Strange gamma-ray flickers seen in thunderstorms for the first time

Nick Petrić Howe

Welcome back to the Nature Podcast, this week: investigating intense gamma rays in storm clouds…

Benjamin Thompson

...and a complete circuit diagram of the fruit fly brain … I’m Benjamin Thompson.

Nick Petrić Howe

And I’m Nick Petrić Howe.

<music>

Nick Petrić Howe

Gamma rays are an invisible type of radiation produced by intensely powerful forces, like black holes, exploding stars and thunderclouds. Yes, around the globe right now, huge amounts of gamma rays are being produced inside storms. Scientists already knew about two types of gamma-ray phenomena in clouds — long gamma ray glows that last as long as a minute and high-intensity flashes that come and go in a few millionths of a second. But there’s a lot that researchers don’t know about these glows and flashes, for example, whether they have a role to play in creating lightning. This week Nature has two papers from a group of researchers who have revealed a lot more about the intense environment inside storm clouds. They’ve found that gamma-rays are much more common than previously thought, and discovered things never seen before, including flickering gamma-ray flashes that blink on and off. But observing all this wasn’t easy. Gamma rays are rapidly absorbed by the air so to see them the team needed to commandeer a very special aircraft that could soar above some of the most powerful storms. To hear more about the work, reporter Lizzie Gibney called up one of the study’s authors Nikolai Østgaard from the University of Bergen in Norway. She asked him about what the team set out to study.

Nikolai Østgaard

We were targeting actually high convective thunder clouds in the Caribbean and over Central America. And when you have this high convection, you know you heat the air — there's a strong updraft inside a cloud, and that leads to separation of charges like a battery, but the voltage here can be up to millions of volts. In that environment, you can accelerate some free electrons. And when the electric field is high enough, then electrons can run away. And as they run away, they also collide, and in the interaction with neutral atoms and molecules, they will produce gamma rays.

Lizzie Gibney

So we know that thunderstorms can produce gamma rays. Before this research what types of gamma rays physicists think could be produced?

Nikolai Østgaard

What we know from before is that when the electric field inside the cloud has reached a certain level, it will start glowing. We have gamma ray glowing over extended area, seen from balloons, from aircraft flying at low altitudes. And then we have TGF’s, Terrestrial Gamma-Ray Flashes, which are only 10 to 100 microsecond long, and they have been seen from space.

Lizzie Gibney

Okay, so let's talk about how you study these gamma rays. Presumably, you've got to go out and find a storm somewhere. How do you do that?

Nikolai Østgaard

We started four years ago, planning for renting the ER-2, which is an aircraft owned by NASA. And what's special about the ER-2 is that it can fly at 20 kilometers, which is above even the high convective thunder clouds. And most of our flights we went down to Mexico, to the Campeche Bay and also to Yucatan. And Campeche Bay was really the Holy Grail – that's where big storms developed.

Lizzie Gibney

And so, you have a bunch of instruments on this very high-altitude plane. What did they pick up? What did you discover during your campaign?

Nikolai Østgaard

We had this hypothesis that if we managed to find a cloud which were glowing in gamma rays, that cloud would probably also produce short TGF’s, which have been seen for decades from space. And in order to ensure this, we had real time data, just low resolution, but good enough to make sure that we were flying over a gamma ray glowing cloud, and then we could instruct the pilot to return to that cloud as long as it was glowing.

Lizzie Gibney

And so what did you observe? Did you see lots of these glows?

Nikolai Østgaard

Yeah, we saw enhancement of gammas up to 10 times the level of the background. It was like on and off every time we passed the cloud, we could see like 10s of these enhancement. And so they raised in 10 seconds or 5 seconds, and then they were terminated by lightning activity. So this is what we write in the paper. It was almost like a boiling pot of gamma ray bubbles.

Lizzie Gibney

Gosh, there's this big hive of activity within that thunderstorm.

Nikolai Østgaard

Yeah.

Lizzie Gibney

What about the flashes that you mentioned as well, these more intense radiation from gamma rays?

Nikolai Østgaard

So, you see the cloud is glowing, and at the same time as we were seeing the glows or the bubbles during those we saw a few what we call classical TGFs – that's kind of the TGFs that we seen from space, which are very, very bright. But then we saw hundreds of much weaker TGFs, and those cannot be seen from space. And that was surprising, because that means that these transient events, they are maybe 100 times more frequent than we thought before, and that puts them into maybe a one to 100, maybe one to 10 ratio with lightning. This makes the connection to the lightning discharge very much more intriguing. In addition, we also saw something that we have not seen before, and we name them flickering gamma ray flashes. First you see an enhancement of the glow, and then it starts to be pieced up in pulses. And each pulse lasted for two milliseconds, but the whole flickering event could last for a hundred of millisecond.

Lizzie Gibney

So these pulses, these flickering gamma ray flashes, that's not something anybody's ever seen before.

Nikolai Østgaard

Back in the 90s, there had been observation that look very similar, but at the time, they were called something very different. And they were seen from space, the ones we saw from the aircraft — 27 I think, in the paper we report 24 of them — what was special about them is that there were no optical signature associated with them, no radio. But after the flickering, there was first a signature of a radio pulse, which is a signature of lightning initiation, these flickering gamma ray flashes, and also the new type of TGFs, they were all followed by intense lightning activity. And that leads us to think these could play an important role in initiating the lightning activity.

Lizzie Gibney

That would be quite a discovery, because lightning has been a mystery for what, hundreds of years, hasn't it?

Nikolai Østgaard

Yeah, absolutely. So what we need to resolve these things is that first we need to know the extension of these glow bubbles. We need to know the extension of the flickering gamma ray flashes, they don't tell us how big they are. So in order to do that, you need to build a gamma ray imager, which is our plan to do, and go up there again.

Lizzie Gibney

Well, it sounds like there's an awful lot that can be seen if you do your observations right. You saw so much more on this campaign than other teams have in the past. What do you think made it so successful?

Nikolai Østgaard

I think the key was to have this downlink of data. So when we entered a cloud that produced all these gamma ray glows, our hypothesis that this is where you're also going to see all these transient short events. And that turned out to be correct. And of course, having the aircraft flying at 20 kilometers, because if you have been over this from space, you wouldn't see any of it.

Lizzie Gibney

It sounds like, overall, they're a lot more common than we thought. If you actually can get in up close like you did, they seem to happen a lot of the time the lightning happens.

Nikolai Østgaard

Yeah.

Lizzie Gibney

It sounds like the gamma rays are just a really clever way to peer inside a thunderstorm. Obviously, you can't go right in the middle, but the gamma rays fly out, so they give you a little bit of a peek into what's going on.

Nikolai Østgaard

That was the goal to get as close as possible without going into the thundercloud. There's so much turbulence, that's not where any pilots want to be.

Nick Petrić Howe

Nikolai Østgaard from the University of Bergen there, to read his papers, and a news article from Lizzie, look out for links in the show notes.

Benjamin Thompson

Coming up. The first complete circuit diagram of an entire fruit fly brain. Right now, it’s time for the Research Highlights with Dan Fox.

<music>

Dan Fox

Europe's oldest battle featured warriors from further afield than previously thought. Researchers analyzed the shapes of 64 bronze and flint arrowheads found at the site of a conflict that occurred more than 3000 years ago in what is now northern Germany. The team then compared the arrowheads with nearly 5000 other Bronze Age arrowheads found across Europe. Most of the arrowheads were similar to implements typically found in northern Germany, suggesting that most of the warriors were locals. But others, such as bronze arrowheads with a barb on one side, were commonly found to the south, suggesting the presence of foreign fighters and that a major interregional conflict took place on the site. Dig up more on that research in Antiquity.

<music>

Dan Fox

The dwarf planet Ceres may host a frozen ocean that is almost pure water at the surface and becomes ‘dirtier’ the deeper you dive. Ceres orbits the Sun in the asteroid belt between Mars and Jupiter and researchers are interested in figuring out its composition to understand how small astronomical bodies evolve. But currently there is some conflict between observations, with some indicating that the interior of Ceres is rich in ice, while others point to a low ice content. To clear up this contradiction, researchers developed a new model of Ceres. In the model, the ice content of the dwarf planet's crust gradually falls from about 90% near the body's surface to 0% at a depth of 117 kilometers. These findings suggest that Ceres had a global ocean that slowly froze from the top down. During this process, the ocean became progressively richer in impurities like salts or mud, causing the frozen ocean's purity to decrease with increasing depth. Dive deeper into this research in Nature Astronomy.

<music>

Benjamin Thompson

Despite having a brain the size of a poppy seed, the humble fruit fly, Drosophila melanogaster, has played a central role in the understanding of how this organ works. One of the key ways that researchers think will help unlock the mysteries of the brain is to map all the neurons and connections between them, creating a wiring diagram known as a ‘connectome’. This week, there are nine papers published in Nature from the FlyWire consortium relating to the biggest map of the fruit fly brain yet. Joining me to talk about these papers is Lauren Wolf, Nature's Americas Bureau Chief who's been following the story. Lauren, welcome to the show.

Lauren Wolf

Hey! Thank you for having me. It's been a while.

Benjamin Thompson

I mean, Lauren, let's talk numbers. So this isn't the first map of a fly brain to be published, but it is huge in scope.

Lauren Wolf

Indeed. Yeah, so, you know, the fruit fly is a model organism. So just like some other model organisms that we look at in the lab, like worms, like C. elegans, scientists are wanting to study these model organisms, because they're just that, they're a model they're– they're really useful. And so, yeah, we have other partial maps of fruit fly brains. I think there was one of a young fruit fly — six hours old — and that was a lot fewer neurons than this new map that's been published. So for that young fruit fly it was about 3000 neurons, and this new one was 140,000 neurons. So this is a big step up. And in the past, they have also done this sort of adult fruit fly brain, but a hemibrain, so it's only part of the brain, not the whole thing. And again, that was a lot fewer neurons. So this time, it's the whole enchilada. It is an entire fruit fly’s brain and adult female fruit fly that they have chopped up into a bunch of slices and imaged and put them all back together and come out with this wiring diagram.

Benjamin Thompson

I mean, you've kind of glossed over how much work this must have been its a Herculean effort, right? So this single fruit fly, let me get this right, they're taking this tiny brain, sliced it up, and then try to fit the puzzle back together again, which it sounds easier to say than to do.

Lauren Wolf

Yes, I think you can think about it as like slicing pepperoni, if you will. So you're slicing it up into a bunch of pieces, you're imaging it with electron microscopy, and then I believe that they use some AI model to kind of put all those slices back together, electronically speaking, to build their preliminary model. And that's only the first step, because you can't just trust all of that – it's a lot of detail. And so you want scientists to then go back in and kind of hand check all of that for errors, which is the next step of this enormous project, which has gone on for years.

Benjamin Thompson

And involved an international team of researchers and other folk as well.

Lauren Wolf

Correct. What they realized was that if we had to have this manually checked for errors, it would probably take 33 years if you just had one person doing it. That’s not efficient. And so instead, what they did was they looked for volunteers in the community. And I think interestingly, some of this work happened during the pandemic, when scientists were bored and looking for things to do when they couldn't go anywhere else. Yeah, so what they did was they built a little game, as I understand it, and kind of trained some volunteers, so both scientists and non-scientists, then they picked the top volunteers from that cohort, trained them how to identify different types of cells in the fruit fly brain, and then they gave them the actual data and had them check everything.

Benjamin Thompson

Goodness. And that was kind of stage one as I understand it. Like putting the pieces in the right place is obviously super important, but then actually saying what those pieces are is something else that needs to be done as well, right?

Lauren Wolf

Yeah. So it's making sure everything is wired properly. You know, when the AI put it all together, but then they do this thing called annotation, which is to label each of the neurons as a specific kind of cell. And so we know that there's a lot of different cell types in our body, and in the brain, it's no different. I think in this project, they identified something like more than 8000 cell types in the brain, and so all of those neurons had to be hand labeled. And I think there's a nice analogy that the researchers use. It's kind of like when you're doing this annotation, it's like, well, satellites can take pictures of Earth, and it can show you where the lakes and the rivers are, in the same way that the AI can patch things together. But what it takes is a person going in and saying, well, this is what this lake is called, or this is what this river is called, because the AI is not going to be able to do that for you.

Benjamin Thompson

I mean having this map, having this circuit diagram, whichever metaphor we're going to use, I suppose, is one thing, but actually applying it to the real world is something else. Now we have connectomes, as we talked about, from other studies there, and they have been used in a lot of research, a lot of research papers. What has this work accomplished thus far do we know?

Lauren Wolf

Well, it's really interesting just to take a step back. I think there has been an argument in the community for a while about whether making these connectomes is even worthwhile, because people think, well, you're just getting this static map of one entity's brain, and what are you going to do with that, right? Is it worth this much time and energy? Because these things are so complex, and as the organism gets more complex, the brain gets more complex. Is it worth our time to do this? But I do think that some of those other examples that I mentioned earlier right, the worms and things like that, you can actually learn something from having this map and that seeing how the organism behaves, and being able to kind of pinpoint one of these neurons, and then seeing how it connects to the others, how that's affected by their behavior, and so they are finding that with this fruit fly. So you mentioned there are nine papers. So two of them are really about the work of mapping this brain, and the other seven are researchers using this data and seeing what they can learn from it, which is a really nice example to kind of put down some of those who had questioned whether it was worth doing these connectomes and that it is and that you can learn stuff from it.

Benjamin Thompson

One of the things that caught my eye having a little look around what's been going on here is that researchers fed this map into a computer to try and make a virtual connectome that they could then sort of prod, but then take into the real world as it were.

Lauren Wolf

Yeah, I think you pointed out exactly the one that the reporter that I worked with on this story told me about where they did use the data to make sort of a virtual computer model of the fruit fly brain. And they know this particular neuron is a sensory neuron that responds to sugar, like the flavor, the taste of sugar. And so if we activate this one, then what happens? And now you can know this neuron’s connected to that one, which is connected to that one, and that one and so on. If you light that one up and have it transmit its signal, you can watch this virtual fruit fly sense the sugar and then stick out its tongue or proboscis, as if it's going to eat the sugar. And you could do that similarly with a bitter taste sensory neuron, you know, you give it a bitter taste signal, and it kind of is in its avoidance mode. It doesn't want that taste or that flavor. And then, in the end, as if all of that wasn't enough, they then activated those neurons in a real fruit fly, and they found that the simulation that they had built was more than 90% accurate at predicting which of those neurons would respond and how the fly would behave. So they sort of confirmed their theoretical model, which is what researchers want to do. You want to align the theory, the model, with reality.

Benjamin Thompson

Which for sort of human brains, is so hard to do because they are so complicated. And I guess the fruit fly is that sweet spot of like it does have complex behaviors, but it has, what, a million-fold less neurons. But I think what you said at the start that this is a single female fruit fly, so in a sense, n = 1, and is there the possibility that this is an unusual brain, and so what's found here maybe isn't applicable across fruit flies in general?

Lauren Wolf

Yeah. Think there's always that possibility. I don't know exactly the details of this, but I'm sure they chose this fruit fly and didn't just sort of pick one randomly. But I think that when they compared this fruit fly brain with the hemibrain that I mentioned earlier, they did see some differences. And obviously this is also a female fruit fly, it's not a male fruit fly, which would have maybe some different behaviors. And so, you know, the idea would be to compare some other brains eventually, and kind of confirm all this wiring. There's another caveat too, in that this is just the wiring diagram of the neurons. There's lots of other cells in the brain, and there's other ways that neurons communicate, even that don't involve just their kind of direct physical connection through neurotransmitters. One way that they do it is through this like Wi Fi, not physically connected way of doing things where, say, one neuron might release a neuropeptide and another one catches it. And so, there's all these other parts of the brain that we haven't even mapped yet people are trying to. So, this picture, this map that we have, isn't the full picture, but it's certainly the most impressive thing that we have so far, and it's a huge feat.

Benjamin Thompson

And have you got a sense of what the research community is saying about this work now these papers are coming out?

Lauren Wolf

I mean, I think our reporter who did this story got some quotes of this is, is really impressive. It's the most complex brain for the most complex organism so far that we've been able to map fully. It took however many years and however many people to do this one. So you can imagine that, as you step up in complexity, how long that's going to take, right? People are working on the mouse brain, for example, and that's going to take an even longer time, and it'll be an even bigger feat. But right now, this is this sort of gold standard for where these connectomes are, and people are pretty excited about it. You know, some of this data has been available to Flywire researchers, the ones that are part of this consortium that worked on the data, and so that's how we have these seven other papers. So the data has been available to those people, and so now it's going to be available to all the public, and it's free to look at. And so I'm sure that there will be even more explanations of awe as time goes by.

Benjamin Thompson

Well, Lauren, thank you so much for joining me today, and we'll put links to those in the show notes.

Lauren Wolf

Thank you so much.

Benjamin Thompson

Nature's Lauren Wolf there.

Nick Petrić Howe

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

Benjamin Thompson

Well, I’ve got a story that I read about in Nature, and it's about a rather unusual chemical bond between carbon atoms.

Nick Petrić Howe

Well, bonds between carbon atoms, I guess, are useful for all sorts of chemistry. But what makes this one unusual?

Benjamin Thompson

Yeah, so this is a covalent bond we're talking about. And you know, I'm sure you remember from high school, that's one of the bonds that forms between atoms – it’s quite a strong bond. And covalent bonds form when atoms share one or more pairs of electrons. Now, in this case, chemists have managed to make a covalent bond between two carbon atoms that is a single lone electron.

Nick Petrić Howe

Oh, wow. I'm guessing this is quite a difficult thing to actually achieve, if it's normally done with pairs of electrons.

Benjamin Thompson

Absolutely right. So this is not the first time this sort of bond has been made, it has to be said. One was made in the 90s between two phosphorus atoms and one between copper and boron in 2013 so pretty rare, though, right. But this is carbon so, you know, a building block of us, you know, drugs, plastics, proteins and so on. So carbon-carbon bonds are of special interest to chemists, but to make a single electron covalent bond has eluded scientists until now.

Nick Petrić Howe

Wow. So, quite a feat then. So how did they sort of encourage the atoms to do this sort of bond?

Benjamin Thompson

Yeah, it's quite a story then. I guess the concept of single electron covalent bonds dates back to Linus Pauling in 1931. Okay, now he proposed them, but researchers didn't have the tools to observe them. And as we've said there, even now, it's kind of tricky, right. Because these bonds are somewhat unstable. A one electron covalent bond will break easily, right. There's a tendency to release that electron or to scoop up another one to get a pair back, to get things back to parity. And chemists theorized that these bonds might form between, you know, short lived intermediate structures that appear during chemical reactions. And the trick was stabilizing them, and that's what they've done in this instance. So they've designed a chemical structure, you know, a shell, if you like, made up of carbon rings. And at the center of this shell is a carbon-carbon bond. Now this bond is stretched somewhat. Now, apparently it's a relatively long length for a carbon-carbon bond, and this makes it susceptible for something to come in and steal one of those electrons away and a process called oxidation, and that's what's happened. So to capture this, the team made a crystallized form of this molecule, and when it's oxidized in the presence of iodine, what happens is this reaction yields a purple salt, so they know that it's happened. Okay, it's kind of a tell. And this shell holds everything together. So this single atom covalent bond is caught in the act, and then the researchers can sort of probe it and see and make sure they've got it and what have you.

Nick Petrić Howe

Okay, so this thing is sort of stabilizing this single electron bond. Now, we've talked about various kinds of ways of making chemical bonds on the podcast before, and typically this is because it will enable chemists to do lots of useful things. So I'm guessing that's the case here as well, right?

Benjamin Thompson

That's an open-ended question. It has to be said. One of the researchers quoted in the article, says, ‘it's hard to know, you know, whether this work will have any applications’. You know, it's a bit of a curiosity, but it will be in the textbooks. But being able to manipulate carbon is something that is so useful to chemists because of its ubiquity. So while we can't predict the future, it is something that is of interest, certainly, and chemists in general know that atoms interact in all sorts of ways, and studying what's actually possible and looking at unusual bond types could help you know, understand what a chemical bond, what a covalent bond, actually is in a kind of an atomic level. And that's what I think the researchers behind this work are kind of interested in.

Nick Petrić Howe

Well, let's see what happens next in the world of chemistry then, I guess, it could be an exciting time for making and breaking bonds in the near future. For my story this week, if you’ll bond with me, is about a perennial topic on the podcast, it's AI chatbots. And this was a story I was reading about in Nature, about a Nature paper about how bigger chatbots tend to be, I guess, kind of a bit more confident, but confidently incorrect.

Benjamin Thompson

Right. So ‘bigger is better’ is not necessarily the correct phrase then when it comes to chatbots.

Nick Petrić Howe

It doesn't seem so. So there's been a lot of work that has looked at hallucinations, or chatbots just generally getting things incorrect. And so, researchers have been trying to keep an eye on this. And this study has looked at three families of large language models, GPT, LLaMa and Bloom and looked at their earlier versions and their later versions to see how well they respond to a series of questions. Now, as they got bigger, they got more accurate, which would be what you expect. They're using more training data. They've been more fine-tuned, that sort of thing, so you'd expect them to actually be more accurate. But the problem seems to be, this research has revealed, that as they get bigger, they also answer more questions. And now that might sound like a good thing, but they'll answer the questions whether or not they know the answer.

Benjamin Thompson

Right. So they've got a lot of training data, but that might not be related to the question they're being asked. They just say, yeah, do you know what? I can answer this and have a go.

Nick Petrić Howe

Yeah, exactly. Now there's a quote in this article that illustrates what this problem is, and it's from a philosopher of science and technology called Mike Hicks, and he says ‘that looks to me like what we would call bullsh*tting. It's getting better at pretending to be knowledgeable’.

Benjamin Thompson

And this presumably, is potentially a significant issue, because people are putting more and more trust into these chatbots, for better or worse, but if they're just chancing it and giving an answer that could have all sorts of knock-ons, exactly.

Nick Petrić Howe

So this is a thing that the research itself has looked at specifically. So in addition to looking at whether or not the chatbots were producing correct or incorrect answers, they looked at people's abilities to gauge whether or not the answers they were being given were correct, incorrect or avoidant. And so people incorrectly classified inaccurate answers as being accurate, really quite often, around 10 to 40% of the time. So, in essence, that means that humans are not able to really know whether or not these chatbots are answering correctly or not.

Benjamin Thompson

So the old phrase ‘Bullsh*t Baffles Brains’, then seems like it might be apposite here.

Nick Petrić Howe

Indeed, that does seem to be the case. And so, the question is, what to do about this. Now, the researchers interviewed in this article do think there are ways to make it so these chatbots don't do this. They think that if the questions are more difficult, then the chatbot should, you know, not answer it. Say this is beyond the scope of what I can say. And that sort of thing is done in certain applications. So for example, in medical applications where AIs are used, they'll routinely say, you know, this is beyond my knowledge. But the double challenge is, if you're making a general-use chatbot and you want to maybe sell that technology to people, it's unclear whether or not you want to basically put into your model its own limitations. You want to really show that it can't answer certain questions. So it'll be interesting to see like, what the creators of these AIs do, whether they'll just try and make them more accurate more of the time, or if they will input things to make the chatbots be honest, basically, and say whether or not what they're producing is correct, or whether it's something beyond what they know.

Benjamin Thompson

Because I imagine, if you just give it more data, this problem just gets shunted down, and then it learns more stuff, but then it gets more confident later on about what it doesn't know, and we haven't got to the root of the problem.

Nick Petrić Howe

No exactly. That's what this research really seems to show. Bigger models – they are giving you more correct answers, but they're also just answering more questions. So they’re just more confident more of the time, which I imagine, if you've swallowed the entire internet, would probably be how humans are as well, but perhaps not that useful if you want to use this as a tool for whatever purpose in the future.

Benjamin Thompson

Right well, another thorny issue for the AI chatbot researchers to look at. But let's leave it there for this week's Briefing Chat. And listeners for more on those stories, and where you can sign up to have stories like them delivered directly to your inbox, look out for links in the show notes.

Nick Petrić Howe

That’s all for this week, as always you can keep in touch with us on X, we’re @NaturePodcast, or you can send an email to podcast@nature.com. I’m Nick Petrić Howe.

Benjamin Thompson

And I’m Benjamin Thompson. See you next time you.