AI takes weather forecasting by storm, and crabs use aspirin

Plus, was Venus once habitable?
06 December 2024
Presented by Chris Smith
Production by Rhys James, James Tytko.

RAIN STORM

Rain Storm

Share

In the News pod, Google DeepMind's weather forecasting AI model outperforms traditional tools. Also, new data from volcanoes on Venus dampen theories it was once a watery world, and is this double action weight loss drug the successor to Ozempic and Mounjaro? Then, we hear the proof that crustaceans can feel pain, and will seek drugs to relieve it...

In this episode

Clouds and Lightning

01:01 - Is AI model 'GenCast' the future of weather forecasting?

Researchers behind the tool say it is faster, smarter and greener than current technology...

Is AI model 'GenCast' the future of weather forecasting?
Ilan Price and Remi Lam, Google DeepMind

A new AI model named ‘GenCast’ has outperformed the best traditional medium-range weather forecast, and it is also able to better predict extreme weather. Unlike our existing weather forecasting systems, which operate on supercomputers running massive simulations of the atmosphere, burning in the process through megawatt hours of energy and undoubtedly contributing in the process to climate change, the new approach uses machine learning to spot patterns in historical weather data - basically, when conditions look like this, this is the outcome, to predict future weather patterns. And it does it with a fraction of the energy spend and in a fraction of the time. The findings have been published in the journal Nature, and Ilan Price and Remi Lam at Google DeepMind have been telling me all about their model…

Remi - The traditional way of making a weather forecast is to use a physical equation to describe how the weather in the atmosphere evolves over time. What this means is that you have to use a very large supercomputer. It's costly and time consuming. It's also error prone.

Chris - We know about the errors. Anyone who's been a victim of the weather forecast knows all about that!

Remi - What we've been trying to do at Google DeepMind is try to uncover the potential of using all of that historical weather data that we are sitting on to improve the weather forecast. By doing so, we believe we can make better weather forecasts and faster weather forecasts.

Chris - How do you do it, Ilan?

Ilan - We do it by training a machine learning model on four decades of historical weather data. The model learns weather patterns and weather dynamics directly by looking at that data, and that's what it uses to make predictions going forward.

Chris - When you say you train it, what is it actually looking at? What do you feed in? What's the input?

Ilan - What it's looking at are historical estimates of the state of the weather in the past, and basically, during training, the model is shown, okay, this is the weather state at time X, make a prediction for 12 hours time. Then it gets shown what it should have predicted and it learns from its mistakes by showing it many examples of this. What are the patterns that it should learn to pick up?

Chris - Did you focus on just one geography or were you feeding this global data?

Ilan - This is a global model. That's important because if you want to be able to predict the weather at medium range, so that's out to about 15 days in our model's case, you really need to be able to model the global atmospheric dynamics.

Chris - I was going to say, because obviously there's this old joke, isn't it there, the butterfly flaps its wings and then there's a hurricane on the other side of the Atlantic, but it really is all interconnected, isn't it? You've got to be able to consider everything. But that has previously been such an intractable problem because of scale that it hasn't been done.

Ilan - Absolutely. Chaos is the name of that phenomenon where very small things can have very large consequences. It's one of the reasons why the weather really is inherently uncertain. We actually know that we can't predict the weather exactly. One of the important features of our new model is that it's an ensemble forecast. We don't try and do the impossible and make exactly one prediction of what will happen. Instead, we make multiple predictions of what might happen and that gives us a sense of the range of different possible scenarios in the future. It lets us calculate, okay, how likely are some scenarios, how likely are other scenarios?

Chris - How much better is it, Ilan? If we compare what our weather forecasters thought was going to happen with what your model suggested was going to happen, how good is it?

Ilan - It's hard to put an exact number, one single number on how much better GenCast is because there are lots of different things that we would like from a weather model and the improvements are different on different tasks. But overall, on the headline metric, that is averaged over all times of all weather conditions over the year, for more than 97% of the evaluated targets GenCast is better, but it's also better on a lot of the specific things we care about. For example, we might care specifically about extreme weather. We can ask, how good is GenCast at predicting a once in a seven year high temperature in a given location. We evaluated that in the paper and we see GenCast improving at that. Similarly, we care about predicting the trajectories of tropical cyclones. These have devastating consequences and the more advanced warning we have the better. We were able to show that GenCast is giving us better predictions of the tracks of these storms. It's giving us about a 12 hour advantage in accuracy over state-of-the-art operational models at the moment.

Chris - One of the things that Remi said earlier was that, in order to do what we do at the moment, it takes a supercomputer to do the sorts of calculations and run these models that enable us to make the predictions we have. How much better in energy terms is doing it your way than running those supercomputers?

Ilan - I don't have a good answer to that in energy units, but I can give you a comparison that makes it quite apparent. In comparison to hours on a super computer, with tens or hundreds of thousands of processes, we're talking about making a 15 day prediction by GenCast in eight minutes, produced by only a single TPU chip. That's a chip just a bit bigger than a computer. There's really orders of magnitude difference in the amount of computation that it takes to generate a forecast with GenCast and machine learning models compared to these traditional physics based models.

Chris - What are the implications of that then, Remi, apart from the fact that you can argue we'll save a lot of energy because we won't have to run these supercomputers and we'll get the results, which potentially are more accurate, more quickly. Apart from those, what are the implications of this?

Remi - I think this is quite a pivotal point in the way we do weather forecasting. It's much faster to make predictions, and it doesn't require supercomputers. What it means to me is that it'll be more accessible to weather forecasting and to conduct research in weather forecasting. We're also making the model publicly available so people can do research on it. I think this is really going to accelerate the progress within weather forecasting, both because it makes the research accessible, doesn't require a supercomputer, but also provides a new way of improving the model, really pushing on the axis of the data rather than purely the compute axis.

Chris - Are there any risks though, Ilan, in the sense that with large language models, I know that's a different technology, one of the things to emerge that's caused some people some headaches has been things like confabulation where it just makes stuff up. Could this make up a hurricane that isn't going to happen and have everyone battening down the hatches unnecessarily? Or could it just miss a hurricane and say, well that's not a hurricane, that's going to be fine, and then we end up with people in danger because of it?

Ilan - It's a really great point and I think that there's a few important things to consider. The first is, of course, no model is perfect, no model is free from errors. As we've already discussed, that's also true of physics based models. But it does raise the point that it's really important for us to be doing rigorous and scientific evaluation of these models, both in peer reviewed research like we've done in Nature, and also that when these models are beginning to be incorporated into operational systems as they are and as we think they will be going forward given these GenCast results, that they be tested by weather forecasters, by meteorologists so that trust is built in these systems. The second really important aspect of that is the prospect of these kinds of mistakes even further highlights the importance of probabilistic models, right? We don't have to rely on a model either predicting that there's a cyclone or not. It's a question of, in how many simulations, in how many scenarios that were predicted by the model, in how many did this occur? It allows us to estimate the risk of these events and if a mistake is made in one of those predictions, then it'll only show up as a very low probability event and we don't have to worry as much.

Chris - Ilan, to finish, I've got a big birthday next year, I'm thinking of having a party. How's July looking?

Ilan - I'll have to come back to you. Go check the model.

global view of the surface of Venus

Was Venus once a watery world?

Space scientists at the University of Cambridge have, they think, answered a long-standing question about our near-neighbour, Venus:  whilst it’s a hothouse today with a surface temperature sufficient to melt lead and sulphuric acid for rain, given its similarity to Earth, was it once a lot wetter, like we are? By looking at the gases that the Venusian volcanoes release, Tereza Constantinou has been able to work out how much water is really there inside the planet. And the picture it paints is not of a once watery world…

Tereza - Earth and Venus are often thought of as sister planets. That's because they're very similar in mass, radius, density and distance from the sun. But now, they've ended up looking really, really different. You've got Earth, where we've got oceans, we've got a nice climate, it's quite comfortable. However, on Venus, the surface conditions are extreme. You've got an atmospheric pressure at 90 bar, so that is 90 times the pressure that you experience when you're standing on earth, and that is the equivalent of being about a kilometre underwater. Imagine the pressure you'd feel in your ears from that, that is extreme. Then you've got the toxic atmosphere made of carbon dioxide and you've got sulphuric acid clouds. So this is vastly different from what it would be like on Earth. However, there's been this theory suggesting that Venus was once very much like Earth: it had oceans, a cool, temperate climate, but that is really unknown. Even scientists are still debating to this day whether that was the case. That's what we were seeking out to answer, trying to see if there's any evidence in what Venus looks like today that speaks to the climate past, that tells us whether Venus ever had oceans.

Chris - We struggled to do something similar for our own planet, that we are on, that's a lot less inhospitable. So how can you get to the bottom of those questions on Venus?

Tereza - It has been very tricky so far. The way it's been done for Venus is using climate modelling. That's very much what people would use when you're checking the weather for here on Earth, but doing that for an early Venus: seeing if it was ever cool and temperate, cold enough essentially to have liquid water at its surface. But we wanted to approach the problem a slightly different way. We wanted a more direct test, something to do with what Venus is like today. That is a very clear clue as to whether the past of Venus ever had oceans.

Chris - It's almost like, you know where we are today, you're going to wind the clock back using some kind of model of how things evolve and change to work out what it would've been like back in the day?

Tereza - That's essentially what has been done so far. But we wanted to do something else. We wanted to see if there was any clue within Venus as it is today that was a signature of this past. Essentially, we wanted to see if Venus in its present day has any clues about it ever having oceans. So, for example, on Mars, just by looking at the surface, you can see where water has shaped surface features. So, much like flowing water through soil or sand will shape channels and valleys, we see that on Mars. However, Venus's surface is really young in geological terms, even though it's a few hundred million years old. But essentially there was a large volcanic event that released lots of lava, covering most of the planet's surface, so if there was any evidence of flowing of water - like valleys - it is now covered, so completely erased. We've had to come up with entirely different ways to figure this out. Different clues to look for.

Chris - And what are they? How do you do it?

Tereza - The way we did it is we looked into volcanism on Venus to find out how much water is inside the planet. Now this is a very key clue to whether Venus ever had any oceans in its past. Studying the composition of the gases being released by volcanoes on Venus, we can see how much water is inside the planet. Now imagine a volcanic eruption here on earth. If you've ever seen any photos of them, you essentially see these large billowing clouds coming out. Most of that is water, and this is directly coming from the inside of the planet where the lava comes from. We wanted to do the same for Venus by studying the position of the gases coming out of Venus's clouds. We wanted to see if there was any water, or rather how much water there is inside the planets and this was a very key signature about the climate's past.

Chris - How are you actually making those measurements though?

Tereza - It would be great if we could send a probe now and study Venus's clouds, and there are plans in the future. There are probes being sent to Venus to do that, but obviously we haven't done that yet. In fact, any probes that we sent in the past died when they reached the surface due to the extreme pressures and temperatures. So we did it slightly differently. Most of the data that we have about Venus or the thing we understand the most about Venus is its atmospheric composition. So we modelled that atmospheric composition. We studied the chemistry of the atmosphere, and we looked at different gases being destroyed in the atmosphere through chemistry, which must be restored by volcanism to maintain atmospheric stability. So we use that to kind of study what kind of gas you would need to be added into the atmosphere to sustain what it is like today.

Chris - So you can basically see the atmosphere from far away. You don't need a probe to do that, and you can work out what's in it and therefore you can work out what must be going into it and what's coming out of it. So therefore you get that dynamic of what the volcanoes must be doing and therefore the amount of water can be inferred indirectly.

Tereza - Exactly. So we've been observing Venus for decades now. We've sent missions there that took some measurements of what the atmospheric composition is, and we've also used telescopes on the ground here on Earth to observe Venus and, again, study the composition of the atmosphere. That's exactly what we've used.

Chris - What does that reveal then? Is it much wetter like the Earth is, or were we wrong?

Tereza - It is really, really dry. There is so little water coming out of the volcanoes on Venus, and this is vastly different to Earth where you've got lots of water. Most of the gases released from the volcanoes are water. But that's not what we found for Venus. In fact, there was very little water, meaning the planet itself has very little water as a whole.

Chris - How do you reconcile that? Because you mentioned when you started you said, 'look, Venus is almost a sister planet to the Earth. There are so many similarities and we know the Earth's got loads of water. It had some to start with, loads of asteroids and comets rained down on it when it was young and they put water here.' Why wasn't the same happening to Venus then?

Tereza - I think Venus ended up being too close to the sun so that it took too long to cool down and solidify into this rocky body that we now know. All the water that the planet might have had only remained as steam in the atmosphere. This gave the water enough time to be dissociated. So this meant incoming radiation from the sun was strong enough to be able to break apart molecules in the atmosphere. Water was broken up into hydrogen and oxygen. Given Venus is closer to the sun, it's got more incoming radiation, and hydrogen is a very light molecule. It escapes your atmosphere at very high rates. Now, if the planet never cooled down fast enough to condense that water and trap it within your system, the water will be exposed to the sun in the atmosphere. It would dissociate and leave the planet, leaving behind a very dry and desiccated planet.

Weight loss

15:11 - Double action weight loss drug reawakens appetite signals

Could this improve on the already impressive results from drugs like Ozempic...

Double action weight loss drug reawakens appetite signals
Randy Seeley, Michigan Nutrition Obesity Research Center

Weight-loss jabs - like Ozempic and Mounjaro - have been in the news a lot lately. They work mainly by simulating the action of one of the body’s signals called GLP-1, which shoots up when we eat. But there’s also another important signal in the form of a hormone called “leptin”, which is released by fat tissue in increasing amounts as we gain weight. And the brain uses it to regulate eating behaviour, except when we start to put on too much weight. Now the pharmaceutical company Novo Nordisk have developed a new molecule, one end of which looks like leptin, and the other end resembling GLP-1. This seems to be able to trigger a population of appetite-regulating cells in the brain, and sensitise, or even re-sensitise them to leptin, significantly boosting rates of weight loss in experimental mice. Randy Seeley is the director of the Michigan Nutrition Obesity Research Center where he’s been working on this potential new drug. He told me what he’s found…

Randy - Leptin is released from your fat tissue and it essentially tells your brain how many stored calories you have in your body. Your brain uses that information to be able to maintain your body weight, to figure out how much you should eat and how much you should burn. What's interesting is, when it was first discovered, people thought, 'Oh my gosh. This is going to be a great therapy for individuals with obesity. It turned out not to be true. The reason it wasn't true is that it turns out that individuals with obesity aren't leptin deficient. That is, they don't have low leptin levels, rather they have high leptin levels and they seem to be resistant to the ability of that leptin to be able to give them information about whether they have sufficient stored calories. So let me give you an example. If you took a lean mouse and you gave it leptin, you're essentially tricking that animal into thinking that it has more stored calories and the animal does something pretty sensible in reply, it stops eating. But when you do that in a mouse that's been made obese, it turns out it doesn't respond to leptin. So the whole trick of this is trying to understand, is there a way to still use leptin from a therapeutical perspective that might be part of the armamentarium that we use to treat individuals with obesity

Chris - In overweight situations, then, it's as though the brain has become deaf to the signal of leptin and you are asking, can we restore hearing to that part of the brain? It's like a hearing aid for the brain. Can we resensitise it so it does respond to the leptin signal which says, you have too many calories on board, you need to lose some weight.

Randy - Exactly. How do we resensitise the brain? How do we turn that hearing back on in a way so that it can listen to both their own leptin and the leptin that comes in the form of this particular molecule?

Chris - So you think that you've got a drug here which can do that. How does it work?

Randy - We identified a set of neurons in a specific part of the brain called the hypothalamus, and those neurons express both the receptor for GLP-1 and the leptin receptor. We verified that that's not just true in mice, but it was true in non-human primates as well, that there's this set of neurons. And then we used a variety of genetic tricks to either add or remove leptin receptors from that area of the brain, in these cells that also express the GLP-1 receptor. It turns out these neurons are both necessary and sufficient for the actions of this particular drug.

Chris - Why should hitting the neurons with this agent that stimulates both the leptin signal and the GLP-1 signal at the same time be an effective strategy to make the cells more sensitive to this fat signal, leptin?

Randy - You're asking a really good question. What we know is that when we add this GLP-1 component, we get a response that you wouldn't get with leptin alone. There are two ways to think about this. One is, if we prime the animal to begin losing weight with the GLP-1 side of the molecule, we now get the system cranked back up and turned back on so that it can start hearing the leptin side. It doesn't matter what you would do. You could do lots of different things and maybe they would all be effective at making leptin able to hear the signal that leptin brings. The other side of it really has to do with these particular neurons. The idea that we're hitting the GLP-1 receptor, turning on that signalling cascade in these neurons that restores their ability to be able to hear the leptin signal. That's the side that we favour. But again, it's pretty hard to prove between those two particular hypotheses.

Chris - I suppose that if you view the weight loss that would arise from this as two phases, because normally if someone loses weight, then the amount of leptin would drop. So they would lose whatever drive they were getting to lose more weight because they're losing leptin. But if you have a molecule that seems to fool the brain into thinking the level is staying high and it stays sensitive to it, you're going to reinforce the weight loss for longer. So in theory, you could actually have a much more effective pattern of weight loss.

Randy - That's what we show. One of the things that Novo cleverly did was to make versions of the molecule that only had one side or the other, but looked similar. We were able to use those kinds of research tools to be able to ask what happens when you only push on one side or the other. And the answer is, you don't get as much response as if you are able to push on both of them at the same time.

Chris - How much response do you get? We know when people use these agents, when people have done trials with the existing GLP-1 agonist, drugs like Ozempic, for example, Mounjaro, that kind of thing, we know that translates over the period of treatment to a weight loss of between 10 and 15% of body weight. So if you bring your new double acting drug to bear, what sort of levels do you think you would get with this?

Randy - I don't think we can know from this study. We didn't do the kinds of comparisons that would allow you to be able to directly compare molecules like tirzepatide or semaglutide. As you can imagine, it's not a trivial thing to be able to do. Mice are different and particularly they're different in how they metabolise a drug, that is, how long the drug lasts. And so trying to do those kinds of comparisons turns out to be really tricky. I think it's going to take some time to tell whether a molecule like this one could be as effective or more effective. We just don't know today.

A crab

21:45 - Crabs will take aspirin to relieve pain

It raises questions about the ethics of cooking live crustaceans...

Crabs will take aspirin to relieve pain
Eleftherios Kasiouras, University of Gothenburg

Anyone who’s ever felt a twinge of disquiet when they selected a lobster and watched it being dropped into a pan of boiling water should perhaps trust their instincts: because scientists have discovered crustaceans do appear to perceive pain. Eleftherios Kasiouras from the University of Gothenburg has found that crabs dabbed on their sensitive parts with strong vinegar will administer aspirin in the aftermath to ease the discomfort, suggesting they really are feeling pain…

Eleftherios - There are some criteria that need to be fulfilled so we're certain, and we can say beyond reasonable doubt, that crustaceans, they experience pain.

Chris - Pain is a difficult one though, isn't it? Because it's a perception. I can define pain, I can say I'm experiencing pain and you'll know exactly what I'm talking about. But how do we know that other animals share that visceral experience or whether it's just an automatic reflex for them and they don't experience the same emotional effect that we do?

Eleftherios - So we worked on shore crabs and we used different stimuli on the soft tissues of the body, like the legs, the claws, eyes and antenna, which are the little pointy things. We stimulated with acidic acid, vinegar, and it's painful to these animals, and we also used mechanical pressure to see how they responded to that. From these two different stimuli and the responses that we got in the nervous system, we could see that the signal transfers through the body to the brain and a response arises and we record these responses.

Chris - So you can demonstrate that they have the neurological capacity to detect a stimulus that we would regard as painful, and it changes brain activity when you put that stimulus into the nervous system. So you're two thirds of the way to showing that they are experiencing pain, how do we then clinch it to get the final tick in the box that this is then registering in their brain as an unpleasant thing in a way that we would say, well, that looks like they're feeling pain.

Eleftherios - We want to inflict pain on them and see how their behaviour will change and then, if we provide drugs such as analgesics, aspirin, if they would prefer the aspirin than the pain stimulus, or they want to go to the aspirin to relieve themselves from pain, that behaviour will tell us that they actually want to avoid the stimulus no matter what so they don't experience that. That's the last step of my PhD to see how the behaviour and learning come into play. That has not been answered yet, but I'm working on that.

Chris - So crabs can take aspirin. Really?

Eleftherios - Yeah. So far we've tried it on lobsters and it seemed to work okay. But more to come about that in the next paper that I'm trying to publish.

Chris - Summarising then, you stimulate them with something we would regard as painful. They flinch effectively. It's like me touching a hot plate. You can then show that they will actively seek out pain relief off the back of that with things like aspirin, which would suggest that it's an ongoing discomfort for them and they're alleviating it.

Eleftherios - Exactly. That's the next step for my experiments.

Chris - Well, what are the implications of this then? Because we traditionally just get crustaceans like crabs and we dump them in pots of boiling water to cook them. Does this mean that we ought to be rethinking this?

Eleftherios - I think so too. The industry also needs to rethink that as we do as scientists when we use them in the lab. Firstly, it's important to find methods to kill them as fast as possible. Then, for distribution we should probably be transporting dead crustaceans and not alive ones. If they're alive, the restaurants or the people that want to kill them should use methods to kill them fast when we catch them. They should be dead when we buy them from the shops.

Chris - What's a good way to do that then? Not that there's a good way to kill anything, but when it's a necessity, is cold temperature, do you think, possibly putting them in the freezer to drop the temperature down so they effectively become hypothermic and unconscious, is that the best way?

Eleftherios - We tried ice slurry because we're on the verge of finding the best method and we're conducting experiments on that. Coal doesn't seem to do the trick, especially on large lobsters. We study Norway lobsters, langoustine, and they didn't die as fast even in the freezer in minus 20. So we think the best method will be electro shock because it renders them unconscious really fast. We are trying to investigate that more, but I think many companies are trying to implement electrical stunning as a better method than chilling.

The Milky Way galaxy.

26:60 - How long do galaxies take to form?

Looking back through time to learn about the structures which populate the universe...

How long do galaxies take to form?

James Tytko asked the University of Cambridge's Public Astronomer Matt Bothwell for help with the answer...

Matt - How long does it take to create a galaxy is a very good question that it's not so easy to answer. A question like, how long does it take to form a planet or a star, is a bit easier because there's a set moment at which it becomes a planet or a star. You start with a cloud of hydrogen and gravity squishes it down, and when it starts burning via a thermonuclear reaction, turning hydrogen into helium, that's what we say is the beginning of a star. But for a galaxy, things are a bit more fuzzy and loose. We know that from the earliest moments of the universe, there were fluctuations where one part of the universe was slightly hotter and another part of the universe might be slightly colder and over millions and then hundreds of millions of years, these patches of the universe started to become denser and attract matter and gas and became giant clouds of gas that began to form stars.

James - We can know all this because although the sheer scale of the universe presents a lot of challenges when trying to study it, one of the really cool things is that astronomers are looking at structures millions or even billions of light years away, which means they have the advantage of being able to understand how they evolve through cosmic time.

Matt - With telescopes like the James Webb Telescope, we can look back around 13 and a half billion years into the past to the first few hundred million years after the Big Bang, and we see baby galaxies. These baby galaxies, through billions of years of cosmic history, grow and make stars and eventually become the galaxies we see around us today. But there's not one obvious cutoff point where you can say, this is where a galaxy has formed. One rough idea might be about a billion years. We know for sure that when you see these earliest galaxies, they don't look much like a galaxy as you might imagine it, they look like slightly chaotic splotches of gas with stars mixed in. If you come back about a billion years later, the galaxy will look a lot more like a grownup mature galaxy with things like spiral arms that we recognise from galaxies around us.

So it takes maybe a billion years to go from a chaotic splotch of gas to something like a grownup galaxy with features that we recognise. But then that galaxy is going to continue to change and the gas is going to be turned into stars. In a sense, galaxy formation isn't finished yet. Our Milky Way is still turning its gas reservoir into stars. Our Milky Way galaxy makes four or five new stars every year. Maybe the galaxy formation process is only finished when the initial reservoir of gas is converted into stars. For galaxies like our Milky Way, that's going to be billions of years in the future.

James - So Dave, your question has potentially lots of answers: maybe a billion years to create a galaxy, maybe the whole age of the universe, and maybe no galaxy has ever reached its final form. You can always count on Matt to leave you humbled at the magnificence of the universe. That was Matt Bothwell, public astronomer at the University of Cambridge's Institute of Astronomy. Next time, from one Dave to another, we'll be answering this question.

Comments

Add a comment