How light-based computers could cut AI's energy needs

Primary Topic

This episode explores how photonic computers, using light instead of electronics, can drastically reduce the energy requirements of AI systems.

Episode Summary

The Nature Podcast delves into the innovative realm of photonic computing, showcasing how it can transform AI's energy consumption. Hosts Benjamin Thompson and Lizzie Gibney discuss the increasing energy demands of AI systems and introduce photonic computing as a potential solution. Guest expert Dong Baowei explains that photonic chips compute using light, making them faster and more energy-efficient compared to traditional electronic systems. The episode highlights a breakthrough where a less coherent light source made photonic computers more effective, simplifying control and boosting efficiency. This counterintuitive approach, replacing precise lasers with broader bandwidth LEDs, promises significant advancements in photonic computing for AI applications.

Main Takeaways

  1. Photonic computing uses light to process data, offering speed and energy efficiency advantages over electronic computing.
  2. A breakthrough in using less coherent light sources can enhance the efficiency of photonic chips.
  3. This technology is particularly suited for AI due to its ability to efficiently perform matrix vector multiplications—a common operation in AI algorithms.
  4. Photonic computing is still in the experimental stage but shows promise in practical applications, such as analyzing medical data.
  5. The developments in photonic computing could lead to broader, more sustainable AI applications in the future.

Episode Chapters

1: Introduction to Photonic Computing

The episode begins with an overview of the growing energy demands of AI and introduces photonic computing as a solution. Benjamin Thompson: "Power requirements and why scientists are concerned about bird flu spread in cows I'm Benjamin Thompson."

2: Interview with Dong Baowei

Dong Baowei discusses the advantages of photonic computing, including its higher bandwidth and energy efficiency. Dong Baowei: "Photonic computing can compute at much higher energy efficiency. Lights just pass through an optical fiber."

3: Advancements in Photonic Chips

Details on the new approach using less coherent light sources for more efficient photonic computing. Dong Baowei: "We use a super luminescence led sled that has a broader bandwidth than the laser."

4: Practical Applications and Future Prospects

Discussion on the practical applications of photonic computing and its potential impact on future AI technologies. Dong Baowei: "We indeed use this system to run some algorithm."

Actionable Advice

  1. Explore photonic technology: Consider how photonic computing could be integrated into current technology projects to enhance efficiency.
  2. Stay updated with AI advancements: Keep track of developments in AI and computing technology to leverage new efficiencies.
  3. Experiment with new technologies: Encourage experimentation with less conventional technologies in research to uncover potential breakthroughs.
  4. Advocate for sustainable technology: Promote the adoption of energy-efficient technologies in computing to reduce environmental impact.
  5. Educate about photonic computing: Spread awareness about the benefits of photonic computing through workshops or seminars.

About This Episode

Computer components based on specialised LEDs could reduce the energy consumption of power hungry AI systems, according to new research. AI chips with components that compute using light can run more efficiently than those using digital electronics, but these light-based systems typically use lasers that can be bulky and difficult to control. To overcome these obstacles, a team has developed a way to replace these lasers with LEDs, which are cheaper and more efficient to run. Although only a proof of concept, they demonstrate that their system can perform some tasks as well as laser-based computers.

People

Dong Baowei, Benjamin Thompson, Lizzie Gibney

Companies

Nature Podcast

Books

None

Guest Name(s):

Dong Baowei

Content Warnings:

None

Transcript

Nature Podcast
The Nature podcast is supported by Nature Plus, a flexible monthly subscription that grants immediate online access to the science journal Nature and over 50 other journals from the Nature portfolio.

More information at go dot nature.com plus.

Us Cellular
Us cellular notice that the way we use our phones has gotten ironic.

Us Cellular
Yeah, we try to put our phones down for dinner, but the menus on a QR code? That's ironic.

Us Cellular
We hit like on social posts that we don't actually like. Ironic.

Us Cellular
Which is why us cellular created us mode to help us reconnect with each other and use our phones less.

Us Cellular
Ironically, a phone company wanting people to use their phones less ironic.

Us Cellular
Let's find us again with us mode from us cellular.

Us Cellular
Visit uscellular.com builtforus to get started.

Us Cellular
That's uscellular.com builtforus.

Benjamin Thompson
In an experiment we.

Lizzie Gibney
Don'T know yet, why is blight so.

Dong Baowei
Far like it sounds so simple? They had no idea, but now the data speaks. I find this not only refreshing, but.

Benjamin Thompson
At some level, astounding.

Lizzie Gibney
Nature welcome back to the Nature podcast. This week, how light could lower AI's.

Benjamin Thompson
Power requirements and why scientists are concerned about bird flu spread in cows I'm Benjamin Thompson.

Lizzie Gibney
And I'm Lizzie Gibney.

AI is everywhere in science, but as AI models get bigger and more sophisticated, the computing power needed to run them is rocketing.

Estimates suggest that training a large language model creates about twice the amount of CO2 that an American would produce over their entire lifetime.

This increased thirst for power has led teams around the world to investigate ways to make AI chips that run more efficiently and sustainably.

And one way is to get some components to compute using light rather than electronics.

Known as photonic computers, these systems can be quicker and much more efficient, and they're analog rather than digital.

So instead of having to convert information into binary ones and zeros, they transmit raw numbers in the properties of light, such as the amplitude or brightness of a super precise laser.

Photonic chips are particularly suited to running some AI algorithms, but their lasers can be bulky and fiddly to control.

This week, a team demonstrates something rather counterintuitive, that by using a different, and, in some senses, worse, light source, they can make photonic computers even more efficient and easier to run. I called up one of the paper's authors, Dong Baowei, to find out more, and he explained some of the advantages that photonic computers have over their digital counterparts.

Dong Baowei
I think there are three major advantages. One is it can provide much higher computing bandwidth. This means it can send data at much faster data rate. For example, if you use electronics to perform computing. It can send data at a rate of a few gigahertz. By the way, few gigahertz means you send 1 billion of data in 1 second. And if you use photonics, you can send data at tens of gigahertz or hundreds of gigahertz. That is much faster than electronics. The second advantage is that photonic computing can compute at much higher energy efficiency. Lights just pass through an optical fiber. You don't lose that amount of energy that dissipates as heat.

Lizzie Gibney
So why in particular is photonic computing being seen as good for AI calculations?

Dong Baowei
This is because at the heart of AI computing, the major computing step is called matrix vector multiplication. It's nothing more than just a massive amount of additions and multiplication, and photonics is just very suitable to do that. And the size of these AI models are actually growing exponentially. So, according to research from OpenAI, the size of the AI models is doubled every 3.5 months, and that consumes enormous energy to perform AI.

Photonic computing is very energy efficient and very fast. That means it can process AI models much faster at the lower energy consumption.

Lizzie Gibney
So how do photonic processing units, kind of the heart of the computer, usually work?

Dong Baowei
So, let's assume you have a vector. So, vector is an array of data. You send each entry of this array to multiple channels to represent these data, and you use lasers to send this. Lasers are basically a very coherent light. Coherence means these lights are very well defined in terms of wavelength and phase. And you send these lights into the photonic processor and they are processed. And because you need to do addition in the processor, that means light from different channels, they need to meet each other. The fact that you are using lasers means that you have to use different colors at each channel, because otherwise, if they have the same color or say the same wavelength, they will interfere. You can think of the light as a sine wave, and when the peak meets the peak, you get a bright signal. And if it's a peak, meet a trough, you'll get a very dark signal, so that in the system there will be perturbations, and that gives you noise and error, so that you have to use different colors to avoid this interference to happen.

Lizzie Gibney
Tell me a bit more about what you did in your research. You didn't use lasers.

Dong Baowei
We actually find a way to completely eliminate this interference problem.

Consequently, we can use light at the same color to enter all different channels. That means we only need one light to process all the information.

We use a super luminescence led sled that has a broader bandwidth than the laser. By broader bandwidth, I mean we have increased the number of wavelengths a little bit, so it's less coherent. Or sometimes people say it's a worse light source. That solves this interference issue.

Lizzie Gibney
And how come it doesn't interfere like the lasers?

Dong Baowei
So, since we cover a broader range of wavelengths, at some wavelength light interfere constructively, give you a bright output, and some wavelength light interfere destructively give you a very dark output. And because we have many of these wavelengths, some gives you dark, some give you bright, the result is actually an average of them, you get a very stable output.

Lizzie Gibney
So rather than trying to avoid a any interference entirely by having these lasers that are of completely different wavelengths, you instead have just a kind of much messier light. And then you get lots of cancellations and lots of adding up, but on average, your signal stays the same.

Dong Baowei
Yeah, that's the trick.

Lizzie Gibney
Okay, so it sounds really counterintuitive, because people have spent years trying to make lasers better and better for this kind of application and for others. And now you're saying actually it's a good idea to make your light in some ways, a bit worse.

Dong Baowei
Yes.

Lizzie Gibney
Where did this idea come from?

Dong Baowei
It's actually a coincidence. So we were trying to solve this interference issue, and we begin with thinking about, is there anything we can do about the photonic processor to avoid this? And one day, it turns out the system just becomes so nice, it does not interfere. The output is so stable, and we cannot figure out why.

Smriti Malapati
I.

Dong Baowei
But it's a good sign. So we try to figure out what is happening, and we realized that it's because of the experiment of a previous colleague. He changed the light source to a worse one, I would say in that word, and he solved the problem. So we were very surprised about this. So I would say it's completely a coincidence.

Lizzie Gibney
Sometimes serendipity does wonders in science.

So this kind of processing unit that uses these leds rather than lasers, what advantages does that have?

Dong Baowei
One is that controlling an led is much simpler than controlling a laser, because the laser, you have to control its wavelength, control its face very precisely. You need to input many power to do this, so that by controlling an LEd, you actually put in less energy to control it. This is the first point. And the second point is that now you can use one single led to send light into many channels, and that just make your system compute at much broader bandwidth. The third aspect that is the system becomes so easy to control.

Lizzie Gibney
And is this just a prototype or an idea that you have or have you actually used it to run some algorithms?

Dong Baowei
We indeed use this system to run some algorithm. So in our recent study, we viewed this prototype and use it in the machine learning algorithm to analyze the working gate of Parkinson's disease patients collected by a false sensor. And our chip can analyze the guide signal from these Parkinson's disease patients and identify their status.

Lizzie Gibney
And how successful was it?

Dong Baowei
Oh, it gives a very high accuracy. So in the experiment, we can identify ten Parkinson's diseases patient at an accuracy of 92%.

Lizzie Gibney
Where do you see this research going now? It seems like the promise of being more efficient and running on less energy, the real positives. Do you think all photonic computers in the future are going to use this system?

Dong Baowei
The fact that you can use one sled to replace multiple lasers is a very promising solution. It saves your effort in control, it saves energy, increase your, then we give all the advantages. So we do hope that this method can be adopted in larger photonic computing systems.

Lizzie Gibney
What do you think your colleagues, your fellow physicists, will think of the idea of chucking out the lasers that they've been honing and refining for so many years?

Dong Baowei
In this work, we are demonstrating that in photonic computing, using less coherent lysos might be a better option. But actually in many other applications, lasers are still the most. For example, if you want to use light to do ranging, which is called glider, you must use laser for that. So we're not saying lasers are bad. Lasers are still good, and they are the must for certain applications. But we just want to emphasize that for different applications, simply tracing higher coherence might not be the optimal way. Sometimes less coherent might give you better performance.

Lizzie Gibney
So the AI computing maybe a messier beam is better?

Dong Baowei
Yeah, that's what we think that was.

Lizzie Gibney
Dong Ba Wei, now at a star Singapore, to read his paper. Look out for a link in the show notes.

Benjamin Thompson
Coming up, what research? Looking at the spread of h five n one influenza between cows is telling scientists.

Right now, though, it's time for the research highlights with Dan Fox.

Dan Fox
The grassy lemony scent that wafts from some rose stems comes from a compound called citronell. Now researchers know how the flowers produce it.

To understand how citronell is made in roses, researchers pieced together genomes for two flowers, one which produces citronell and one that doesnt.

This data, along with sequences from nearly 40 other rose species, revealed that roses contain extra copies of three key genes involved in the production of terpenes, a group of volatile aromatic compounds of which citronel is a member after working out the genes needed for terpene synthesis in roses, the team inserted them into flowering tobacco, producing a plant that could make citronylol. The work could help produce more fragrant roses and generate new plants that can produce rose essential oil.

Sniff out that research in current biology.

A stroke in the brain can inflame the heart, but blocking some immune responses could limit the damage, according to a new mouse study.

Strokes temporarily block the flow of oxygen to the brain, but they can cause long term damage to other areas of the body as well.

To understand why, researchers investigated which genes were active in mouse immune cells one month after a stroke.

They found that strokes caused inflammation in multiple organs, particularly the heart. Production of an immune protein called il one beta spiked after a stroke. This protein altered the DNA of immune cells called monocytes, in a sense training them to promote inflammation, which in turn damaged the heart. Mitigating the effects of this protein caused the animals cardiac function to improve. The authors say that similar treatments might prevent further health problems in people who have had a stroke. You could read that research in full in cell.

Benjamin Thompson
Next up on the show, reporter Nick Petrich Howe has been finding out about the latest bird flu research.

Nick Petrichow
A particularly pathogenic strain of h five n one, known commonly as bird flu, has been devastating bird populations across the globe.

Increasingly, it seems that it is able to spread in mammals, too, with even some human cases reported.

Recently, Nature published a paper investigating the spread of h five m one influenza in cattle, something that could increase the risk of human transmission.

Smriti Malapati, a senior reporter for Nature, has been covering this topic and joins me now to discuss this paper. What else researchers know about the virus and what can be done as it spreads. Smriti hi.

Smriti Malapati
Hi Nick. I'm happy to be here.

Nick Petrichow
Well, it's good to have you here to sort of discuss through this. So I thought a good place to start is just to have a little bit of a roundup of what we know about it. I think h five m one has been around for a while. So what's different about this particular strain?

Smriti Malapati
Yeah, so what we're talking about is a highly pathogenic strain of h five n one. And why we're talking about it now is because it has been identified in cattle in the United States. So in March, the United States first identified this particular strain of h five n one in cattle in Texas, and it has since spread to more than 100 farms across the country in multiple states. And maybe a handful of people have been infected from that strain of h five n one. Also, cattle are mammals like us? And so that kind of increases the risk profile of these viruses potentially jumping to people.

Nick Petrichow
And so I wanted to talk to you about the paper that has come out, nature. What does this tell us about this sort of spread in cattle?

Smriti Malapati
Yes. The study was a preprint before it was published in Nature. So a colleague of ours, Max Kozlov, has reported on it previously when it was a preprint. So one of the big questions that researchers are trying to understand is how the virus is spreading between cattle. So, in this paper, the researchers got samples from some of the farms in the United States that were first infected with this virus, and they collected different samples from these cows that were infected, including blood samples and tissue samples and milk samples.

And they also studied the genomic sequences of the virus that they isolated from these animals. And what they find is that the virus is probably spreading from cow to cow, which is something that they had to establish.

And they also find that there is a really high amount of infectious virus in the milk, and a key target of this virus in the infected cows is the mammary glands.

And so it suggests that this virus is probably spreading via the udder root, so via infected milk, which makes it potentially simpler to control because it's probably spreading via the milking equipment. So you can just manage contamination that way. The risk that researchers are really concerned about is if it was airborne. So if this virus could spread via the respiratory route, then that would increase the risk of it potentially spreading to people that are exposed to cows.

Lizzie Gibney
Yeah.

Nick Petrichow
You've reported before that there was some evidence to suggest that perhaps cattle could be infected through inhalation, through sort of airborne particles. Right?

Smriti Malapati
Yeah, there's another preprint that came out a few weeks ago. So the previous study that we were talking about, you know, researchers took samples from cows who were infected at actual farms. In this newer preprint, which hasn't been peer reviewed, researchers experimentally infected cows. So they did a challenge study with cattle, and they experimentally infected them in two ways. One of them was via the respiratory route and another one was via the udder. And they found that the cows could get infected via both routes. So it's possible that cows can get infected via the respiratory route, but it does still seem that the mammary glands and the udder are a key target of this virus.

Nick Petrichow
So do these new studies, these preprints, do they give researchers any more concerns about this virus?

Smriti Malapati
Yeah, I think the big concern for researchers is that if this virus continues to spread in cattle, it increases the risk or exposure to people. And so there's one concern that, you know, then this virus, as it continues to spread in cattle, it could evolve to better infect mammals and therefore evolve to better infect people. But the other risk is that as it continues to spread in cattle, it increases the exposure to people, and then that access creates more opportunities for the virus to jump to people.

Nick Petrichow
So we've talked a lot about how this is spread in cattle, but there have been a smattering of human cases as well, right?

Smriti Malapati
Yeah. So there have been, I think, a handful of cases that are linked directly to the cattle outbreak in the United States. But then recently, there were also some cases in poultry farms that have been potentially linked to the cattle outbreak. And from those farms, some people have been infected from poultry. And so it hasn't been a lot of people. It's not clear whether all of the cases in people have been detected. You know, that's linked to being able to test all the people.

But of the cases that we know that have been detected, there's maybe less than a dozen people. So it's not a huge amount of people. And the risk to people remains quite low at this stage, at least from what researchers are saying.

Nick Petrichow
Yeah, there doesn't seem to be any sort of human to human transmission, which I guess would be real alarm bells for researchers.

Smriti Malapati
Yeah, definitely no evidence of human to human transmission.

Nick Petrichow
So you've been speaking to researchers about this. What are the sort of strategies that are being proposed to help control this outbreak?

Smriti Malapati
Researchers are really concerned, and so many countries are already starting to prepare for the possibility of a pandemic. I mean, we have to be clear. At this stage, the risk to people remains quite low. But some researchers have said to me that with influenza viruses, you just have to be on alert because the virus can change in a way that could make it better at infecting people, and that is a concern. And so many countries are ramping up surveillance to monitor cattle and other animals so that they can detect an outbreak when it does happen quickly. Some countries have started purchasing vaccines against this particular strain of h five n one, which so far, the vaccines that are available do seem to still work against this strain of the virus.

And another thing that researchers are doing is they're trying to study the sequences of this virus to look for whether this virus is evolving in ways that would potentially increase the risk to people.

Nick Petrichow
And, you know, we all live in, I guess, the memory of 2020. Have lessons been learned from the previous pandemic the researchers believe were better prepared.

Smriti Malapati
This time around, I think there are two things. One of them, from the pandemic is that we saw this new vaccine, the mRNA technology being used. So some researchers are already starting to develop mRNA vaccines for influenza viruses, and some companies are already working on that. And so these could be potentially a way to manufacture influenza vaccines against specific strains quicker.

And another thing that the world has learned is the huge inequality in the way that vaccines were distributed. So I spoke with an individual at CEPI, the coalition for epidemic Preparedness innovations, and they are really trying to ensure that any kind of response that involves vaccines is equitable.

They want to ensure that low and middle income countries aren't left behind when it comes to being able to access vaccines. And that's one of the things that we have learned from the Covid-19 pandemic that could be a potential problem if this does become a pandemic, although at this stage it's not looking like it at the moment.

Nick Petrichow
Well, let's hope that those lessons have been learned from 2020, and we'll continue to monitor this on the podcast as this situation develops. But for now, Smriti, thank you so much for joining me.

Smriti Malapati
Thank you.

Benjamin Thompson
That was Smriti Malapati talking with Nick Petri Chow. For more on this story, check out the show notes for some links.

Lizzie Gibney
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. Why don't I go first this week? I've got a story that, as a massive space nerd, is totally exciting, or hopefully very exciting. This was in space.com, but to be honest, it was absolutely everywhere. This is a very exciting rock that's been found by the NASA rover perseverance on Mars. So I think I would describe it as probably like the best evidence yet of ancient life on Mars.

So this is one of the big goals is like, you know, we think that probably at some point in Mars history, there was water, there was potentially some kind of microbial life. And perseverance is one of the missions that has been looking for it. And there are future ones planned.

Benjamin Thompson
We often joke in our production meetings, is it aliens? And I think the answer here is, well, probably not, but maybe.

Lizzie Gibney
So we're talking about ancient life here, right? So we're talking about probably microbes and living probably billions of years ago. So this rock was in a place called Ceyava Falls, and it's got three very cool things about it. So number one, it's got organic compounds, you know, which we know are in a lot of living things and used to make living things, and it's got veins of calcium sulfate running through it. Now, that's a mineral that gets deposited when water runs through rocks. So again, it's suggestive that there was once water there. And finally, there are these leopard spots, which are like millimeter sized splotches, and they're ringed with black, so they really do look like little spots on a leopard. And the rings contain iron and phosphate. So perseverance saw this rock, has zapped it with a load of x rays and lasers to get all of this information.

And some of these things have been seen before, but never these three things like that all together.

So those kind of leopard spots is something that's been seen on earth as a result of chemical reactions by microbes that are like living in the substrate of the rock. So it's very, very promising. But, of course, there is always a possibility that that's not the explanation.

Benjamin Thompson
Right. Well, of course. I mean, I guess perseverance can only do certain experiments and provide this evidence then, for something that is potentially analogous to something seen here on earth. But the through line between the two, there's a lot of gaps to be filled in, I'm sure.

Lizzie Gibney
Absolutely. And, you know, we are studying a completely different world to our own that's only been seen in this remote way through rovers in the past. So the kind of mechanisms that were at play there billions of years ago, we are very much, you know, speculating about. So maybe some kind of volcanic activity could have baked these markings into the rock somehow. Maybe it's a coincidence that that's where water used to be. It's possible. And perseverance really has done everything it can. There was a lovely quote in the press release that the NASA team put out saying, you know, perseverance has reached the end of what it can do here. It's thrown everything at this rock, and that means the only thing left to do is bring a little bit of the rock back to earth so we can do much more extensive studies here. And that is ultimately the plan. But it's a rather long term plan.

Benjamin Thompson
An expensive plan as well.

Lizzie Gibney
Very expensive. So perseverance has taken a sample. It's got a whole raft of samples now that it's collected on Mars.

And NASA ultimately wants potentially with ESA to bring these samples back.

The plan has gone up and up and up in cost. So I think it's now looking about $11 billion, which is a lot of money. But NASA's now exploring alternative plans. So is there a cheaper way that it could try to bring these samples back? So, anybody out there, dear listener, if you've got any ideas, send them on a postcard to NASA. But, yeah, so it might be that NASA has discovered ancient life on Mars, but we might only have it confirmed in about 1020 years time.

Benjamin Thompson
So a tantalizing potential that life could have existed on Mars. But the flip side of it is it could just be a strange rock, and it's going to be a long time till we find out.

Lizzie Gibney
It could be scientists really excited about this. This is exactly the kind of thing that they were looking for.

This is probably one of the best examples that a rover could hope to find and analyze. You know, there isn't much more that they can do. So I think it's quite a big win. And, yeah, something to look forward to getting those samples back eventually.

Benjamin Thompson
Well, I'm not sure, Lizzie, how I'm going to top that in this week's briefing chat, but there is a through line to my story, actually. It's not an outer space mystery, but it isn't under the sea. And I know the old adage that, you know, the surface of the moon is better studied than the ocean floor, but something is going on in the ocean floor, and it seems that oxygen is being made there. And how this is happening is kind of baffling scientists. Now, something is pumping out large amounts of this stuff, but what it is is currently kind of a shrug. But some researchers have published what they hypothesize is going on in nature geoscience.

Lizzie Gibney
So let's back up a bit. So oxygen needs what to be produced?

Benjamin Thompson
Well, typically, we think about oxygen being produced by photosynthesis, right? Like that's how it gets made, algae and water or plants on the land. But what's weird about this one is that it's at the very base of the Pacific, like thousands of meters down, where there is no light. So it can't be photosynthesis. Right. And this story has got a bit of a backstory, I suppose. In 2013, researchers were studying ecosystems in this area in the Pacific between Hawaii and Mexico, right? And they dropped these probes that kind of sink to the sea floor. They run these automated experiments. They're kind of essentially maybe an upside down glass bowl, maybe, right? So they trap some seawater inside, and oxygen usually gets into ecosystems, certainly deep down via kind of surface current spreading it around. Right. And you'd expect oxygen levels in this kind of trapped bit of seawater to drop as it's used up not in this case. It went up. And the researchers, understandably very confused and thought their senses were wrong. I read in one place that they sent it back, like, several times, saying, there must be something wrong with this equipment. Can you have another look at it? And they kept on getting this result, and different techniques showed it, too. So one of the researchers here is quoted as saying, like, he'd been ignoring a potential new process for making oxygen for years because I guess he thought it was a mistake. So I know the question you're about to ask is, what on earth is going on then, right.

And hopefully I can help out.

Lizzie Gibney
Yeah. What are the theories?

Benjamin Thompson
Well, the current hypothesis is that this oxygen is being produced thanks to these things called polymetallic nodules. Okay. They're about the size of a plum. Our colleague Davide describes them in a nature article he's written about this work. Or maybe a potato as well. You know, that sort of thing. So quite small. There are a bunch of different sorts of metals, maybe nickel, maybe lithium, copper, cobalt, all these kind of things, right? And they've been laid down over millions of years. Right. Dissolved metals in seawater have collected around something, maybe a fragment of shell. So they've gathered around this thing and the way that it's been described in a bunch of places. And I'm going to use heavy inverted commas here. They're kind of acting like batteries, maybe. Okay, so if you think if you drop a battery in water, don't try this at home, team, obviously. But, like, sometimes you have enough voltage to split water and you get oxygen out, and that's maybe what's going on here. So the researchers did a bunch of tests, and as I understand it, these little nodules on their own don't create enough voltage. But if you place them side by side, like you might do in a torch, for example, or a flashlight, I suppose, that would give you enough voltage to split seawater. Now, this is a hypothesis, okay? There are lots of questions to answer about what's going on, right? If this is a chemical reaction, presumably the chemicals would have depleted a long time ago. So what's going on is sort of shrug.

But what's interesting as well is that it raises a bunch of other questions about how we could make use of this potentially, right? Because if this is indeed what's going on, perhaps there's a way to use this chemistry up here on dry land to make better catalysts. All right? But also these nodules are being investigated by various mining companies as a source of these metals, right? Lithium, we need lithium for stuff.

Lizzie Gibney
First thing I thought when you mentioned those metals, I was like, well, that sounds valuable.

Benjamin Thompson
Well, exactly right. And so they are valuable metals, but if they are in some way related to being the source of this oxygen, taking them out could damage the ecosystems on the seafloor. So there's an ecology aspect to this, too. But I did say there was a throughline to your story here, Lizzie, and it's kind of an interesting one that I hadn't considered until I read about it in the article. And it's that if we're looking for extraterrestrial life, in many cases researchers are looking for the signature of oxygen on exoplanets. Right. And I think it's fair to say that, you know, we might assume that this oxygen is made by organic life. It's biologically made. But here, if this is what's going on, there's potentially no biological aspect to this at all. It is just an underwater inverted commas battery. So we need to be even more cautious when thinking about what the signatures of life might be out there in the wider universe.

Lizzie Gibney
Gosh, yes. That is such a challenge. And has this just been seen in one place on the seabed, or are there potentially other sources of dark oxygen that haven't been found yet?

Benjamin Thompson
Well, this is the first time this has been sort of described from this area in the Pacific. But I think what's really interesting is it shows that there is another source of oxygen that isn't photosynthesis. And it's so interesting that there's loads of chemistry out there that is yet to be discovered. We cover it all the time like nobody knew this until now. And so your fingers crossed that, you know, researchers can get to the bottom of this and what may be going on.

Lizzie Gibney
Amazing. Well, thank you so much, Ben. And for more of those stories and for where you can sign up to the nature briefing to get more like them. Check out the show notes for some links.

Benjamin Thompson
And that's all for this week. As always, you can keep in touch with us on x. We're at Nature Podcast, or you can send an email to podcastature.com.

Lizzie Gibney
And if you fancy leaving us a review, you can do so wherever you get your podcasts. I'm Lizzie Gibney.

Benjamin Thompson
And I'm Benjamin Thompson. Thanks for listening.

Nature Podcast
The Nature podcast is supported by nature, plus a flexible monthly subscription that grants immediate online access to the science journal Nature and over 50 other journals from the Nature portfolio.

More information at go dot nature.com plus.

Benjamin Thompson
Hey, corporate types. Billy Idle here just because you use workday to drive long term success, it doesn't make you a rock star. Rock stars drive fast cars, not business operations.

Lizzie Gibney
Be a finance and HR rockstar with workday.