Instant Genius - How AI is changing the world of scams

Episode Date: August 13, 2023

AI, deepfakes and advanced technology has made scams more advanced than ever. We spoke to Oli Buckley, a professor of cyber security at the University of East Anglia to learn what these scams look like and how to avoid them. Learn more about your ad choices. Visit podcastchoices.com/adchoices

It's peak pollination season, and my business is scaling fast. To keep the nectar flowing, I need a phone plan with top priority data speed. That's why I chose Google Fi Wireless. My connections stay strong even when the hive is buzzing. Plus, unlimited plans start at $35 a month. Now that's a deal that doesn't stay. Explore Google Fi Wireless plans today. Plus taxes and government fees.

GoogleFi Wireless is not subject to data traffic deprioritization during times of high network usage. You said this place was steps from the water. We just haven't found the steps yet. How much did we save? Enough. Enough to get lost. Or you could book a stay with Hilton.

Welcome to your oceanfront room. Just steps from the water. The Hilton sale is on now. Book on Hilton.com or the Hilton app and save up to 20% to get the stay you expected. When you want savings, not surprises. It matters where you stay. Hilton, for the stay.

When you need to build up your team to handle the growing chaos at work, use Indeed sponsored jobs. It gives your job post the boost it needs to be seen and helps reach people with the right skills, certifications, and more. Spend less time searching and more time actually interviewing candidates who check all your boxes. Listeners of this show will get a $75 sponsor job credit at Indeed.com slash podcast. That's Indeed.com slash podcast. Terms and conditions apply. Need a hiring hero? This is a job for Indeed sponsored jobs. This podcast is sponsored by name, audio and focal. Streaming has made music more accessible than ever,

but true listening is about more than ease. It's about quality. British audio experts name audio, alongside French acoustic specialist focal, combine handcrafted tradition with cutting-edge innovation and high-end materials, delivering digital precision with analogue warmth, so you can experience exceptional sound at home. Music just as the artist intended.

visit name audio.com to learn more. From BBC Science Focus Magazine. This is Instant Genius, a bite-sized masterclass in podcast form. I'm Alex Hughes, staff writer at BBC Science Focus magazine. This week, we're talking about scams and how future technologies are changing how they work. I'm joined by Ollie Buckley to discuss this topic. He's a professor of cybersecurity. at the University of East Anglia.

He explains how deepfakes, AI, and a host of other technologies are changing the world of scams, adding in new risks and creative tricks for you to watch out for. A deep fake is something that uses an artificial intelligence algorithm to create content that looks realistic or sounds realistic. So that could be something like a video or it could be recreating audio. They need very little training data to get them to work. and you could create something that's quite convincing with a standard kind of laptop you'd just buy in curries or something. And so with this kind of technology, it could be replicating someone's

face, someone's voice, it's a variety of things. Yeah, absolutely. So deep fakes kind of hit the news and gain prominence on the video side. So things like taking an actor's face and putting it somewhere else. As with most evolutions in technology, this took kind of a slightly see. deedy route to begin with putting actors faces into porn or something like that. And that, but sitting alongside that is this idea of an audio deep fake as well where you can use AI to mimic someone's voice to a really good standard with, again, very little effort and data. And so for, to create this kind of technology, what they're doing is, I assume, in the context of it, taking large amounts of video or audio, which is why you tend to, tend to,

to see deepfakes that are celebrities or people that have lots of footage of themselves out there. Absolutely, yeah, that's absolutely right. So a lot of the stuff that we see is based on on data that's out there. So as you right said, that celebrities have a lot of stuff out there so you can get a movie or a podcast or whatever of someone's image or their voice and create quite convincing deep fake. Pretty widely shared one in 2022 of Ukrainian president Zelensky, telling Ukrainians to lay down arms. And again, that was footage taken, data, and to create this fake version of him giving this message. If you actually look at the video, it's not that convincing.

You can see the movements don't look quite natural. The intonations don't look quite there. So it wasn't the best version, but you could do it much better. But as we all live our lives online now, you can get data about someone quite easily. So I don't know, one of my students could download one of my lectures. And that's more than enough content to recreate my voice, I would guess. So a couple of companies report that you can create something that's about 80 to 90% convincing with about three to 10 seconds of audio. So it's not loads of stuff that you need to create something that's convincing.

And so that can be really, it can be really quite powerful. And if we step beyond just the technical accuracy, if you start to pull on people's emotions and add time pressure and make things a little more stressful and chaotic, then that can be a lot more convincing. to people. So whether it's, I guess, AI, deepfakes, any of this kind of technology, how do they actually work at their core? So a lot of these technologies work with something called a neural network. A neural network is an AI algorithm, an AI approach that tries to imitate the way that our brains work to solve problems. It's a really powerful tool. It learns patterns in data. So let's think a really simple example, you're trying to teach a computer how to recognize different fruits. You might show it lots of pictures of apples, of oranges, of bananas.

And the neural network is like a virtual brain with interconnected neurons. Each neuron takes a piece of the picture, so an element, a feature, something about the fruit and the way it looks, and uses that to classify it. So we might take the color and the shape and the size of each fruit. And very broadly speaking, a neural network is split into different layers, an input layer where you take in the data, and then lots of interconnected nodes that work on different features of what we're looking at. So in the case of our fruit example, the first layer might look at the colour, and then the second layer might combine the color with the shape, and then the following there might combine everything together to try and guess what it is. And then we have an output layer

which says, I think the fruit that you show me is an apple or an orange. And the way that these work is the more data that you can train them with, the more data you can feed them, the better they get at identifying things. So it is really like teaching a toddler how to recognize shapes or recognize fruits or cats or dogs or different animals. And so that's how a neural network works under the hood. That's what they're looking at. Deepfakes and some of this AI technology use something called a gamut.

a generative adversarial network. And this is a special kind of neural network that is made up of two parts. We have the generator and the discriminator. And these are two algorithms that are working, that are two halves of the same algorithm that were competing against each other, trying to beat each other. So the generator creates some data, and the discriminator is trying to see if it's false.

So let's say we are taking a deep fake. We're trying to create an image of someone doing something they hadn't done. the generator would start out creating some random data. The discriminator would say, no, that's absolutely not correct. There's a 0% probability of that being the real image, and it feeds us back. The generator tries again and tries to improve its image to get something that's going to beat the discriminator. It's going to beat the person who's the algorithm that's checking what it's doing.

And they bounce back and forth lots and lots of times until the generator, the thing that's creating the fake data, creating the images in this case, is able to beat the discriminator until it's able to. fool it into thinking that it's real. And that's how we get our output at the end. So it's these two bits working together against each other, if you like, to try and convince the discriminator, the thing that's trying to identify if it's real or fake.

When it can trick that, then that's our image. That's where we get things like deep fakes or chatGBT or Dali AI, or Dali AI creating images or something like the website, this person does not exist, which creates images of faces that don't exist. and say someone manages to get enough footage or audio of you, how could this technology be used in the context of scams? Yes, so we're seeing, we have been seeing for quite a while now,

deepfakes being used to create disinformation and misinformation, so we're making people say things they wouldn't say, we're making people do things they wouldn't normally do. And that in itself is quite powerful. That's quite a powerful scam that we're putting stuff out into the world that didn't actually happen. But there's also this idea of we've seen a rise of AI scams, voice scams, where someone will receive a phone call or even a video call, we're starting to see now. And it's a loved one saying that they're in trouble, that they've been in a car accident or that they've been kidnapped and they need money urgently to pay the ransom or to bail them out of jail or to do something.

And these are just pulled from data that's available on the internet, I would guess. They don't have to be 100% accurate because if you imagine as a parent or a partner that you get a phone call out of the blue and your loved one is on the other end saying, I'm in real trouble, I need some help. It's quite a desperate situation. Then you'll panic and you'll start to overlook those inconsistencies and it becomes that much more immersive and believable. Ambition comes in all shapes and sizes. At First Citizens Bank, we roll with your goals because we're built for what you're building. fit for your ambition for Citizens Bank. Study and play. Come together on a Windows 11 PC. And for a limited time, college students get the best of both worlds.

Get the Unreal College deal, everything you need to study and play with select Windows 11 PCs. Eligible students get a year of Microsoft 365 premium and a year of Xbox GamePass Ultimate with a custom color Xbox wireless controller. Learn more at Windows.com slash student offer. While supplies last,

June 30th terms at AKA.m.m.m. slash college PC. This podcast is sponsored by name, audio and focal. With over 100 years of combined expertise, name and focal have been bringing music to listeners just as the artist intended. Since day one, this mantra has shaped every innovation in high-fi design, technology and acoustic engineering, balancing craftsmanship and tradition with pioneering thinking. Name audio. pushes cutting-edge technology to ensure digital precision whilst sustaining Pratt, pace, rhythm and timing, the elusive quality that makes music feel alive and gives it emotional texture.

Today, in partnership with French acoustic specialist focal, name audio creates systems that deliver exceptional sound, and unforgettable listening experiences at home. Try it for yourself at a focal powered by name boutique. Visit focal powered by name.com. for more information. And you mentioned a little bit there about the things that might seem off about it or the things that maybe your child wouldn't say or that your loved one might not do. What can people do to prevent these kind of issues or deal with this when they're in one of these calls?

Is it a case you just don't panic and look for inaccuracies? Yeah, it's really easy to say sat here and thinking about it with kind of the cold light of day. but it is about being that bit critical. So, yeah, does it sound like them? Is it something they'd say? There are pieces of software that you can use to detect if they're fake, but the chances of having one of those to hand are pretty slim. So it is about thinking, yeah, thinking objectively,

thinking critically. So in the case of deep fakes that we might see of, for example, the Zelensky's example of him telling the Ukrainians to lay down arms, that doesn't seem particularly realistic, and it certainly doesn't seem like something that just be put out on YouTube in the middle of the day. Similarly, if you receive a call from a loved one that you're not expecting, you can call them back. You can text them to check where they are. You can, yeah, it's about thinking, yeah, it's really difficult to say. Actually, you take that step back and you think

would this really happen and you do some kind of checking, I guess. And obviously, as we've mentioned, these kind of scams are using your voice or images of you to recreate realistic versions. I've got loads of audio of my voice out there. Yeah. And lots of video of me out there. Is there anything people can do to reduce their risk of being targeted? Or is it a little bit too late for that? It's tough.

The genie's kind of out of the bottle in that we do live our lives quite publicly now. Particularly COVID as well, saw this sense of online community growing as we were kind of physically separated from each other. There was a shift towards living our lives online to a degree and maintaining those friendships and those relationships and that public persona on the internet. And so, yeah, there's tons of photos, there's videos, there's audio out there of me, you have lots of people. To try and get that back would be really difficult.

You could completely erase yourself online and remove your digital footprint, but that's going to be, that's a big ask. We do have the legislation with the right to be forgotten and all of the stuff. You can ask Google to remove you from search results, but yeah, it's about, I don't think there's much we could do to prevent it from happening unless you go and remove your entire digital footprint. It's about people being objective when they do come up against this kind of stuff. So hopefully take the example of my students, if they were to go through and take one of my recorded lectures and use that to create a deep fake. Hopefully my loved ones would know that I wouldn't say some of the things that might be put into one through a deep fake, but who knows?

And we're talking about this in this sense of it being quite widely used. Is this a technology that anyone could create, or does it take, I guess, someone with a higher technical ability and more money that could be doing realistic deepfakes? Sadly, no. Sadly, you could do the pieces of software that you could go and you could download now,

and it'd be pretty convincing. You need its seconds rather than minutes or hours of data to create an audio fake, deepfake. and anyone with a bit of time and YouTube could figure out how to do it and create one of these that would look reasonably convincing. Yeah, so it's one of the benefits and the curses of the kind of AI boom that we're seeing at the moment that there's this amazing technology that we would seem, I don't know, 10 years ago, it would seem science fiction to have a piece of software on the internet that can have a realistic

conversation with you or can create an image from just a small text prompt, something like Mid Journey or Darlie AI. And that's great. But then there's always kind of the flip side to how this technology is deployed, I guess. And as we're talking about this, we're mentioning it. I guess the focus is very much voice and its video. Those are the two things that the technology is advanced the most on. But as you mentioned, their AI is developing rapidly.

Is there, I guess, this chance of deepfix moving into images being sent to you, video calls, you know, more advanced versions of this? Absolutely. So there have been recorded instances of video calls where it looks like somebody else. So you can have this kind of FaceTime type call. Sorry, I got stuck on, I thought, was Skype for a minute then, which is quite often. But we've seen this with images as well. So earlier this year, there was some images put out about the Pentagon being under attack

that were posted on Twitter. They came from users that were verified with the blue check and showing photographs. of smoke and debris around the Pentagon. And then these pictures were picked up and amplified by new sites with the gold tick, with the new paid-for verified tick. And yeah, so these were turned out to be air-generated images, but they got picked up and then it kind of snowballed into this thing that got picked up by, I think, a Russian news station in the end.

I think it was RT in Russia, detailing about an explosion near the Pentagon and then at the White House. So we've already at the stage where we're seeing images like this. And I'm sure some of your listeners have tried things like Dali AI where you can give it a prompt and it will do you a picture in any number of styles of pretty much anything you can think of. And I mean with all of these things, AI, deep fix everything, it's like you were saying earlier, you can't just go about it and say,

you know, oh, it's easy in the moment. It's difficult when you're actually in that situation. but is it a case of just stepping back and I guess trying to address things with as much logic as you can? Yeah, pretty much. So deep fake videos are an interesting example of the technology that you can use to spot them. So when they first came out, they were quite easy to spot. There were things like blinking was something that was really hard to get right because the way that the deep fake either wouldn't blink or would blinking in a weird way.

So as that was used as a detection method, then the people writing the deep fake, deep fake algorithms would then, they improved that. And so it kind of became this arm race, arms race of the algorithm developers and the people writing detection software had this, yeah, a race to see who could do what. And so I think we are kind of beyond the point of technology now, and it is about that objectivity and that humans doing what they're good at. So the computers are great at creating this automated content and working with vast amounts of data and making it into something tantal.

and coherent. And then the role for humans alongside AI is to think critically, is to think objectively, to think, would this really happen? And so, yeah, I think that's one of the big areas where we can take that control is thinking objectively. So we've been talking mostly on the topic of deepfakes. I guess that's probably the most prolific version of what the scam has become. Yeah. Are there any other forms? of, I guess you'd call it advanced technology that could be used to create scams and that are being used in that context? Yeah, there's a few. There's a really interesting idea called facial morphing. So that's a task of creating an image. It's where you take, say, an image of two

individuals. And you morphing can fully transform one face into another face. But if you stop the process midway through, then you get this idea of a morph, which is a non-existent person that resembles both of the people you were trying to create in the first place. So there's stuff online where you can go and create a morph of what it would look like if you cross two of your favourite actors, for example. But this is also something that's being used to create ID documents that look realistic. So you could register a face on a passport that would, you could pass two people for that face. So because you've morphed the two faces together, then it's taking the facial landmarks or the eyes, the nose, a note.

the mouth and where they are and the key geometry of the face and blending them together. So there have been cases where a group of activists merged together a photo of a high-ranking EU politician and a member of their group. And they created that and obtained a German passport, which could have passed for both of the people. So that's a really, maybe calls the wrong web, but it's a really cool use of the technology that you can create these images that can fool the systems that are checking these ID documents. And that's, I mean, that's obviously quite advanced and it's in the sense of,

it's not being used so much to scan people, but it is kind of a sign of where the technology is going for that risk of the average person being caught out on things. Yeah. Yeah, it's, one of the things that they found, what the research is found with looking at facial morphing is that humans aren't very good at spotting them. They're not great at identifying them. And the results that you see on those experiments are kind of 50-50.

So it's as good as random guessing as to whether it is one person or the other. So the answer to that is technologies to try and identify them. But yeah, I think the technology is really interesting at the moment. It's providing so many opportunities and developing so many things that would have seemed, again, like science fiction, only five, ten years ago. and it's having it's changing the way that we think and the way that we we have to work and i'm curious um what you see as being i guess the the most common sort of scams these days what's the ones that are becoming prolific and the ones that everyone seems to be

pulling out the bag um they're variations on a theme they're all variations on classic scams if you like so it's i'm sure again i've had many of them you might have and your listeners might as well, that you have the phone call from someone saying your machine, your computer at home has been compromised and you need to log in and give them rights to this to try and fix it. And they say they're from Microsoft, or they say they're from some other name that you would trust.

And it's the scams are all just playing on our own innate fears. They're adding time pressure. They're adding stress. They're making it seem, they're aiming to short circuit that rational part of the brain. So we're going to panic mode. into fight or flight and we have to make a quick decision. And the scams that we're seeing now, so whether it's a phone call that comes from a fake voice using AI saying, your partner's

been in a car accident, or whether it's a video of a famous politician saying something they didn't say or a celebrity in a compromising position, they're all about kind of tweaking that bit in our brain that makes us panic and makes us think there could be something really wrong. So it's less about the technology, really. It's about the psychology. of what it's manipulating. Yeah, as a computer scientist, I genuinely think, and the work that I do at the moment, that human bit is the bit that's going to be really interesting.

The technology is evolving so rapidly now. And if we look at the start of the year when chat GPT launched to where we are now, there's several versions of chat of GPT, the answers it gives you are so much better, it's so much more convincing. And so it's going to be our role. In that, it's going to be how humans interact with that kind of technology and how we let the computers do what they're very good at, which is processing lots and lots of data very quickly. And then humans doing the bit that they're really good at, which is that logical, objective thinking and that creativity, I guess.

So with the rate that technology moves, I mean, we're talking about right now that in the last year AI has just gone leaps and bounds above what anyone would have thought. Yeah, there's likely to be, I guess, a whole host of new scams, not too distant future. What do you think is like the future of that? Do you think it's going down this sort of root of face morphing or is it just that no one's really going to be able to see where this goes? I think that's a really tough one to predict. I think that we've seen a huge rise in, well, disinformation and misinformation have been a real hot topic in the last, I don't know, where are we now, seven, eight years. And it's something that I think is going to accelerate with this, where you can't necessarily take the fact that something exists as a video or as an

audio recording to be, it definitely happened. There are technical methods to look at that and to see if they've been created synthetically. But yeah, I think that, I just really see that area of not being able to just believe your eyes with the first thing that you see. and having to think a bit more widely about it. I think that kind of misinformation, disinformation, blurring reality and fiction is going to be the big issue. And for the most part of, I guess, the scams we're talking about here, it's a case of preying on someone's fear

and the more psychological effect on that side. But, you know, in the past couple of years, scams across all areas have kind of just really attached themselves to technology. There's the famous, you know, Tinder swindler story. There's these fake NFTs, cryptocurrencies, you know, as all these things have made headlines for themselves. But that's more about giving you an opportunity for success.

Do you think that's, when you get into that world, is it just a case of if it's too good to be true? Yeah. Again, it's, we are all very well aware of email scams now. we're all aware of getting that email and saying that someone has tens of millions of dollars in their bank account and they need $500 from you to release that and you'll get a million dollars. And again, this is just a variation on that.

It's promising you something big for a very small investment. It's an offer that's too good to be true. And it's the same thing in the Emperor's New Clothes. It's just a fancier version. It's something that's more convincing, that's more immersive. But at its heart, scams are still the same kind of thing. They still play on the same innate human desires, whether it would be to find a partner in the case of the Tinder one, or whether it's to make some money, which is quite appealing to everyone at the moment.

Yeah, I think they're still appealing to the same root causes. It's just the delivery method is slightly fancier now. So we've been speaking mostly about the, I guess, the individual's ability to reduce, their risk when it comes to fraud. But is there any sort of improvement from, I guess, a software perspective as AI rapidly improves, could that be used in fraud prevention

in a different kind of way to what we see now? Yeah, there are technologies that you can deploy to see if something has been deepfaked. You can look to see if an image has been or audio has been created synthetically. But there's also, it's a tough one. the technology is outpacing the way that we can think about it and the way that we legislate for it.

So I think we're kind of waiting for everything to catch up. So there's a huge push at the moment to talk about how we manage AI and how it's leveraged in the future. And I think that's going to be a really interesting conversation that evolves over the next couple of years. So it's basically once it passes a test of this is believable enough. Yes, absolutely. And it could be millions of iterations between. The first random data that sends in and the final image, just bouncing back and forth rapidly to try and get that. And the success of that is based on the amount of training data that's available.

So the amount of data that you can feed into these algorithms to say, this is what a face looks like, or this is what a picture of a horse looks like. And the more data you give it, the better it's going to be at fooling those algorithms. Thank you for listening to this episode of Instant Genius. That was Ollie Buckley, talking about the future of scams. The Instant Genius podcast is brought to you by the team behind BBC Science Focus magazine, which you can find on sale now in supermarkets and newsagents, as well as on your preferred app store. Alternatively, you can come and find us online at sciencefocus.com.

This podcast is sponsored by Name, Audio and Focal. The texture and emotional depth of music can be lost through digital sources or poor signal. Name Audio believes you can have digital precision without. analog warmth. Alongside French acoustic specialist vocal, Name creates high-end audio systems combining innovation with craftsmanship so you can listen to music, just as the artist intended. Discover more at name audio.com.