This AI has been designed to assist mathematicians with their research and to get a sense of what it is and how it works, I called up Davide Castelvecchi, who, among other things, covers mathematics here at Nature. Davide, great to speak to you. How are you doing today?
Interviewee: Davide Castelvecchi
Great. how are you?
Interviewer: Benjamin Thompson
Yeah, I’m doing okay, thank you very much. We’re here today to talk about maths and a new paper coming out that’s getting an AI involved in helping mathematicians do their work. But before we get into what that is, maybe let’s zoom out a little bit, and there’s certain types of maths that are very much involved in looking at patterns between two things.
Interviewee: Davide Castelvecchi
Indeed, and this is actually one of the kinds of mathematical research that mathematicians consider most profound. It’s when you have two different kinds of objects that maybe arise in two completely different branches of mathematics and you find hidden connections, so some kind of dictionary that translates from one kind of object to the other.
Interviewer: Benjamin Thompson
And trying to find that dictionary then, I mean how is that traditionally done?
Interviewee: Davide Castelvecchi
Well, traditionally, mathematicians have recognised patterns just by looking at a lot of examples with maybe a lot of calculations, or drawing pictures, looking at tables. And maybe more recently, in some cases, they started using computers do calculations, for example, numerical properties of certain geometrical objects.
Interviewer: Benjamin Thompson
And in this new paper that’s come out then in Nature, the sort of computer aspect has been taken a little bit further and, in this case, a team at DeepMind have been using an AI to kind of help look for these patterns then between these two different things to try and find this dictionary. Broadly speaking, Davide, how does it work and what does it do?
Interviewee: Davide Castelvecchi
This approach is intended to guide the mathematician’s intuition. So, mathematicians that do a lot of calculations with a lot of special cases and identify patterns that linked these kind of two different mathematical worlds and then they could formulate some kind of general principle that then they would try to rigorously investigate. But traditionally, it was a mathematician’s own ability to spot the patterns that was crucial. And here, the computer scientists are saying, ‘Well, we have this technology that is able to spot patterns that maybe humans couldn’t see before, and so if you let the computer find these possible patterns, even if it’s just a guess, it can guide mathematical research and then the mathematician, the human, can take over from there.’
Interviewer: Benjamin Thompson
So, it points them in the right direction, saying, ‘Hey, there is a link between these two things. Now, go ahead and use your mathematical skills to try and figure out what that is.’
Interviewee: Davide Castelvecchi
Exactly, and this is something that the authors of this paper have emphasised, that their work basically consisted of a lot of back and forth and interaction between the mathematicians and the computer scientists at DeepMind who sort of tailored their AI algorithms to the problems that the mathematicians were looking at.
Interviewer: Benjamin Thompson
And what sort of problems were those, Davide?
Interviewee: Davide Castelvecchi
So, for this paper, they worked with three mathematicians, two of them who are experts in the theory of knots, and another one, who studies symmetries and a field called representation theory, who worked on a longstanding open question looking for connections between two approaches to symmetries.
Interviewer: Benjamin Thompson
And so, did the AI point the mathematicians to anything new or interesting to look at that they hadn’t considered before?
Interviewee: Davide Castelvecchi
Yeah, so in the case of representation theory of the symmetries, the algorithms did find an interesting pattern, and the mathematician thought it was so compelling that he formulated what’s called a conjecture. So, that’s a statement that appears to be true in all the special cases that a mathematician can think of or that a computer can check but which hasn’t been proven rigorously in all generality yet. In the case of the knot theory, the AI discovered a pattern which led to a conjecture, and the mathematicians were actually able to prove this conjecture, so it is now what’s called a theorem.
Interviewer: Benjamin Thompson
So, it seems that in this case then, this AI is helping to point mathematicians in a direction where there could be some interesting maths to think about. What’s the broader community saying about this when you’ve spoken to them?
Interviewee: Davide Castelvecchi
So, people are, I would say, cautiously open-minded. They also point out that this particular way of using AI doesn’t apply to all kinds of mathematical problems because, first of all, you need problems where you can calculate a lot of special cases before you can identify a pattern. So, you have to be able to feed a lot of data into the algorithm before the algorithm can find a pattern, and that’s not necessarily true in a lot of mathematical problems. Also, a lot of mathematical problems don’t necessarily involve finding patterns between two different classes of objects. A lot of mathematical questions are not of that kind. But that said, there are a lot of important mathematical questions that could arise this way.
Interviewer: Benjamin Thompson
Well, Davide, longer-term listeners to this podcast will recall you being on the show at the start of the year talking about a thing called the Ramanujan machine, which is another different sort of AI that was helping mathematicians with their work, and now we’ve got these two systems that work on very different sort of branches of maths. But do you think that we’re maybe at a stage now where AI becomes very much entrenched in the mathematical process?
Interviewee: Davide Castelvecchi
Oh, that’s difficult to tell. I’m guessing that most mathematicians, at least for a long time, will continue to work in the traditional way. Some may also start using AI, not to come up with conjectures but to check their proofs. Once a mathematician writes a paper and claims to have some kind of rigorous sequence of deductions, it can be helpful to have a computer check that there’s no human error in there.
Interviewer: Benjamin Thompson
Well, where do you think it goes next? What have the researchers been telling you about?
Interviewee: Davide Castelvecchi
The mathematicians who were involved in this paper, they’re still using the technique. This is ongoing work where it could yet lead to more results. In principle, the tool are tools that other mathematicians can use and the computational resources, compared to other applications of deep learning where you need the supercomputers that only large corporations have, this one is relatively modest in its requirements. So, for example, the co-authors of this paper are now able to run these AI models on their laptops.
Interviewer: Benjamin Thompson
Finally, Davide, if this system then is helping to point mathematicians in the right direction, why does the system not just do all the maths itself? I’m guessing it can calculate a lot faster than a human can, so why is it not just a one stop shop that goes all the way from start to finish, do you think?
Interviewee: Davide Castelvecchi
That is a great question. Deep learning and machine learning in general are techniques that are inherently statistical. They’re inherently about guesses. So, all these algorithms can do is make guesses that have a reasonable probability of being true, and that means they’re very useful for finding patterns that could lead to actual mathematical laws. That doesn’t mean that the laws are necessarily true. They could just be coincidences. And so, it still takes a mathematician, using the rigorous techniques of mathematics, to prove that such a mathematical law holds.
Interviewer: Benjamin Thompson
Nature’s Davide Castelvecchi there. Head over to the show notes where you can find links to the paper and a News and Views article about the work.
Host: Nick Petrić Howe
Finally on the show, it’s time for the Briefing chat, where we discuss some of the stories that have been highlighted in the Nature Briefing. Ben, what have you found for us to talk about this time?
Host: Benjamin Thompson
Well, Nick, you know we love to talk about mammoths on the Briefing chat, and I’ve certainly got a mammoth-adjacent story for you today, and this was reported in Nature. And in this case, it’s the discovery of what could be the oldest known example of decorated jewellery in Eurasia made by humans, according to archaeologists.
Host: Nick Petrić Howe
Okay, so you said this is mammoth-adjacent, so is this some jewellery made from mammoth bones, is it something like that, or mammoth teeth? How does the mammoth come into this?
Host: Benjamin Thompson
Well, Nick, you’re certainly on the right track, but in this case we’re talking tusks, and this is a pendant carved from a woolly mammoth tusk that was found in a cave in southern Poland.
Host: Nick Petrić Howe
And you said this may be the oldest jewellery found in Eurasia. Exactly how old is it?
Host: Benjamin Thompson
Well, in this case, radiocarbon dating has put the piece of mammoth tusk used to make this pendant between 41,730 and 41,340 years old.
Host: Nick Petrić Howe
Wow, so this is really ancient then. What does it look like?
Host: Benjamin Thompson
Well, yeah, I mean, that’s a great question. So, in this case, it’s oval-shaped and it’s kind of smooth round the edges, about 4.5 centimetres wide, and it’s got two holes that have been sort of drilled into it, and it’s got 50 kind of small puncture marks that are in this kind of curved loop. Now, what these puncture marks mean is an open question. I think we’ve covered on the podcast before the origins of numbers and counting and so forth, so there are speculations that these marks could be some sort of counting system or could be for lunar observation, maybe used a hunting tally. But it’s a wonderful thing to look at. And that wasn’t the only thing that was found in this cave as well. Other things have been found there too. For example, a pointed tool was found, used for making holes, and that was shaped from a piece of horse bone. Now, that’s not necessarily used to make the holes in this particular pendant, but it shows that there was a lot going on in this cave at that sort of time.
Host: Nick Petrić Howe
Wow, so potentially this tells us a lot about the people who made it, but does it tell us anything about the mammoth itself?
Host: Benjamin Thompson
Well, what the researchers are suggesting, Nick, is that the tusk probably wasn’t a lot older than the time that the decorations were made because a very, very old tusk would have been hard to shape and difficult to work, so probably not too much a difference in time there. But of course, with these old objects, there’s always a little bit of controversy, Nick. So, in this case, the researchers are saying that this piece of jewellery was thousands of years older than similarly decorated objects from other sites with these kind of dotted patterns, but some researchers are saying, ‘Well, you’ve only compared to other objects with dotted patterns, not to other things that could be jewellery.’ So, whether it is the oldest piece of jewellery is unclear, but it does maybe give some more information about where Homo sapiens were and at what particular time in Europe and maybe about the understanding of these kinds of symbolic behaviours about putting markings on things as well, which is obviously super, super interesting stuff.
Host: Nick Petrić Howe
Well, we know we love a mammoth story on the Briefing, so thanks for bringing that one to me, Ben. And for my story this week, I’m bringing us right up to the modern day and also a little bit into the future because I’ve been finding out about lab-grown fish.
Host: Benjamin Thompson
Well, Nick, you put a big of weighting on the phrase ‘lab-grown’ there, so presumably you’re not meaning when two fish mate and then a baby fish grows into an adult fish.
Host: Nick Petrić Howe
No, this is a story that I found in Nature Biotechnology, and I guess a better way to put it would be ‘cell-grown’ fish. So, this is from the harvesting of cells that have been produced in a lab that are fish cells that can be used to produce meat.
Host: Benjamin Thompson
Right, and why is this being done then, Nick?
Host: Nick Petrić Howe
Well, this relates back to the story that I was talking about earlier in the show, about us trying to be more sustainable with our diets, and this is another way in which people can do that. They could, perhaps, instead of growing animals on land and them eating plants and harvesting from the animals, instead perhaps we could skip some of those steps, take some cells from animals and grow them in a lab, and that would make it a more sustainable thing. And for fish in particular, there’s a lot of pressures on the ocean – there’s overfishing which can reduce biodiversity and things like that and also pollution and stuff going on – so it might be good to have an alternative source of protein and meat that comes from a more sustainable source.
Host: Benjamin Thompson
Right, and so is this just growing fish protein then because I know that people have been trying to do lab-grown meat for quite a long time. I guess this seems similar to that?
Host: Nick Petrić Howe
It is very similar to that. So, the principle is the same. You want to produce cells that become muscle cells and fat cells – the components of meat – and grow them in the lab to make some sort of food product. Now, the difficulty is, if you just have cells, it’s going to be kind of a mush, so what researchers have been doing with lab meat and with this lab-grown fish is they’ve been using like scaffolds to make them into more sort of the things that we’re used to when we think of meat, so they are in the right shape and of a similar sort of texture and things like that. And that’s sort of where lab-grown fish is at the moment, but it’s got a few other difficulties that aren’t present for lab-grown meat.
Host: Benjamin Thompson
Pray, tell, and what are these difficulties then?
Host: Nick Petrić Howe
Well, one of the things with sort of terrestrial lab-grown meat, if you want to call it like that, is that we actually only eat a few kinds of species. You’ve got your chickens, you’ve got lamb, you’ve got cows and that sort of thing, so there’s a limited amount of things that you could get cells from and produce their meat in a lab. Whereas with fish, there’s a whole range of species that we eat, and for each one you then have to get a cell line from them and get that cell line working in the lab, and that’s not trivial because, basically, you want to make these cell lines so they’re sort of immortal, and that means that you don’t have to keep going back to an animal to keep getting stem cells. You have got a line of cells that you can keep producing forever. And then the problem then after you’ve done that is you also have to scale it up. It’s okay to produce a little bit of meat but then getting it to the sort of vast quantities that we need as a society, that’s quite tricky.
Host: Benjamin Thompson
Well, scales are certainly important when it comes to fish, Nick. Which species have they chosen then? Is it ones that are particularly well eaten across the world or is it ones that are particularly highly nutritious or somewhere in between?
Host: Nick Petrić Howe
So, there’s a range of different fish that are in development and there is companies all over the world that have started work on this, and so the majority of them, I would say, are very, very sort of common fish, so you’ve got your tunas, you’ve got your salmon, you’ve got your carp. But you’ve also got lobster and even caviar.
Host: Benjamin Thompson
Well, goodness, I mean, that leads beautifully into my final question, Nick. How much is this going to cost and when do you think we can expect to see lab-grown fish protein on the shelves at supermarkets or in restaurants and what have you?
Host: Nick Petrić Howe
Well, it’s hard to say how much it will cost until they get to a point where they can produce it on a big scale, but we know, at the high end for lab-grown meat, it can cost US$20,000 per kilogram, so it could potentially be quite costly. But I think the idea is to make it economically feasible, and the way to do that is to get the scales up. And at the moment, only very, very small amounts of this fish has been produced, so I think it’s some way off yet, and there’s a lot of optimising that needs to be done because, unfortunately, you can’t just be like, ‘Oh, this is how this works. We’ve grown the cells. They look like meat. And then just times that by a hundred.’ Things get more complicated as you get up to those bigger scales, so it’s not just a linear thing. So, it could be some way off yet but there’s a lot of people involved in this and it seems like the field is starting to hot up.
Host: Benjamin Thompson
Well, Nick, that is a fun story. Thank you very much for bringing it this week. Listeners, if you’d like to know more about these stories, we’ll have links to them in the show notes. And if you’d like more science news delivered directly to your inbox then make sure to sign up for the Nature Briefing, and head over to the show notes as well for a link where you can do so.
Host: Nick Petrić Howe
That’s all for this week. As always, if you want to get in touch with us, you can send us an email to podcast@nature.com. Or tweet us – we’re @NaturePodcast. I’m Nick Petrić Howe.
Host: Benjamin Thompson
And I’m Benjamin Thompson. Thanks for listening.