Tiny satellite sets new record for secure quantum communication
31:05 Share

Adios.

BetterHelp Online Therapy bought this 30-second ad to remind you, right now, wherever you are, to unclench your chock.

Relax your shoulders. Take a deep breath in and out. Feels better, right? That's 15 seconds of self-care. Imagine what you could do with more. Visit betterhelp.com slash random podcast for 10% off your first month of therapy. No pressure, just help. But for now, just relax.

Welcome back to The Nature Podcast. This week, a tiny satellite for ultra-secure quantum communication. And how to give an AI helpful feedback. I'm Sharmini Bandel. And I'm Nick Pachaciao.

Ultra-secure communication over long distances may be a little bit closer, as a new paper in Nature is demonstrating a tiny satellite payload that can harness the oddities of quantum physics. And this will be a step towards a practical quantum network. I'm very optimistic we will see the future realisation of quantum networks. This is Jianwei Pan, one of the authors behind the new paper.

Now, a quantum network would be one where many quantum devices are connected to each other. And thanks to the quirkiness of quantum physics, these devices would be able to communicate in a very secure way. You see, data can be encrypted using what's known as a quantum key. This relies on the quantum state of a particle, which can represent 1s, 0s, and something in between, a superposition of 1 and 0.

Someone can use this key to encrypt their data, then send the key to someone else to allow them to decrypt it. This is called quantum key distribution. The reason that this method of quantum communication would be so secure is down to the quantum nature of the key.

You see, whenever you measure a quantum state, it gets disturbed. So any would-be eavesdropper trying to figure out what the key looked like would leave a trace. So the eavesdropping process will unavoidably introduce some noise. So if the noise is beyond some threshold, we know the key is no more secure. Achieving a secure quantum network like this would have many applications, but they're tough to make.

In their work, Jianwei and the team were sending a key in a stream of photons in superpositions from one location to another. But even this in itself is tricky. One of the major difficulties is how to send such a fragile quantum state over long distance. If you were to send these photons through fiber optic cables like the ones used for internet services, then many of the photons would be lost.

In fact, Jianwei estimates that if you were to do this over a distance of a thousand kilometres, only one photon would be able to make its way from one end to the other every 250 years. Probably too long to wait for a message. To overcome this, Jianwei has looked to the skies, trialling sending quantum keys between satellites in space and ground stations here on Earth, using lasers to beam photons between them.

And as space is a vacuum, photon loss is rare because there's less stuff for them to bash into. As a proof of principle, Gemway and a team of researchers launched such a satellite capable of receiving and sending quantum keys to and from a ground station in 2016.

but their setup had its drawbacks. The satellite is very big with a payload more than a few hundred kilograms. And also for ground station with a weight of 13 tons. So it's very, very big.

So it's of no practical application because no one will install such a thing at your home with a weight of 13 tons. The new paper, though, demonstrates that it's possible to shrink things down, describing a device that can be loaded on a satellite with a mass of only 23 kilograms. Its corresponding ground station is also a lot smaller, weighing in at a measly 100 kilograms.

So we really tried to significantly reduce the cost and also reduce the weight by one order of magnitude for the satellite.

And also for the ground station, I mean we reduced the weight of the ground station from 13 tons to less than 100 kilograms. So it's by two order of magnitude improvement. To accomplish this, the team made two key changes to the design. First, they combined the laser that sends the photons with technology that helps control the beam, something that was previously two separate devices.

And secondly, they integrated the technology to make sure that the satellite can track the ground station with another system that controls the orientation that the satellite is pointing. So with these two significant improvements, we managed to really reduce the payload weight. These changes also made the system a lot cheaper. Jamway estimates it cost them only around £1 million, rather than £14 million for the previous version.

With that in place, they were able to send an ultra-secure message from their ground station up to their slimmed-down satellites and back down to their collaborators' base station in South Africa, some 12,900 kilometres away from them, the furthest distance a quantum communication has been sent yet.

And as for the content of that secret message... We did not really send a secret message. We just sent some text, right, some figure and so on. In fact, amongst other things, the team sent a picture of the Great Wall of China. Maybe not quite top secret. But in principle, such a communication method would allow anyone using it to send messages without fear of eavesdroppers.

And by making the satellite and ground stations smaller and cheaper, it makes creating a quantum communication network that much more practical.

Nature physics reporter Lizzie Gibney has been writing a news article on this new paper and she gave me a sense of what the reaction has been from physicists. Yeah, they've been excited. The idea of having a satellite involved is you can connect any two points around the world but to do that you need to have ground stations that can go anywhere. So now this is 100 kilograms, that really is actually portable versus previously it was many, many tonnes.

So what people have told me is that, yeah, this does seem to be a milestone, a step on the way towards actually having a quantum network that connects up the whole globe. But to get to a global quantum network, there are a few more steps. For a start, the current satellite is limited in how often it's connected to

to the ground stations. This satellite is rotating around Earth, so it only goes overhead for maybe an hour or two every 24 hours, and so you're a bit limited in your communication with it.

So either you need to have a network of satellites in the sky, a constellation, in order to have constant coverage, or you need a satellite that's much higher up in this orbit called geostationary orbit, where it stays then permanently above you as Earth rotates. Jamwe and his team are planning to send up four new microsatellites in 2026. And in 2027, they are also planning to send up a geostationary satellite in order to build such a network.

And if a full network covers the globe, using satellites or any other technologies currently being explored, it would enable people to communicate extremely securely, which could be a boon for privacy, but also could be used for other purposes such as military applications.

To be frank, that is perfectly possible. And, you know, we haven't seen a quantum satellite come out from the US, but that is to say that we know of. So it's perfectly possible this stuff is happening behind closed doors as well by other countries around the world.

What we know at least about what's happening here with Zhang Weipan's group is that they are publishing about it. They're very open about what's going on and they are talking about it as being useful for privacy and for commercial applications. So for banks, for governments to use.

On the ground, networks like this are already used in Switzerland, for instance, to ensure that its elections are fair and secure. Jamway believes that the content of messages sent are really up to the people who send them. Privacy, he says, is a human right.

And he also sees this technology, relying on international collaboration as it does, as being a way of bringing people together. I mean, we are very open. From time to time, all our guests from abroad can really see our laboratory, can take pictures and so on. So I still believe, I mean, yeah. So international collaboration is very important for the future of quantum information.

To me, I mean, the quantum information technology will benefit to the whole mankind. That was Jiang Wei Pan from the University of Science and Technology of China. You also heard from Nature's Lizzie Gibney. For more on that story, check out the show notes for a link to Jiang Wei's paper and Lizzie's news story. Coming up, a new way to improve AI by giving it written feedback from another AI. Right now, though, it's time for the research highlights with Dan Fox.

Powerful storms called atmospheric rivers could replenish Greenland's vanishing ice sheet if their timing is right. Atmospheric rivers are narrow streams of water vapor in the atmosphere. When they hit land, they release their water in the form of heavy rain or snow.

In Greenland, extreme rainfall from these rivers can accelerate the Greenland ice sheets melting. But when the temperatures are low enough, the rivers produce snow that can recharge the ice sheet. Researchers looked at an ice core collected in southeast Greenland that contained a layer of snow dumped by one of these atmospheric rivers in March 2022.

Analysis suggests that the storm dropped more than 11 billion tonnes of snow and as a result the ice sheet lost 8% less mass that year than would have been seen due to the effects of climate change if there hadn't been a storm. The extra snow also made the sheet more reflective, delaying the start of the melt season by 11 days. You can read that research in full in Geophysical Review Letters.

An innovative prosthetic hand can distinguish a wide range of objects by touch almost as well as a human's. The human hand's combination of rigid and soft anatomy allows it to effortlessly grip and identify objects of differing shape, texture, and weight. But prosthetic hands typically lack the real appendage's hybrid structure, preventing them from easily executing those tasks.

To tackle this, researchers developed a prosthetic hand that unites a rigid internal skeleton with soft robotic joints. The device has three layers of touch sensors inspired by the layers of human skin and can be controlled using a person's forearm muscles. The new hand can securely grasp 15 everyday objects ranging from a soft toy to a metal bottle and tell them apart at an accuracy close to 100%.

And the prosthetic can differentiate between textures correctly more than 98% of the time, surpassing other robotic fingers. Get a grasp on that research in Science Advances. Next up, reporter Adam Levy is here to explain how AI can improve itself. I would love to pretend that I made this podcast piece perfectly first time round, that I handed it in and there was simply no room for improvement.

But let's be real, that's not how these things work. I hand it in, then my editor, thanks Nick, tells me all the things that can make the piece better. Then I try to do all those things, and as much as I hate to admit it, Nick's feedback does end up making my work much better. And that's why feedback is important, so that we can improve our work.

And it's just as important for the work coming from artificial intelligence. But the more complex AI systems get, the harder that becomes. ChatGPT and other AI systems have billions of parameters that essentially encode all their knowledge. And lots of the cutting-edge systems are built of several interacting components.

And all this makes it hard to automate this feedback so that an AI can give feedback to another AI and improve its results. Well, this week in a paper in Nature, James Tso and colleagues have created a new tool called TextGrad, which uses one artificial intelligence to give feedback to another by effectively talking with it, just like Nick spoke with me to improve this introduction.

To improve my understanding of how they did it, I called James up and asked why optimizing modern AI systems is such a challenge. So the standard way of optimizing AI, basically for the last 15, 20 years, is really by using a single algorithm, which is called backpropagation.

And the main idea of backpropagation is that you essentially want to pass what we call numerical signals or numerical gradients from one part of the model to the next part. So it turns out that this approach no longer applies to many of the current generative AI systems. If I'm a user of a chat CPT, I cannot actually go in very easily to change all of these billions and billions of parameters.

And a second big barrier is that generative AI models like ChatGPT, they can actually use external tools, right? For example, you can have ChatGPT that uses like a Google search engine.

And I cannot really go in to optimize a Google search engine. Google won't let me do that. So in a paper out in this week's Nature, you're publishing an alternative way of optimizing AI tools. Can you outline the fundamental idea behind this tool? We proposed a new and very generalizable framework for optimizing any of these generative AI systems by essentially optimizing them through human language feedback.

When you say that we're using human language and essentially chatting with these AI systems,

Is this a person doing this? Who or what is actually giving the feedback? It's not a person giving that feedback. We basically rely on another large language model to provide this kind of feedback. Then what happens is that you have a first AI system that's performing some task and you have a second AI system, like a large language model, that looks at feedback.

the performance and their outputs from the first AI, and then provide some critiques. We can view this as sort of like a way of AI to self-improve. How does this tool you've developed, TextGrad, know what feedback to give in the first place? How does it actually evaluate how good a job the AI system is doing? It's often actually turns out to be easier to give critiques than to solve the problem, right? So we see this in both human tasks, but this is also true for AI algorithms.

So what we found is that the current AI large language models are actually quite adept at providing these kind of critiques. Let's take a look at one particular application of this tool, perhaps the most tangible application, which is in radiation therapy treatment for cancer. A good therapy should give enough radiation to the target area while protecting surrounding regions from radiation.

Could you walk through how TexGrad works to improve the results that an AI system would give for radiation therapy? Design of these radiation therapy plants is actually quite challenging.

In the standard way, the clinicians often have to manually tune certain parameters to try to balance that trade-off. In this case, what TexGrad does is that it actually looks at the proposed radiation plan. It actually provides some feedback as to, maybe you're getting too much radiation in this organ, and you should reduce the radiation in this area by doing this. So then that feedback, which is in the form of natural language, then goes back into what

algorithm that's generating the radiation plan to optimize it. And we found that by doing this, using tax grads, we can achieve quite good, sometimes as good as the clinicians optimize radiation plans that minimizes side effects. And when you say it's using natural language, is it phrasing it in the same kind of terms that you just phrased it to me? That's exactly right. Yes.

This is sort of the verbatim text produced by the algorithm. So it says that the current weight for the rectum and bladder are relatively low, which is not sufficient to protect these organs from receiving too much dosage. And the benefit of that is that first, it's easy for other large language models like Chachbi to understand that feedback.

But it's also easy for the human researchers and clinicians like ourselves to look at that feedback and say, is this reasonable? Whereas if I just see like a big list of numbers, it's hard for me to make sense of what's going on. Now, the way you've explained it here, it sounds very intuitive. These AI systems use natural language. So automate the feedback to them with natural language.

So why hasn't this actually been done before? Essentially, what we had to do is to create what we call sort of like a calculus of language. The calculus that we learned in school, that's sort of the underlying basis of the traditional backpropagation algorithm.

So when we had to replace those numerical gradients with these text gradients, we essentially had to create a new calculus of text, where you can take one piece of textual feedback, decompose that into two or more separate feedbacks that targets individual components of the AI system to update each of those separately. One of the, I suppose, crowning achievements of modern AI systems is just how diverse their applications are.

Could you give a sense of the diversity of applications where you've already applied TexGrad to give feedback? In our paper, we also demonstrate how we can use TexGrad to design new molecules that have good drug-like properties.

And we also show on how we can use TextGrad to optimize other AI agents and also to improve question answering and coding tasks. What does it mean to you as a researcher to see TextGrad working in all these very different systems?

I think it's very exciting. And in many ways, it's also, I think, very thought-provoking to me as a researcher. Previously, optimizing machine learning algorithms requires quite a lot of human engineering and human ingenuity.

But with TechSquad, now the AI is actually able to self-improve to a large extent. And I think that really opens up a lot of very exciting possibilities to solve harder and harder problems. And actually, TechSquad is open source. There's been a lot of talk about the inequalities in access to AI systems and in the ability to develop these systems in different parts of the world.

Could an open source tool like this in any way level the playing field? That is indeed one of the main motivations for us to develop this tool and to make it open source. And what we've shown is that you can actually use this tool to optimize open source AI models and also gain substantial improvements from these open source models.

That was James Zoe of Stanford University in the US talking to Adam Levy. For more on that story, check out the show notes for some links. Finally on the show, reporter Max Kozlov joins Benjamin Thompson for the latest update on the state of science in the US in the wake of the Trump administration's ongoing budget cuts.

Max, thank you so much for joining me once again. Of course. So we are talking on Monday afternoon. And of course, this is and remains quite a fluid situation. And things may have shifted by the time listeners hear this. But let's talk about what's been going on over the last little while. The first thing to chat about was the Stand Up for Science March. A bunch of these

events took place across the U.S. And you were actually at one of them in Washington, D.C. Yeah, there were thousands of people that showed up on the National Mall right in front of the monuments in Washington, D.C. to protest some of the actions that this government, this administration has been taking, specifically about bio

medical science and some of the actions at NIH in pausing research or terminating research. And I saw a lot of signs about the National Weather Service because people are worried about impacts on their ability to do their work and to provide data for scientists and researchers to use. So there was a lot of energy there. So yeah, you mentioned the NIH, the National Institutes of Health,

there. And you've been reporting extensively on what's been going on there, specifically around things like grant terminations. Yeah. So a few weeks ago, I reported that they were halting these grant review sessions to review applications for new research to be funded. And they were kind of gumming up the process and making it really difficult to schedule those meetings. And that still is actually going on to this very day. But

What you're touching on is they've also begun terminating existing grants. So this is grants that have already gone through the process of being reviewed by scientists, and they've been deemed the creme de la creme of the research to be funded. And now the NIH has been instructed to go through and look for certain keywords and to send these termination letters, often with only hours notice. And the

And this has to do with Trump's many directives and executive orders, but specifically the ones pertaining to what he calls transgender issues or gender ideology. There's a laundry list of words or topics that

They've been instructed to flag now. We talked about this anti-DEI focus across the board last time you were on with us. And so what researchers are literally getting letters in the post saying you can't have any more money to continue your already accepted study. Pretty much. I mean, and these are researchers who are on year three, four, five of their project.

And they're being told, stop working on this. We're pulling your funding, essentially. And I think a lot of scientists are perplexed for an administration that has talked so much about efficiency and stewarding taxpayer dollars. There's nothing more inefficient, they say, than stopping a research project in the middle before they've been able to collate or to

gather and present their data. And in your story, I mean, you say that at the time that sort of a handful of these termination letters have been sent out, but it could number in the hundreds. Is that right? Yeah. There's kind of, like I said, an ever-growing list of these topics. The most recent one was they terminated a slew of vaccine hesitancy projects

These are projects that research why people might be reluctant to get a certain vaccination. And this is important because for many infectious diseases out there, say measles, it's important to have a certain proportion of the population to be vaccinated for the vaccines to actually do their job of protecting people. And so it's important, researchers say, to study this. So

public health officials can tailor their messaging and understand why people might be hesitant to get the vaccines. And so dozens of those projects have now been terminated. And that's in addition to studies focusing on LGBT plus health, transgender health,

and research that involves studying DEI, diversity, equity, inclusion, in the scientific workforce. And that's a big deal because the way they've defined it is so ambiguous that it could really encompass a pretty wide range of studies that the Trump administration doesn't agree with. And obviously, you've spoken to many researchers who have been affected by

Do you have any stories that really stood out to you? I spoke to a program official at the NIH being tasked with combing through some of the existing grants as they come up for renewal and flagging them for termination. And, you know, because there's been so little specific and clarifying

And of course, this ambiguity then means that potentially everything's on the table, right? And there are different categories of what the administration deems as cultural rights.

acceptable or not acceptable. We published in our story the categories that NIH staff are being tasked with grouping or lumping the grants into that might involve terminating the grants. And let me just also be clear that terminating grants is extremely rare. It normally does not happen unless there's been some kind of serious misconduct or

or there's been no progress made, or some kind of fraud. And so the fact that there are these en masse terminations happening, that is absolutely unprecedented. Unprecedented in scope as well. It's not just this focus they have on all things

DEI. There's other stuff as well. Yes, it's not just limited to DEI. Grants that fund research that goes on in China or research that relates to climate change or environmental justice, all of these are being flagged. And it's still unclear whether all of this will be terminated. Research that has been terminated so far includes DEI work, research that includes transgender individuals and vaccine hesitancy. Now, NIH staff get these requests for

for list all of the studies that have to do with this, and then they send them along. And a recent one that came up was mRNA vaccines. So this is the vaccine technology that gave us the COVID-19 vaccines. But this vaccine technology is being piloted for many other different diseases and infections, including everything from dengue virus to cancer. And

These grants have not been terminated yet, but recently NIH staff have been asked to produce a list of all projects about this. And a lot of people, both inside the NIH and researchers outside, are worried that this means that mRNA vaccine research could be next, given some of the skepticism among senior Trump officials about vaccines and specifically COVID vaccines. And of course, all

All of these grants represent a huge amount of funding dollars, right? And in another story, you were looking at what the effects of suddenly removing or threatening to remove a large amount of funding can have to an institution. And you were involved in a story about Columbia University in New York. Yes, exactly. We have covered this extensively that the effects will be felt most strongly by the next generation of scientists.

So we've already seen dozens of universities across the United States rescind or cancel their graduate student programs or rescind postdoc positions because there's so much uncertainty about funding in higher education right now. And higher education is being singled out in this country such that researchers don't know if they can sustain keeping their labs open or

hiring new talent in their labs and early career researchers are having to make difficult decisions about the future of their careers. I know you mentioned Columbia University. There, the NIH has canceled more than $250 million in funding, which is about 400 research grants, because of what the Trump administration calls continued inaction in the face of persistent harassment of Jewish students.

So they're terminating grants at Columbia for something that has nothing to do with the research that they're doing there. It's just that they are targeting the university specifically. And the Trump administration is investigating at least 50 other universities as well. So this might not be the last termination of grants at one specific university either. Well, finally, then you've, of course, been speaking to researchers across the country.

Is there a prevailing thought, a prevailing through line that they've told you, do you think? Something that I've been hearing is even if the clock turns back tomorrow, the damage done here will be lasting in that how can researchers outside of the United States trust or collaborate with researchers in the U.S. now that

They understand that their funding is uncertain and they might not be able to publish manuscripts using certain keywords because they might be blocked or censored in the United States. And so I think that the effects of these actions in the first two months of the Trump administration will be felt for a very long time. Well, let's leave it there. Max Kozlov, thank you so much once again for joining me. Thank you.

That's all for this week. If you want to keep in touch, you can follow us on X or Blue Sky or you can send an email to podcast at nature.com. I'm Sharmini Bandel. And I'm Nick Petrychow. Thanks for listening.

I know I'm not alone when I say adulting can be overwhelming. And what we all could use is a drink. That's where Apple & Eve juice comes in. As the rulers of the juice box, they've been making juice joyful for 50 years. With refreshing juice blends bursting with bold flavor, one sip sends you right back to childhood. So when the grind dulls your shine, remember to kid yourself. Apple & Eve has delicious juices for at home and on the go. Shop today.

Dinnertime. It's where little moments are cherished. With Blue Cash Preferred, get 6% cash back at U.S. supermarkets and bring everyone together. I did say everyone. Learn more at americanexpress.com slash explore dash bcp. Terms and cash back cap apply with Blue Cash Preferred.