This podcast is proudly produced by Stripe. The autonomous Fink Crime Department. Welcome to the laundry, the podcast connecting AML compliance and financial crime to the real world. Sometimes you just have to pay your respects to the OGs. And everyone in this industry knows Graham Barrow and Ray Blake, who have been unpacking the finer details of financial crime prevention on their podcast, The Dark Money File since 2019. So it was an honor when I became the first person ever to take up a third chair alongside them to discuss all things AI in financial crime. We brought you half of that conversation last year, which you can find in our podcast feed. And now we're bringing you the second half too. I'm Madid CEO, it's Stripe, and in this special takeover, we are handing our podcast feed over to The Dark Money Files. Listen as Ray, Graham, and I discuss a $20 million Hong Kong deep fake heist. Why just ask Cheshire Pt to do it is the worst advice in compliance and would AI ever have come up with the Beatles? And what does that mean for the future of compliance jobs? Okay production team, let the takeover begin. Hello and welcome to a special edition of The Dark Money Files in which we shine a light into a murky world. Today it's called The Dark Laundry Files and this is part two of a crossover podcast that we've done with The Laundry. So normally I'd just be introducing my friend and business partner, Graham Barrow, Salogro. Hello Ray. Today I also have to introduce Marit from Stripe. Hello CEO of Stripe, absolute industry superstar. Oh my God, thank you so much. And it has to be said the very very very first person ever other than Ray myself to say anything on The Dark Money Files podcast. Oh, I'm so honored. I'm so honored. I'm honored. Yeah, well, I mean it's lovely to be here with you, Marit. And thank you for flying in specially. I mean, I'm just honored to be here and also just got to say, I mean I came into the industry a bit later than you guys, but some of the content that you guys were posting was part of what made me find the passion for the industry as well. So yes, human aspirations and leading stars. Thank you. Wonderful. Well, look, one of the things that we wanted to do was to get proper value out of meeting with you. And both of us have a bit of a Achilles heel in compliance because the flavor of the month or the flavor of the decade probably is artificial intelligence. And it's a terrible thing to confess, but I mean, I don't know very much about it much less than I feel I should. And I suspect many of our listeners will feel the same. Is that fair, Graeme? I start with ordering the evidence, Ray. Well, indeed. Sometimes any intelligence would do to be honest. So I thought what we could do was use this special episode to present if you like everything you were afraid to ask about artificial intelligence in the past. So Marit, tell us about how you got involved in artificial intelligence in the first place. Well, so I did study robotics and. Well, that's fairly direct. Robotics and software development. They're including AI. So that was. So I came from a technology background. And then I was part of working on this AI project at the university in Norway. And it got spun out as its own company. And we did various AI projects. With various type of clients, one of them being a bank who then said, "Okay, we did a project. It wasn't that successful." And they were like, "You know what? What you did there, not so important." But we have this thing called KYC. That's pretty hard. You guys, maybe we should do a project together on it. It's quite complicated. And at the time, we used a lot of what's called graph technology or graph databases to kind of piece together a lot of data points. And it turns out that was really good for being able to piece together ownership. And that is, of course, very handy when you're doing KYC. So that was my entry into this. So you're just mentioning graph databases. Now, one of the things about AI, Mora, is it is replete with terms that not everyone might understand. So we've got things like machine learning. We've got supervised machine learning, large language models. We've got generative. AI, I've got things like agentics and help us out. Just take us through some of those terms. And if you can, explain them in kind of relatively understandable. We'll try to make it super easy. So let's just start with machine learning. So that is basically when a computer learns from examples instead of being explicitly program. And kind of bringing it back to our field. Let's say instead of programming, you have a company and you know all these attributes that make it shady, like a shady shark, you could be if it has this, then this, if this, then this instead of programming all of that explicitly, what you can do is then take a set of companies that you know to be shady, give it to a machine and see if it can then learn the patterns so that when the next new company comes in, you can classify it as shady or not shady. That is kind of the super, like a super basic, basic explanation of it. So you give it examples that it learns from like supervised machine learning. You give it examples, it learns from hopefully it then learns enough that when you present it with something new, a new data point, it can then classify it into one or the other. So large language models. And that is something, that's something different. And that is also, it's like, it's trained on vast amounts of text. So unstructured data language, you know, it's unstructured data text and they can understand and then generate human language or text. So for compliance teams, this means that a large language model can then take a lot of different data points and then summarize it, for instance, or it can, if you give it a lot of data points, it can help you create some new text as well. So it's something a bit different. And the architecture that's being used a lot is this GPT. GPT, and I kind of intriguingly, I'd sorry, and I had to start advertising on UK television and the app was about asking for something called for a first date. It was quite effective, but the other one I'm really interested in because it's quite new to me is a gen tick. What's that about? So that is kind of taking this LLM's or generative AI one step further. So instead of having generative AI, which kind of means that you can generate something and now with these large language models, you generate text a lot or you can generate images and you can generate a video, but instead of just generating something new, you take it one step further and you can have them computers also taking actions and making decisions upon what it has learned. So for instance, in the financial crime world, you can think of it as suddenly you can use an agente AI to decide whether or not there's also false positive or not, for instance. So it's kind of like the next step beyond that. And of course, it is the new shiny thing of the month in a sense. Everyone is throwing this around. Everyone says it holds a lot of promise, but it's still in its early days. OK, now that's a really interesting point, because I can't be the only one who has started reading some material that's quite critical of some of the public models and what they're actually doing with them. So some people suggest that LLMs are just sort of super computer implementation of Eliza, you know, the sort of 1960s program that they used to emulate an analyst and try and pass the Turing test. I remember being a computer hobbyist in the 80s and having a version of Eliza in basic that did a pretty convincing job, I thought, but I was 14 at the time. Do you think that's a fair criticism? We're just chucking a huge amount of computer power at something.
that apes rather than displays intelligence? I think Eliza, so first of all it's crazy to think now that oh they had a chat part in the 1960s, just how long the ideas and how AI has kind of been in the making, but Eliza, it had this you could ask it, you could say to it like I am sad. Yes. And then Eliza would answer why are you sad? Yes. I kind of had a few hard coded these ones, but it just kind of mimicked and reflected what you said. And of course it fell through when you asked it something outside of what it was programmed for. Who won the FA Cup in 1952? Yeah, then no idea. So I do think then the depth is really the difference here because now imagine all the different questions you can answer. You can ask chat GPT or any of the other LLMs and solutions out there, whether it's Google's Gemini OpenAI's chat GPT or you can ask Claude. So it's the depth I would say that's quite different. And I'll be coming in today. I couldn't help but notice that an Australian branch of one of the big four consultancies is just being called by 700,000 pounds. So doing a proposal for the Australian government that actually contained made up references to historical events. Yeah, they had to reimburse the customer because they had to made up references and a lot of a few scientific articles I saw it too. So chat GPT and others isn't self-controlling. It will happily make stuff up. It does. The general itself is nature. And you will not have any notion unless you check it. I mean, there have been issues in the America where legal cases have been cited which are completely fictitious. So it is a dangerous territory. Or do you think it has danger? I think it's definitely worth being mindful and cautious. So of course you can if you're using some of these models, you can put on guardrails, etc. But it is generative in its nature. So one need to be aware of the limitations. And of course, I mean in financial crime, the obvious fault pits is like, "Oh, suddenly it makes up a pep or a sanctioned individual." You know, so it does have some pitfalls. But yeah. You can also say, "But where does it generate it from?" Because if it's making stuff up, this is not sitting in sorry. So it is trained on billions and billions of data points, right? So it's kind of trained on the internet and tax that it input that is. But these are entirely manufactured cases. This is not stuff that's read anywhere. This just actually made it up. So I just think. Based on everything that it knows. So it's kind of like. Yeah. So it's job really as I see it is to be plausible rather than to be truthful. Which makes me think of criminal applications because that's also a valid objective of your average criminal, isn't it? So how are the bad guys using this? I mean, you guys have probably seen some of the same stories as I have. Of course, now you can also fabricate images and these things. So fake ideas, definitely. I mean, the scam texts are just so much better now than it used to be. It can be hard to actually spot one because I almost fell into this pitfall not too long ago. And I thought it was a credible email, but it wasn't. It was just total bullshit. You can have AI now that can call. You don't need necessarily call centers with people. You can have call centers with AI. That is one thing. And yeah, so those are the main ones that we saw. I guess you both saw also the incident. Was it in Hong Kong when they wired? Was it $20 million? Or yeah? Because they thought they were on a call with the CFO, but it just wasn't. Exactly. So all these things, these things as well. I mean, it's crazy. Can you imagine? You jump on a call. You think it's the CFO. It turns out it's AI and you lost $20 million. I can't imagine. Well, I've got $20 million to lose. I don't know. Have you guys seen any interesting or nuts or scary interesting, like the ways criminals have used it? What have you guys come across? Well, I think we haven't probably. So because I mean, here in the UK, you don't, in our world, you don't need to do it because you can do frankly laughable stuff and still get away with it as you know from our link to info. Yes. But it won't take long. So I guess the, you know, the follow question is, how's AI going to help us in that arms race then? Yeah. I think also just to final point on the, on the criminal side is just that now it's just, it's just so much easier to do a lot of things. Maybe before you needed to have a lot of technical expertise in order to let's say set up a fraudulent website, you needed to like, to have a more knowledge about how certain things work. But now that knowledge, like the threshold is so much lower, you can easily set up a very nice website in a few seconds using some of these, some of these tools. If you have any questions about how to do certain criminal activities, you can just ask some of these, these models. So I think in general, it just more access to knowledge around how to do fraudulent activity as well. Did you hear that? It doesn't add to us playing email tennis with the clients, chasing documents, waiting on data, sending another follow up, and another, and another. Strides, ultimately routine cases that should never have needed this human in the first place. The ones that are low risk, predictable, and take up time that belongs elsewhere. Analyzed, stay in the loop for the decisions that actually need them. And every decision is log traceable and ready for audit. Welcome to the autonomous fin crime department. The one with humans in the loop. Look at them, let's try today, aye. So how do we find park? Yeah, that's the good question. I don't necessarily have all the answers. And I feel when I talk to a lot of people in the industry, everyone says, we need to use AI to fight back. But the recipe on how to is still not made. I mean, the one thing that is one of the things that both, that you hear a lot about, that I think the industry is kind of aligned on, maybe we should do that, is of course use AI to empower the first and the second line of defense. So maybe you can reduce the time analysts spend on gathering information, analyzing information. Maybe you can have AI help summarize, present, show data. So everyone kind of, you know, their knowledge can increase. Everyone having their own personal assistant, imagine what you can, how much more, how much more every person could do if they, let's say, 30% of their time, 40% of their time. Was taken away from doing manual repetitive work, but doing analysis, thinking, understanding new patterns. That is one of the kind of promises that AI holds in the industry. So it's the classic complaint, isn't it? That the job is ruined by the paperwork. So if you can take that away from the job, you're doing much higher value stuff. Okay. And also that's also one of the, you know, when there's a new big technological shift coming now, such as AI. And then you can see the like overall productivity in every industry kind of increases. It doesn't necessarily actually play out in the sense that it will disrupt something or completely change something. It could be that these technologies like AI just goes into every industry everywhere and just increases the productivity by a certain percentage. And we could see that that is actually what's happening in compliance to instead of just imagining that, oh no, all the jobs will be gone. It will just be AI agents running around dispassitioning false positives, filing SARS, etc. So I think the future is still yet to be, but I think somewhere in between, in between there. I saw this example with, I didn't, I saw this example in the passing. I think it was some American fast food chain that had received 18,000 orders of water. Okay. And of course a human who had received that could immediately understand like something that was ordering 18,000, but an AI, the AI system that didn't. So it could be that, you know, all the human work in this field is just elevated a bit. You sit instead of gathering information yourself, you're more, okay, does this make sense or not? Okay. Okay. I don't know what you guys are sitting listening, which is fascinating conversation is it absolutely takes me back to not that I was there, but to the industrial revolution where this is not.
all these machines are going to take all of the jobs, you know, and actually what happened was by a large that people got different jobs, new jobs. We probably, it was definitely the start of the service economy because people then were freed to do more things, you know, they were able to have some leisure time because of the savings caused by machines. So I did notice Sam Altman the other day saying that he thought that in the decades to come AI would be regarded as a basic human right, which I thought was quite an interesting proposition that actually you couldn't survive as a human being without access to AI, which is quite a Wi-Fi. Yes, slightly scary, but you can't stop here. There's genius out the bottle. So I guess our challenge is to try and make sure that we make it work for the benefit of the many and not the few at the moment. And also you did mention the Industrial Revolution there again, the general productivity and the economy, that was also a big boost from it, so it could be that we see. But I think AI is some of the things that I have noticed and that I've spoken to people about not just in the compliance industry, but more sort of every industry. And that is that, you know, it kind of brings back the ability to do more things yourself. So for instance, if you're a senior in an industry, maybe before you had a junior that helped you, let's say, write a presentation, doing these writing a presentation, doing, let's say if you had a public speech or you needed some ideas or whatever, you can kind of bring back a little bit more of that yourself because you have the tools available with AI. It's a good creative sparring partner. Like if you're stuck on something, hey, give me some ideas. It might not do the work for you, but it kind of gets you started. So I think as well, you have discussed with other people is that there's fewer, maybe entry level junior jobs because that is the seniors do it more themselves. So that is something that I have noticed a bit. Do you have to be a senior then? Yeah, exactly. So that is a big question. That is a challenge that I do think we as a society, not necessarily in compliance, we'll need to face. One of the things I was particularly keen to ask you about is that as the CEO of a company that promotes AI to a wide range of people, I'd love to hear what you think people get wrong about AI. What do people believe about AI, which is just misguided? And I'm talking here about CEOs, emolaroes and compliance heads, team leaders, IT people even. What do people believe that just isn't true or is misguided? Oh, it's such a good question. But sometimes I do feel, and maybe this is more the IT people or hardcore AI users. Sometimes it's just, oh, just ask your GPT to do it. And then I'm like, no, this is maybe down the line. There will be tasks we can use AI for. But this is first and foremost just that we as human beings need to sit down 30 minutes and get on the same page about what is the actual problem that we need to solve. We don't necessarily just need to use AI to create a solution right away. We're not aligned on the underlying problem. If you know what I mean, that is one thing that I've noticed a lot. A lot of people are just, oh, just get GPT to do it. I'm like, do what? We don't know what we want to do yet. We don't know what the problem is. That's fascinating because now that you mention that I realize that 90% of the value that you get is out of understanding the problem, not just fixing it. Yes. And that is one thing that I have noticed a lot. And I think actually it has matured me a bit too, because previously I could be, I could very well be like, oh, let's just jump into solution mode. Of course, I still love general coming up with ideas for all that. But I think I've noticed it a lot now. So I'm been becoming a bit better at like, okay, let's take a step back. What is the problem to be solved? What's the kind of, what's the target we want to achieve here? We don't necessarily, the AI is not the answer to everything because we don't know what we're doing. And there's a kind of general rule at work there. I can run, I've both come from a training background. And I can't tell you how many times that we've been involved with firms, we've said, oh, we need to do some training. Why? What is it you think the training is going to do for you? Because what is the underlying problem? And is it actually one that can be solved by training? Actually, quite often, and I think Ray, you were just a little bit to this going through that thought process of why do we think we need it? And what's the problem we're trying to solve? It's almost as beneficial as finding the solution because very often the solution is, is a very different one from the one you start out thinking you need. And I guess one thing that AI might well do is force me, which I think much more seriously about what ultimately is our problem and is AI the answer. And I think a lot of the time actually the answer will be yes, but not always. Yeah. And also, maybe you can get to the solution faster using AI, but you still need to just like align on the problem. Another thing that I have noticed a lot is, and this is not necessarily in compliance, but also just the fact, you know, we run a podcast, we do a lot of marketing. And you know, when these tools become available, oh, let's create, we can create video with AI, we can create photos with AI. And here is a good quote from like Jurassic Park, we use it in an episode once is like just because you can, doesn't mean you should because now I feel creativity and original ideas are becoming even more valuable because it's so easy now for everyone to create a video about anything, but it's very quickly that you become in sense like desensitivized to like AI generated content because the human machine is still very good at detecting if it's not made, if it's just like generic, you know. So having original ideas and putting even more effort into thinking, how will this stand out? How will this resonate with people? Is this something that's like truly felt, like comes from someone intended at an audience with like true care? You know, because it's so easy to just create something generic. So I feel like the emotional aspect of emotional aspects become even more important. Maybe that's just me, but that's something that I've felt a lot myself. Is my kind of final thoughtness is would AI have ever come up with the Beatles? Yeah. Yeah. You know, because there's never been anything quite like them. It's quite hard to explain to people you won't around at the time and you wouldn't want to because it would make you as old as me. But there was nothing like them before. Where did that come from? You know, and the fact that they changed so much and they would different elements. And yeah, I'm going to hang on to that as a really, you know, what there's always been for people because I don't think the Beatles would have ever arisen any other way than from human beings. And I also do really, I agree. And also I do really believe what you said in the beginning is that AI will come in. It can change things, but then it will be new jobs, different kinds of jobs. Okay, maybe we are able to alleviate some pain points from the first line of defense to sift through a lot of false positives, manual work, but then the nature of the work they do will be different. So maybe they are more overseers of AI. They kind of come to work in the morning. There's all these things that you need to kind of look at. Yes or no, you kind of, but instead of going into the manual data point gathering yourself. So it will change. So I do believe it's very important for organizations, CEOs, like leaders to be very to foster a culture of curiosity, innovation, testing, you can run some experiments, don't be afraid to fail. Have all these human aspects of testing something new and bring that in because I mean in the beginning, the AI, maybe it will fail, it will do a lot of things wrong, but at least let your people try it and foster creativity around it and see the solution. I really don't like it when people come in and be like, oh, it's going to take all of our jobs. Everything will change and yeah, now we can do some really cost cuts and AI will do everything. I'm like, no, let's take a human approach. Let's be curious, interested. And yeah, let's see how we kind of go. And what will be the thing that makes AI no cold fashion tonight of day? Oh, good question. I don't know. What do you think? I genuinely, I don't know, but it will happen at some point because it always does. And I don't know what it will be, but it will be spectac. You will know this quite better than me, but it's about becoming indistinguishable from magic. Oh, yeah. Any technology which is sufficiently advanced is indistinguishable from magic, not the sea clock, I think. Yeah. So I wonder in 50 years time, people look back in the AI. Look what happened after that. Yeah. I don't know. Maybe like quantum computing has a major breakthrough and it becomes like super commercially available and like, cryptation just becomes, you know, obligated and there's like a whole new paradigm that we need to face. I don't know. So if you can merge AI and quantum computing and 3D printing, maybe we can have a Star Trek replicator. Maybe. That's what I would particularly. A new universe. Yeah. Yeah. Absolutely.
- All right, well I think on that note, that's a bum shell. Yeah, that's a good speculative note to end on perhaps. Thank you so much, Mara, for enjoying. - I did enjoy it. - Yeah, for indulging. - That's not wrong. - For indulging two people who just feel they to know more about these modern things. - Yeah. - And giving us the benefit of your expertise. We're very grateful, I'm sure. - All in all, that you invited me. - Thank you. - Thank you so much. - I'm invited myself as well, but you know. - That's okay, that's good, that's good, that's what we rely on. Okay, well hope you enjoyed this experiment in having a third voice, let us know, and we'd be happy to do more of it if that's what you want, listeners. - Yeah, and if you feel that you could be a third voice that we would enjoy talking to. - Right to Graham. - Yeah, snail mail please. (laughing) (upbeat music) - That brings this takeover of the laundry to an end. If you've enjoyed this episode, check out the back catalog of more than 160 episodes and follow the laundry on your podcast platform of choice. Please share the podcast on your social media channels, telecolleague to subscribe, maybe give us a five star rating. If you think we deserve it, it really helps people find us. We got your own podcast and want to take part in a collaboration, message me a LinkedIn or email
[email protected]. The laundry is proudly produced by Stries. We are building the autonomous thin crime department where humans and agents collaborate on the compliance work. Visit strries.ai for more information. See you next time. This podcast is proudly produced by Stries. The autonomous thin crime department. (upbeat music)
PodcastTranscript AI