Science Friday - Is Each Fingerprint On Your Hand Unique? | In This Computer Component, Data Slides Through Honey

Episode Date: February 8, 2024

A new study uses artificial intelligence to show that each of our ten fingerprints are remarkably similar to one another. Plus, honey could be the secret ingredient in building a more eco-friendly “memristor,” which transmits data through malleable pathways.Is Each Fingerprint On Your Hand Unique?We often think about each fingerprint as being completely unique, like a snowflake on the tip of your finger.But a new study shows that maybe each person’s fingerprints are more similar to each other than we thought. Researchers trained artificial intelligence to identify if a thumbprint and a pinky print came from the same person. They found that each of a person’s ten fingerprints are remarkably similar in the swirly center.Ira talks with study author Gabe Guo, an undergraduate at Columbia University majoring in computer science, based in New York City.In This Computer Component, Data Slides Through HoneyA honey bear is probably one of the weirder things you’d see in a science lab, especially in a lab making computer parts.“It’s just processed, store-bought honey,” said Ph.D. student Zoe Templin. “Off the shelf — a little cute bear so we can put it in photos.”But for Templin and her colleagues at Washington State University, Vancouver, the honey is key.“It is cheap and it is easily accessible to everyone,” said master’s student Md Mehedi Hassan Tanim.The honey also has natural chemical properties that make it a promising foundation for a new kind of environmentally friendly computer component — one that could make computing faster and more energy efficient while reducing the impact on the environment.Read the rest of this article on sciencefriday.com.Transcripts for each segment will be available the week after the show airs on sciencefriday.com. Subscribe to this podcast. Plus, to stay updated on all things science, sign up for Science Friday's newsletters.

Can honey be the sweet solution to making computers cleaner and faster? I mean, honey, you just need a little water, right? And it dissolves. It's Thursday, February 8th, but it's also Science Friday. Sweet. I'm SciFri producer Shoshana Bucksbaum. Electronics are hard to recycle. That's because computer components like silicon are toxic when they're broken down.

So researchers in the Pacific Northwest are replacing silicon chips with something you can pickup in the grocery store. Honey. Substance has synaptic properties that make it a powerful tool for electricity conduction. We'll talk about that story in just a bit. But first, Ira talks with a computer science undergraduate who figured out that each of our 10 fingerprints are more similar than we previously thought. We've been told for decades in all kinds of ways. TV, court dramas, newspaper reports, you name it, about how each of our finger, is unique, kind of like a snowflake, right? Right there on the tip of your finger. But a new study shows that maybe each person's fingerprints are more similar to each other than we thought. How?

Researchers trained AI. What else? To identify if a thumbprint and a pinky print came from the same person. Joining me now to talk about what they found is Gabe Guo, study author and an undergraduate at Columbia University, majoring in computer science. based in New York. Gabe, welcome to Science Friday. Yes, thank you for having me today, Ira. Nice to have you. All right, let's get into this. What did you find? How similar are one's persons' fingers to each other? Yeah, so in summary, we trained in AI to find very statistically significant similarities among fingerprints from different fingers of the same person. So that means,

for instance, we now have characteristics that can link your pinky finger to your thumb. And as for what those characteristics are, it's mostly due to the ridge angles near the center of the fingerprint, a region known as the singularity. The singularity. We love that word in science, don't we? Yes, we do. How is artificial intelligence able to determine if the fingerprints from the same person were similar? We use a type of artificial intelligence known as a deep contrasting network. And I know that sounds really hard and fancy, but it's actually pretty simple. I'll break it down for you. All it is is an AI that tells you if two pictures are of the same thing or of a different thing.

So for instance, if I take a picture of my dog Bob today and they take another picture of Bob tomorrow, and they pass them both into this AI, it'll say, this is the same dog. But if I take a picture of my dog Bob in the picture of my cat, Jin, and pass them into the AI, then it'll say they're different. And we use that exact same strategy for fingerprints. Now, as for how we get this AI model to actually learn this, we just fed it a lot of data from a U.S. government database. And this data had fingerprints from all different fingers, some of them from the same person, some of them from different people. And over time, after looking at this data, the AI just got better and better at finding the similarities and differences. So let's go into what prior research suggests about the similarity of fingerprints.

didn't we already know that one person's fingerprints are similar to one another? What's new here? Yeah, so there was previously a hunch that this was the case, but up until us, nobody was able to quantify it, identify the features that made them similar, and certainly nobody built an automated matching system. And yes, there is previous research that influenced us, so I'll list some of them. There was some research that came out a few years ago

that talked about the genetic determination of fingerprint patterns, and that was actually what led us to investigate this. because if all fingerprints come from the same DNA but then the same person, then there should be some similarity. And then another work that also inspired us to do this was a work that showed that if you have a fingerprint matching system trying to distinguish twins, I can still distinguish them, but the error is slightly higher when you're distinguishing twins versus people who aren't twins.

Interesting. Speaking of interesting, how did you get interested in studying fingerprints? It's not something anybody sort of wake up. up in the morning and says, hey, I want to study fingerprints. Well, maybe you did. Right, totally. Yeah, so as with many Americans, this project of mine actually came about because I was super bored during the COVID pandemic.

You know, I scooped up at home in Buffalo, New York, wonderful town. So I was having a chat with a professor from the University of Buffalo, Wanyoschew. He's a great family friend. And at the end of the chat, he just casually posed to me this question. Gabe, do you think all fingerprints are really unique? And they said, what do you mean, Professor Shoe? And they said, well, specifically those fingerprints on the same person. I mean, they come from the same genetic material.

Surely there must be some sort of discernible similarity, right? And they said, I don't know, Professor Shue, I'll have to investigate that. And little did they know that offhand 30-second exchange would literally take over the next three years of my life. So, you know, I worked on it while I was still in Buffalo that COVID year. and then I loved it so much that I asked Professor Shrew if I could take him to Columbia with me when I started at Columbia. And he gave his consent. So then we collaborated with Professor Hodlipson, who's also very wonderful, and now we're here. Wow. Wow. So tell me how the process of getting this published work. You got rejected a few times, right? Yeah. And I think that whole process was very funny because, you know, while we were trying to get it published, one of the most common criticisms from,

reviewers is that, oh, we know that every single fingerprint, even from the same person's unique. And, you know, that's, that's kind of the, it speaks to the power of AI, right? It's really upending these conventional beliefs we previously had. So when we were faced with that criticism, we just kept running more experiments, providing better visualizations, and eventually it got to the point where our quantitative and qualitative evidence was so strong that it was incontrovertible. Yeah, you came up with the same roadblock that we see with doctors facing AI. Oh, I'm better than AI because I just know my stuff. Well, then you test it out and you find

out, uh-oh, maybe I was wrong all these years. Right. And I think to your point, I don't see AI as something that will replace human experts. Rather, I see it as something that can augment the capabilities of human experts. Because actually after this study came out, We got a lot of interest from, you know, police departments, forensic scientists, biometrics companies who, you know, are inspired by our work and want to use it to further the state of the art. All right. We've all watched Law and Order or some other of these crime shows, right? And we've seen them taking all these fingerprints. Tell me how your research might be helpful in the field of forensics or even beyond.

At crime scenes, fingerprints are usually kind of hard to pick up. So typically at a crime scene, you'll only get maybe one or two discernible fingerprints, let's say the right index finger. But the issue is, what if law enforcement or your private investigator only has fingerprints from different fingers on file? Let's say they only have pinkies. Well, with previous techniques in fingerprint matching, it would be impossible to link them. However, with our technology, we're the first in the world to find a way to actually match

these fingerprints from different fingers of the same person. thus we can, you know, catch this criminal and make sure that the criminal doesn't cause problems. But of course, you know, this is America and we believe in freedom here, not just putting people away. And we show in our paper that when we use this method, we can actually narrow down the number of false leads that the law enforcement has to investigate by over tenfold. So that means for every one criminal that we choose to prioritize investigation on based on our technology, that's at least nine other innocent people who don't have to be unfairly investigating. And I think that's a win for everyone. Is it precise enough yet to be used in crime scenes and in court? As currently constructed, we can't use it as deciding evidence in court.

But as we show on your paper, it's very useful for generating leads. And I also want to add that in our research, we saw a trend where as you added more data, the precision and accuracy went up and up. And we only trained on 60,000 fingerprints. So I'm sure that the FBI decides to train this with, you know, there are millions of fingerprints. They probably could get it to a level where it might be used as well. evidence. But I'll wait for them to make that call. Well, they haven't, your phone's not ringing then from the FBI. I'm not allowed to disclose that information. Is it ringing from other law

enforcement people? Or are you now in a some sort of patenting situation where you can't disclose anything? Yes, it is being patented and yes, law enforcement agencies are contacting us, but I can't speak more about that. I know you're graduating and may, congratulations. Yes, thank you. What's next for your research and your career? If you can tell us, that is. So I'm going to be pursuing a PhD in computer science in the fall.

And as for this research, you know, this isn't just a new paradigm of fingerprint recognition or even biometrics. It's really a new era in AI because for, you know, most of AI history all the time, effort, and funding. went into teaching AI models to do things that, no offense, any human dolly could do. You know, is this a cat or is this a dog? Cool, but now we have AI that isn't just recurigitating information, not just doing simple things. It's literally making new scientific discoveries.

And not just new scientific discoveries, new scientific discoveries about our fingerprints, which were in front of our plain eyes for hundreds and hundreds of years. And yet none of us noticed this until we had our AI look at it. So we now have AI and knows our own bodies better than we do. And to me, the implications of that are just massive. So I want to embark on a journey of AI-assisted scientific discovery. That's what's next for my research.

That's terrific, Gabe. And will you keep us in the loop, come back and tell us what you're finding? Oh, most definitely. Good luck to you and to your graduation. Yes, thank you very much, Ira. Pleasure to speaking to you. You too. Gabe Guo, study author and undergraduates.

soon to be a graduate at Columbia University, majoring in computer science right here in New York. The world produces 50 million tons of electronic waste each year. Only a small amount of that can be recycled because components like silicon aren't easily repurposed. What if the solution to this could be an inexpensive sweet treat you can buy at the grocery store? Honey could make computers not only more environmentally friendly, but fast. One laboratory at Washington State University is working to make this a reality. Joining me to talk about this is Jess Burns, Science Reporter at Oregon Public Broadcasting based in Portland. Welcome back to Science Friday. Hey, thanks for having me. This is a honey of a story about computing, isn't it? Totally. Totally. Sorry. Who would have thought about honey and computers? Tell us about that. Yeah, you know, broadly, they're thinking about using honey as a substitute for, you know, silicon and other semiconductors that are used in electronics.

So these are the materials that control the flow of electricity. But more specifically, the researchers are using honey to create a newer kind of electronic component called a memorister. They dilute the honey, they layer it between electrodes, and then they essentially bake the whole thing. to try to dry the honey out. That's ridiculously simplified, by the way. But, you know, watching them make these chips, it was just such a weird juxtaposition of the absolutely ordinary because they got a honey bear on the table, on the lab desk,

from a big box store. And then you have this crazy high-tech equipment all around us. We're working in a clean room in these clean suits. It was just, it was so bizarre to be there for this. And these things called Mem Risters, why are they such a hot topic in computing there? This took me a lot of time to figure out what Mem Risters do because in many ways they feel a little bit like magic. So it's a circuit and it was theorized like back in the 1970s, but it took until about 15 years ago for someone to actually figure out how to make one. there are a key component of this new kind of computing system called neuromorphic computing,

and that's modeled after how our brains work. Floating around in our skull, we have the most efficient computer on the planet, right? The amount of information we can process and store in our brains compared to the amount of electricity and energy it takes to do that is pretty astounding. It just blows our normal computing systems away. So, memoristers essentially mimic the way our neurons, specifically our synapses in our brain work, so the connecting pieces between our neurons. They can store information, they can process information, all in the same location.

And computers that we have now do these things separately. And consequently, they're a lot slower than neuromorphic computers would be. So what is it about honey that makes it? It's so good here because it has this synaptic property? Yes, it is, but to get a little bit into the chemistry, I'm going to let the expert talk us through this one. This is Washington State University engineer Fing Zao. So they have a big molecule chains in the material, and we need those molecule chains in order for us to make our devices. Yeah, so basically the honey is a substance.

that you can engineer to have these memoristive qualities. You know, they behave like a memory and processing information when different levels of electricity are applied to them. They also like honey because it's relatively cheap. It doesn't spoil. And you don't have to mine it and process it like you do minerals. Like when you think about it, like bees are actually helping the environment. We're not tearing apart the earth to get these materials. What I'm wondering about is I know that honey is mostly.

sugar. So could you swap out different high fructose corn syrup, for example, instead of the honey? You know, they tried fructose. They tried all kinds of things in this kind of sugar category. They tried glucose. They even tried things like aloe vera. But in the end, honey ended up showing the most promise for what they were looking for. Okay. So how does the memorister compare to something more traditional like silicon, like a computer chip? Well, you know, This is such a new technology that I haven't seen head-to-head comparisons, but it's probably safe to say the silicon tech is further along in its development, right? Because people are using silicon to make memberisters.

But a really interesting comparison, I think, comes when you're ready to throw the component away, right? The end use of the life cycle. So, I mean, you mentioned 50 million tons of e-waste each year. Right. Only 20% is recycled. And our current silicon-based computing systems, they contain toxic components. They require some pretty caustic chemicals to break them down into something that's recoverable.

I mean, honey, you just need a little water, right? And it dissolves. There you go. Wow, that would be interesting. Now, of course, is this Memrister, is it faster than our silicon counterparts? The computer system, the architecture, if you will, that it's being designed for neuromorphic computers, those systems have the potential of being much faster than the style of computers we currently use. The reason for this, in part, is because of the processor and the memory,

they happen in the same spot. So the computer doesn't have to go and actually, like, physically go retrieve pieces of information in another location. And, you know, we're talking tiny distances, but when it's compounded so many times and so many processes that a computer has to go to, it is faster because, you know, the information isn't having to travel so far. Yeah. So how soon are we going to see this? And is it going to melt on the dashboard of my car? It won't melt.

It won't melt. And I asked them about moisture, though. And they looked at me and they were like, already most of our computers can't take any moisture. And I was like, oh, yeah, that's right. That's right. So anything that's, you know, the ceiling that we would use for our standard computers would be enough to protect this from moisture. As far as like getting this out into the world, there's a long way to go. It's kind of the story of science, right? You get this idea, you test it and you see if it works and it shows a lot of promise. But then there's probably like 20 more steps before it actually gets implemented. Well, Jess, we'll keep looking out for this and have you report back if something happens, okay?

Sounds good. Desper and Science Reporter at Oregon Public Broadcasting based in Portland. That's all the time we've got for today. Lots of folks help make the show happen, including Kathleen Davis. Diana Plasker.

Beth Ramey. Danielle Johnson. And many more. Tomorrow, a roundup of the top science news of the week. Catch you next time. I'm sci-fire producer Shoshana Bucksbaum.