Digital identification revolutionized by AI – a talk with Husayn Kassai

Digital identification revolutionized by AI – a talk with Husayn Kassai

As we are moving our lives towards the digital space – starting with online banking, shopping, health services to social interactions and more – digital identity will likely become the door opener into this online sphere.

In the real world, we have institutional practices in place for proving and verifying who we are – such as personal identity cards, passports, health insurance cards, driver’s licenses etc. But in cyberspace, identity is even harder to verify due to a lack of trusted standards and face-to-face interactions, not to mention the proliferation of identity fraud.

Our conversation with Husayn Kassai, CEO of Onfido, the market-leading digital identity verification provider, revolves around how proof of identity can be digitally exchanged between users and service providers in a secure and efficient manner. Identity authentication is the very first step towards establishing trust between two parties in the online space and ultimately “underpins the whole digital economy,” as Husayn explains.

The realization that identity verification methods call for novel approaches came to him from multiple sources. His family’s story regarding the hassles around opening a bank account without being registered with the Credit Bureau made him aware of the failures of the existing identity infrastructure.

The “smartphone generation” and the demand for user-friendly, digitally accessible services was a key driver for them. The sophistication of cybercrime and identity theft gave them another strong impetus to rethink digital identity security. All this happened in a booming tech ecosystem – with all its powerful newcomers like mobile phones and cameras, connectivity, advancements in biometrics, machine learning and artificial intelligence. This gave them the means to design their best-of-breed digital identity service.

Onfido’s pioneering approach is geared towards offering an AI-based software development kit that can be embedded into the provider’s browsers and applications. When a new user is registering on the platform with a copy of their government ID and a short video selfie, the software powered by machine learning starts comparing them with a myriad of patterns to conclude if the identity is genuine or fake.

As he moves onto telling some thrilling customer stories and illustrating the array of use cases, it becomes evident that the future looks promising for this field. Their early adopters were neobanks and internet-only services like Revolut, but mainstream banks are also slowly catching up with the digital trends. In the category “digital access”, they see endless opportunities in the online and offline world alike. From airport check-ins and demonstrating a COVID-19-test to voting online, there is no limit for using digital identification.

As we start exploring the phenomena “artificial intelligence”, “machine learning” and “automation”, Husayn puts these into the context of industrial revolutions. Ever since the First Industrial Revolution, tech advances have started to gradually substitute human labor - from the steam engine and manufacturing machines to modern day self-checkouts in supermarkets. Wit AI, we tap into the field of simulating human intelligence. According to Husayn, the harnessing of narrow AI at Onfido is more concerned with "spotting patterns or repeatability" in a range of areas such as ID classification, extracting characters and detecting fraudulent patterns. The essence of Onfido’s self-improving digital ID verification system is that the more data is available, the better the system becomes, with higher and higher success rates.

But as with all revolutionary developments, AI can also be a double-edged sword, when it gets into the wrong hands and remains unregulated – think surveillance capitalism and unbounded data abuse. So it is Husayn’s vision that the future will be fundamentally shaped by the power of AI. But it is up to all of us – providers, regulators, and users – to counteract its misuse and degeneration and create an ethical foundation for its adoption.

Listen to the episode on Spotify or Apple Podcasts to learn more, and let us know what you think. Scroll down to read the transcript of the recording.

Show transcript Hide transcript

Paul: Hi, everyone! Welcome to the 10th episode of "under CTRL". My name is Paul Bartlett, and on today’s show is Husayn Kassai, who is the CEO and co-founder of Onfido. We will discuss how digital identification helps companies such as Revolut, and why it is a more secure alternative to physical identification. We also talk about how artificial intelligence is a game changer in their sector to prevent frauds and other criminal activities. Hi, Husayn! How are you doing?

Husayn: Hello!

Paul: We've got Husayn from Onfido on the show today. So this is going to be an interesting one. We're going to cover the topic of AI and personal identification, how that's moving from the physical world to the digital world. So I'm really happy to have you here on the show today, Husayn. Welcome! So I want to get started with handing it over to you. Okay? And giving us some insight into your background, and about how Onfido came around.

Husayn: Of course. It's good to be on. So, I am one of the three co-founders and the CEO here at Onfido. We started eight years ago, back in 2012. And the main reason was that we could see that more businesses are moving online. And yet, they didn't seem to have an effective way to verify that the people they're onboarding as customers are who they claim to be. So our approach is that we help 1500 businesses integrate our software development kits into their applications. It could be online browsers or smartphone applications, whereby if a user is registering, onto an online remittance platform, for example, they are asked to take a photo of their government ID, such as a driving license, and a short selfie video. And then behind the scenes, we're taking that photographic ID and we're verifying if it's genuine or fake, using machine learning, which we'll come onto. And then we're comparing the person's face to the photo on their ID, to make sure they're the owner of that identity, so that they can be onboarded. And that's it in essence, it's building that layer of trust for this new digital economy.

Paul: Okay. And so where did it kind of, you know, start off for you then. You said eight years ago, that you started out. And you saw the potential? Or you just happened to be in a place where this discussion was taking place with a potential customer, and it grew from there? Was it really your own...?

Husayn: It's a number of things. So for me personally, when I turned ten, my parents moved from Iran to the UK, and I remember it taking them quite a few months to be able to open a bank account and rent in their own name, just because they weren't registered on the Credit Bureau. So growing up, I recognized the importance of identity verification and authentication. And that basically, being able to prove your identity digitally underpins the whole digital economy. And so we, myself and my two co-founders, when we started to research this space, we saw that the identity infrastructure is broken and getting worse. The three main issues are: On the one hand, any face-to-face interaction, which is what we're still mostly used to, inside a bank branch, when you're registering for the first time for a bank account, for example, that's not convenient. And it's time-consuming. It's not accessible to many people either, if they don't have a local bank branch. As sort of- this is a very global issue, global problem. The second is this fact that half the world's population are not registered in the Credit Bureau. So the Credit Bureau infrastructure is not fit for purpose for them. And the third one is around fraud and security. Given that data breaches are details, such as date of birth, name and address and so on, are online, on the dark web. So if that- if those are the data points that are required to prove you are who you claim to be online, then it's- there's not much security, 'cause fraudsters can cheat it. And our approach of the government ID and a facial biometric is geared towards making it convenient, giving access to anyone with a camera enabled device and making it secure at the same time.

Paul: Good. And you mentioned you've got customers, many customers all around the globe. And I'd imagine that a lot of them are financial institutions. So I would imagine as well that the- how did you get to talk to these customers and build the trust with them? Because certainly the banking industry, back in those days, was very traditional. "Look, if you want to start opening a bank account, if you need to verify yourself, you got to go into the branch. You got to make an appointment." So how did you start building that trust with those banks or those institutions, where you think that this product could be suitable?

Husayn: So we actually started in a different industry. Eight years ago, we started in the online market place, in the trust market places. That's when you want to share a home or car, or have a nanny or cleaner or tutor come to your house. And there's no face-to-face interaction there, so as you can imagine there's this need for verification to happen, and to happen well. Then we started in the fintech wave, where you had online neobanks and payments remittance landing and so on. And it's only in the last four years that we've been able to make progress with the mainstream banks. It's essentially as they have started to truly go digital. And in the UK, we service four of the five mainstream banks. And the trust that came as a result of them looking at the neobanks, and learning how the neobanks innovate, and it's part of that journey that they discovered that under the bonnet, it's companies like us that are powering this move to the digital economy, and by extension, they started to come to us directly.

Paul: And... I mean, was it just a transition with technology? Or when you think about, you know, like the physical identification process moving to digital, where did you see the necessary step in that? Is it because of efficiencies that- making the customer experience or customer-centric journey for these banks a lot better? Or these institutions, where- whichever customer it may be. Where did you see the shift?

Husayn: Well, we had- our starting place was four building blocks that we wanted to focus on. And the first is to help 98% of people access services digitally, should they want to. Second is to stop the 2% of bad actors, fraudsters. And the more effective you become at stopping fraudsters, the more easily you're able to enable the other 98% the access to services. Third is to make this as seamless and as frictionless and as user-friendly as possible, the whole journey. So it's much more akin to twenty-first century essentially. All online, all digital, a few clicks and you're kind of- be able to sign up to your own bank account, for instance. And the fourth still called privacy. So now we're achieving the first three, but without having to store personal information or use personal information. Now, your question around what were the trends and the factors that made it obvious that there's a need for this? There is a supply side and demand side, but even before that, fundamentally, taking technology, fintech, everything out of the equation, it's when as humans, whenever we're offered something that is more straightforward and essentially a simplified version, we're always going to take that over something more cumbersome. So I would want to sign up online if I can, as opposed to go inside a bank branch, for instance. More likely than that. Second is offering greater security is always better than less security, and so on and so forth. So the supply side taking logical building blocks that having now enabled these to be achieved. The key ones are the smartphone. Fundamentally, camera quality on them and so on. Internet connectivity, improvements in biometrics. Service going online, becoming more accessible. And all these are supply side. On the demand side, there are fewer factors. But as consumers, as we've increasingly been exposed to nicer interfaces, like Facebook, or before that, MySpace, and smartphones... being able to do everything on your smartphone, the extension is, "Well, why can't I open a bank account or make payments or even rent a car and book a travel service and so on?" So the demand side is driven by what has been made possible by other service providers.

Paul: Right.

Husayn: And we just became relevant to that eco-system. And you have to remember that a lot of the smartphone generation now, you know, let's say your average twenty-five-year-old, from the age of ten, they've been exposed to smartphones, or some sort of... from their early teens, essentially. So when they turn eighteen and twenty, and they want to open an actual bank account, to them, going inside a bank branch, that is alien to them. Being- having the ability to open accounts and access digital services on your smartphone, that is what is perfectly normal to them. And what happened with the mainstream banks is they started to see a big divergence between traditional customers that prefer face-to-face, preferring branch experiences and so on, and this new wave of smartphone generation users, who have a completely different behavior and therefore have a different demand. And that's when the mainstream banks started to offer digital propositions, and that's where we are most relevant.

Paul: It's interesting you mentioned the demographics there. And then almost I was thinking about the geography part of it as well, because I- well, it was assumed that- one would assume that, you know, Europe would be the ones that would- Europeans and US would be the population that adopts that the most. But I've also been seeing trends down in continents like Africa, which really are leapfrogging ahead. Because, of course, there's so much distance between cities and villages and things like that. But technology is there to be able to do that. Do you see that similar trend as well, amongst your customer base? That there's certain parts of geographical regions which were adopting the technology a lot quicker? And is it in the developing countries over the developed countries? What do you see happen?

Husayn: Yeah... It's- so in Southeast Asia, where we're very active, there are definitely examples of that. There is a factor around the personal journey of digitizing for many people. So I (inconclusive), and I'm guessing you did. And in a few cases, so maybe it was fifteen years ago, when we went into a supermarket for the first time when they had self-checkout tills.

Paul: Yeah.

Husayn: Versus (inconclusive). It was odd for us. For me anyway, you know. How does it work? It's a bit frustrating. You know, it keeps buzzing. The red light keeps going off, and someone has to come and put in their barcode and unlock it for you and so on. But now, most people kind of prefer that. It's more predictable. It's faster. There's usually fewer people queuing and so on. So that journey takes time. Now, what has happened with COVID is that it has accelerated that journey. So we all have relatives and, you know, an aunt that just keeps asking you to skype or zoom with you. Whereas previously, they would have only called you on a speaker. They experience a digital approach, and they really like it. And they want more of it. So when you're talking about emerging countries, often they don't have an alternative approach. They don't have a local bank branch or anything similar. And so they have to resort to the digital options, and that's when adoption is strong, because it's pretty scalable.

Paul: Yeah. I mean... and just going back to that point, as you mentioned, going through and it was quite alien to see a self-checkout service. And to think, of course, we'll come onto the future part in a minute, but I can certainly relate to that when, for example, being in Tesco's and just pushing the trolley up and queuing up and then seeing these self-service tills of going through. And yeah, you see those things just evolve, whereas a generation before us, or the current generation, they just see- they grow up with it and see it as the norm. So we were talking about your customers and we talked about the geographic. I mean, who is now, would you say, the main customer base that you have? I mean, is it- we saw a lot of digital banks coming up in recent years. They're everywhere. You got your Revoluts and your TransferWise and kind of, yeah, disruptors, right? So they're disrupting the bank industry, and now these traditional banks are trying to keep up and catch up with changing their business models. But is it just the banking sector? You mentioned earlier you started out with somewhere else. What do you see where your technology is being deployed and where it's been effective?

Husayn: So our category is digital access. And three quarters of the customer base are financially regulated in one way or another. And... so a quarter are- is very longtail. So other financial ones, is some of the ones that- well, all of the customers or brands that you mentioned. But on the longer tail, it's a wide-ranging... It's the Swiss Federal Railways, Vienna Insurance Group, Orange, Zipcar and so on. It includes digital access in an offline-world setting. So it's CLEAR in the US, you are checking into an airport for domestic flights. So we go from curb to gates, and typically, you're checked about five times. You know, when you drop off your bag, when you go through security, when you go to the Delta Lounge, for instance. And then on your gate when you board. You don't need any form of identification. You just take your CLEAR acc, and you seamlessly sort of walk through. Equally, check-in to hotels is the same and so on. So these are- this is just a bit of a flavor of the different use cases.

Paul: Yeah, it's funny you should mention the aviation business. It's obviously tough times right now, but I was also very involved in the aviation business, and remember having to go through and show your passport and- at different places: check-in, and then security, potentially security or passport control, and then, of course, before you get on the aircraft. So are you saying that all of these stages, all of the different organizations that you interact with at the airport are moving towards digitization, where your technology is being applied?

Husayn: So there are early adopters of it. Specifically, the one that is public so far is in fact the largest one, and it's called CLEAR, in the US. And that's for domestic flights in the US. I believe there are four million customers. That means that any one of those four million individuals, if they want to fly domestically in the US, they can go through the fast path, the fast queue, which is the CLEAR queue.

Paul: Okay. And so what about other use cases that you have out there? I mean this was one interesting use case. What about the more unusual ones that you've found? Is there any...?

Husayn: So one... one of the more recent ones is around connecting your test results. Say you want to go from Australia to New Zealand. And in the morning of your flight you test, you carry out a test, and the test says you don't happen to have the virus today. You are able to attach that test result to an app, and this app are- the apps we don't build, our partners do. And then we're connecting the app to your offline identity. So we take your offline identity, and we anchor that into this digital app, this account, using your government ID and face, so that you can't then give your phone to a sibling or flatmate or someone else.

Paul: Right.

Husayn: And that is a use case whereby in the morning, you go and get the test. You anchor it to the app, you anchor your physical identity to the app. Then you go to the airport, you show that. You're then able to fly. Then, when you land on the other side, you're able to equally prove that you've already been tested that day, and therefore you can go down the fast path and go home, for instance.

Paul: Right.

Husayn: So COVID-19 is a new use case. Just generally, there are- there's more of a consideration around us- on us physically touching places. So if you want to access a building, if you have to put in a PIN into the entrance security gateway, before you enter, well, if everyone going into the building is touching that pad, that has a hygiene and a virus contagion issue. So now there's a question of can I just scan a barcode on my phone and be able to be granted access automatically? And if I do that, how again do you anchor the person's offline identity, physical identity, to that app that carries that barcode? So it's physical access to buildings, is another new one. We are moving to a cashless society, and are part of it, it's that as well. The big one in the US, as you can imagine, we're not so- I think we're five or so weeks away from the elections, and- so there are conversations around voting and how in the future, the hope is that we'll be able to vote online. It's already happening in a few countries. (inconclusive) companies like Agora and others that help you, for countries that permit it, to digitally vote online. That is one that is just more topical now, it's that- if not obviously, live in the US yet. Nor are there any plans for it to be. It's just purely at a discussion stage. But there's some- we're touching on some of the use cases that I'm hopeful will come in the coming years.

Paul: Wow, that's pretty impressive. That's amazing that you got potentially something for COVID and- so you're actively working with an airline, to do this? Or the airport authority?

Husayn: It's not- there are no airport- so the specific (inconclusive) is- Delfin Health is one. I mentioned CLEAR, CLEAR now have CLEAR (inconclusive) Health, which is for you to enter stadiums for sporting events. So a few weeks ago, over the weekend, there was a hockey league in the US, and sort of 3000 or so hockey players went into the stadium, along with friends, family, and journalists. And all of them were checked through CLEAR. There is Sidehide, which is used live today for guests who- Nobu Hotels in Miami, Florida. So these are like more pockets of examples, as opposed to... there are no airlines using it at the moment.

Paul: Right. I'm just curious for myself, because coming from- having a history in the airline business myself, and my heart is bleeding at the moment, seeing what's happening out there. And I'm hoping, of course, that there was going to be some kind of solution, what you're presenting here, about how you can potentially travel. Because, obviously, the fact that we might end up with a vaccine, it still could be some time. And clearly, that presents us a solution there. So one of the things that I wanted to touch on about collecting this is around the security aspect. I mean, obviously, you must be using certain security technologies to make sure that that's keeping that information safe, right? That you keep it locked down, so maybe it's- you use encryption or what other methods that you use to do the personal identifiable information for that person?

Husayn: Sure. So as a- as with most other tech companies, the- once you register or start offering your services to a mainstream bank, then you typically say bank-grade security, because it's (inconclusive) in the industry that mainstream banks in particular are strict, rightfully so, around all the services that they use. And, as you can imagine, the first bank that we signed up a few years ago, it was actually a six-month process, just to go through all the compliance requirements and make sure everything is completely watertight. So that's on the general security, and then your question in part relates to how secure are we at saying, if the person is who they claim to be, i.e. the extent to which their identity and their face that they're using, is it fake or is it an impersonator? Or is it actually genuine? So when it comes to that, we explain to clients that it's a little bit like an antivirus software. That we do not or are not able to, no one is able to commit to catching a hundred per cent of the fraudsters, just like the way if you catch- no antivirus software could say that they catch a hundred per cent of the viruses. But the way that we've developed the technology, a) we can confidently show that we're better than any alternative and certainly much more than the human eye. And secondly, that it's continuously evolving and improving. So as we sit in the middle of all the different businesses, when fraudsters attempt to cheat the system with fake IDs, as the system learns, all the other clients benefit from those attack vectors as well.

Paul: Right.

Husayn: And that is a benefit of- that you sort of evolve, and you're starting to focus more on the most sophisticated fraudsters. And we produce an annual fraud report that shows- has interesting insights. One of those is that, over the weekend, the fraud rates seem to drop. And that's not necessarily because the volumes go down over the weekend or anything else, it's actually the contrary. But what that- the reason for that is because sophisticated fraudsters are doing this as part of their job, it's a career. So they also take weekends off. They also have sick pay, they have holiday pay and everything else. So it just goes to show, as an interesting example, of fraudsters becoming more sophisticated, and therefore stronger machine learning power technology being relevant and needed, to counter it.

Paul: Yeah. And then that's the- I wanted to come to that point, around the security aspect as well. Because of what you're doing it's like the- obviously, the ability to be vulnerable to attacks and having that level of security in place. I mean, of course, here, what we do is we do full end-to-end encryption. And we also have a zero-knowledge policy. But that doesn't stop, still, those people out there wanting to test and try and test to see if they can gain access to the system. So is that something you see as the biggest challenge for you? Or are there other big challenges out there, trying to deploy this technology and keep this technology managed in an effective way?

Husayn: As an industry, it's fair to say that, well, of course, giving access to more people whilst stopping fraudsters is the key constraint and is the key challenge. So yeah, that would be very much the case. So obviously, as a company, who got a number of different priorities, this is one where it's kind of the way it's structured is continuously improving. And the gap of the performance vis-à-vis alternatives is sort of growing, in terms of it being demonstrably better. So it's not top-of-mind, in part because it's organically improving. But there are definitely other areas that, you know, continuously need to be worked on.

Paul: And we were talking just earlier, at the beginning, about a topic which is artificial intelligence. Are you deploying and using artificial intelligence in combating that security threat or with the identification as well? Is it- is it something that's benefitting your organization, in deploying AI?

Husayn: Very much so. So artificial intelligence, in some ways it could be thought of as a generalizable model that is trying to replicate the functionings of the human mind. Just the way you could think of the Industrial Revolution and the steam engine and so on was trying to replicate the physical arm, artificial intelligence is trying to replicate perhaps in a way the cognitive mind. And machine learning is more narrow artificial intelligence, or specifically like a direct application of artificial intelligence to a repeatable task. For us, it's image recognition, for example. You know: Is this a driving license or is it a passport? For instance. So we deploy artificial intelligence, specifically machine learning, or narrow artificial intelligence, quite a lot. And it's a range- it's embedded in our core products of government ID and facial biometrics in a number of different ways. There's a different model to classify the ID, different model to extract the characters, and a range of different models to detect different fraudulent patterns and match that to the person's face and so on. And that has been why we've been fortunate in that being able to be effective.

Paul: And how long does it take to get AI to work in an effective way? And to deploy. It's an exciting area, an exciting field, about what capabilities it's got there.

Husayn: So it depends... the extent to which- what your model is. But if it's a straightforward AI model, to classify whether an ID document coming in is a passport or a driving license, you know, in theory it could be a few days. The more important thing is you have to feed it millions of IDs. Not this simple problem, but more sophisticated problems. Millions of different IDs to detect patterns and fakes and sort of aid the improving. And naturally, AI is- or narrow AI, machine learning, it's not applicable for every problem. So there's some areas where you're able to be more effective without using AI than you are. One other thing to mention is that AI intrinsically, or machine learning specifically, is more around spotting patterns or repeatability. The challenge with fraudsters is they're, by definition, the anomaly. So that has to be thought through well, and you have to configure it in different ways, to optimize for helping people, the 98%, to gain access, while blocking the 2%.

Paul: So... just something that sprung to mind, because we were talking about the aviation sector earlier on, and artificial intelligence, and just to help me get my head around this as well, is that when you go through those airport scanners - so you go up and you put your passport in - is there a suggestion that that is something that potentially could be obsolete in a few years time? Because what you're bringing in, along with artificial intelligence, you would not need to scan your passport, or you'll just be able to pick up a facial recognition and compare it to something digitally on your phone? Rather than a physical passport?

Husayn: In all likelihood. In all likelihood. It's going to be a lot more like the instructions for credit cards, where the majority will use this- more of a credit card digital base, where making a payment. And yet, for those who are more comfortable, they can continue to use a cash approach. And so it's hard to see a surely digital approach, without there being a cash alternative running in parallel, or an analog approach in parallel. But if the timeframe is more thirty to a hundred years from now, then there is a world where it could be done fully digitally, if done properly.

Paul: Yeah. I'm just thinking about the- like here in Hungary, in Budapest, we've got these number plate recognitions, and you just drive from one area to the other, and it just scans it automatically. And you don't have to buy tickets anymore for motorways, and of course, speeding fines as well. It's not done by individual cameras, it's done over the average. So like all of these things that I'm thinking out there that allow us- or involving artificial intelligence, are just learning all the time. Correct me if I'm wrong, but... yeah, now it just seems to be that's something that's being deployed in facial recognition and identification. Digital identification.

Husayn: It is. There is this clear application for what we do, because there is a lot of pattern recognition involved. With the example of what you just shared: We have that use case as well, in Italy. The company is called Telepass, that helps you drive through the tolls. And you can automatically register and make the payments automatically, but then you register your identity first, and your car, and then you register your credit card, and then you can go through automatically. One other point around artificial intelligence is that there needs to be a better global regulatory framework around it. In that as with all technologies, and this has been around for decades, but it's now more applicable and accessible, because of computing power and other factors that come into the equation. But there are downsides, right? Autonomous robotic warfare is (inconclusive) AI as well, for instance. And so there is a bit of a lack with the regulators catching up and making sure it's done appropriately, all from a data privacy side and ethics side and so on and so forth. So that is an area that we're involved with, and we're looking forward to being even more involved with.

Paul: Good. Good. I just wanted to come onto when you think about AI: I mean, AI and automation, in laymen terms, for the person out there that's listening, who's the business listener, for example - sometimes they can't distinguish between the two. But I know that you kind of- you're doing both. You're working with AI, but you're also looking for more automation. I think, globally, this is a global trend that automation's coming in. We get asked also a lot about automation capabilities as well. Is that something as well that you can separately distinguish and facilitate with your technology?

Husayn: So... yes. The way I would think about it, or the way that I think might help is: Automation has kind of been around for a very long time. It's specifically the First Industrial Revolution, which was around the steam engine and machines that automated the factory line. And so essentially basic robots doing basic functions, like the conveyor belt. It's just the basic functions of things moving along automatically, right? On a conveyor belt, as opposed to having to be- it being pushed or dragged along. That automated and made redundant the human labor needed, as far as your physical strength goes. Like the human arm was replaced in large part by automation in that era. The one now... and naturally, computing power came later, the internet came later, and they all made their contributions. The current one, AI, is somewhat different in that it's part of the same evolution, but this stage, it's more the cognitive functions of the human mind.

Paul: Yeah.

Husayn: The components that can be, again, spotting a pattern of whether this is a passport or a driving license. The human eye can do it. It's just, AI can- or narrow AI can do it much better. And that means that some services, human labor, that are specifically geared towards the cognitive mind and using your brain, essentially, is increasingly becoming less relevant if AI can do it. The call center is a frequently sort of quoted example. And so for small businesses, or any, there are two sides to this: One is if you're in an industry that can be impacted by AI, then you should think of additional skills or alternatives or how you could leverage it to basically offer a better service. And for, equally, another consideration is: You can now leverage other tools. And they're pretty powerful, because they're using AI to offer a better customer service to what you're doing. So even if you're not involved in AI in any way as a business, nor is it relevant to what you're doing in any shape or form - let's just say you're a dental practice - that doesn't stop you from being able to leverage other softwares and other tools to help you with your bookkeeping, with your financial investments and everything else.

Paul: Yeah. Yeah. I mean, we've got, now, a lot of inquiries about how they can- people can collect identifiable information, so personal identifiable information. How it can be sent into especially healthcare, because of course, you're not going into the clinics anymore, so they're looking for tools which can remotely collect this information in a very secure way. And I think this is something that we're seeing as well is that because of COVID, potentially things could change forever, in the way that people do things. So, not to revert back to going in and, as you said earlier on, it's like having to meet a cashier or show some form of ID, is that this will all be done either through collecting photocopies. Or even more so, is it one step ahead? Where you're doing it digitally, with what you guys are doing. Okay, so... what we're going to move onto now, I think we got ten minutes or so left... But I'd like to challenge you or get your feelings about the future of your, in particular, industry. But I think also about the direction in which artificial intelligence is going as well, because it's a fascinating topic, and you mentioned narrow AI - and does that mean that there's going to be a broader AI out there in the future? I mean, is society going to be- should society be worried about AI? Or should it be potentially embracing it in certain areas? 'Cause you touched on something a little bit earlier around the ethics committee and how much you want to do, going forward, to bring in some kind of regulation and standard.

Husayn: Sure. So your first question is: Is there like a generalized AI, generalized, well, AI model that our company is working on? It's basically what they're looking to achieve is essentially as close as you can get to a directly replicable human mind and all of its functions. But it is quite a bit a way away. Whether it's achievable or not is more currently like a theoretic exercise, academic discussion. As for the specific benefits to society, it's unquestionable. You know, if you have a scan of your chest and you have a cough and you are concerned: Is this cancer? Is this just a benign disease? Well, because of AI, the patterns of your lung and the way that it's sort of shaped and the colors can be logged, and the actual condition that you have, in a few years, also logged. And over time, as that is added to, well, millions of other samples, then you can develop AI models or machine learning models, to predict what condition you may have, over and above your standard data science algorithms and so on. So you can imagine that there are very many benefits. I just used the health example one. But in parallel, I just mentioned that, briefly touched on, this is as with any technology like... nuclear technology can give you nuclear power, it can give you nuclear weapons. In not too a dissimilar way, you can have autonomous drones and sort of warfare, purely by machines and without human involvements. And that creates significant risks. So, as with any new technology, the way to make sure you're maximizing the benefits and minimizing the costs, is to have regulation. Both in the sense of how and when should it be applied, and what should be the boundaries and guidelines? So is it okay, in places like China, where you have machine learning or AI-powered facial recognition tools to essentially surveil the people, have a mass surveillance program, and identifying political dissenters or whenever there's someone or a group that are showing political dissent, or you're targeting specific individuals and locking them up for political reasons, you know, that surely is not okay. Whereas there are many examples like that. So first is to have rules and regulations around limiting the access, but second, consider the societal implications, if it means there are going to be an increasing number of people who won't have a job, because machine learning and AI have come in (inconclusive) computed, or been able to offer a better service, then what alternatives are there for them? Is it a basic income approach? Is it retraining? Is it- what is it going to be? So there's going to be a need for these questions, and these to be thought through, properly and effectively. And for there not to be a "laissez-faire" kind of- or "let the market run as it wishes" type of approach.

Paul: What do you see, going forward, for your organization around the deployment of AI? Is this something that you're going to enhance, become more dependent on, when it comes to digital identification? Do you see there's the possibility to evolve even further? Because, obviously, technology is always evolving. What does the future hold for you guys?

Husayn: So the- what the future holds both for us and then the industry, if the question is "as a result of AI", then the answer is: The reason why there's so much interest in AI is that it can help a company gain a competitive advantage, and grow that competitive advantage disproportionately over time. So for- in our example, we have the market-leading approach of verifying if government IDs and facial biometrics are genuine or not, because of our machinery models. That means we're signing up more businesses. That means we're getting more data. And that means our models get even better, which makes us even more attractive to future customers. And that gap between us and others is growing. So in a similar way, the big tech giants - Google, Microsoft, Facebook, Amazon, and others - they're doing the same. And they- whenever they apply sort of artificial intelligence or machine learning, too, because of their privileged position and gaining access to more data, they're able to offer something that no one else can come close to competing with. And that creates significant inequality in the marketplace. Monopoly power and everything else that comes with it. So from a company perspective, like I'm not going to complain about the fact that we're actually the market leader, and we'd been able to become a market leader, because of leverage in technology machine learning. Going back eight years, we were the smallest company. And now, we're sort of the biggest. But in parallel, from an industry perspective, there is a... issue of all the bad things that come when you have any concentration of power. AI just accelerates that manifold. This ability to be a significant concentration of power. And again, the solution seems to be a lot of smart and thought-through regulation, so that society benefits from these technological advances, and it doesn't have to suffer the costs of them not being regulated properly.

Paul: Yeah, I was just thinking, when you mentioned... you know, some of the countries round the world that are deploying AI and identification. I just remember reading a story about China, about grading citizens about the way that they behave and things like that. With like all these video cameras all over the place. I don't recall exactly what that article was and where I found it, but I just remember thinking about something going along the lines of, you know, a police state or some kind of rewarding state for being a good citizen, which is probably, in my imagination, a push too far. But do you think, I mean, as well as it can be used for the good, I mean, it can be used for the negative as well. So what's the dangers of AI in the future then? I mean, could it suppress freedom of speech, for example? Or could it encourage freedom of speech for this? Will we see things in the elections going on, over the last couple of years, is AI to blame for that or a part of that?

Husayn: So AI can help accelerate some of the most dangerous parts. So if you want to figure out in a large population what reaction people have to different adverts, sort of ads, and based on their preferences and likes and dislikes, then you running machinery models just helps it accelerate, and then you gain much more powerful predictive algorithms, essentially. That means you can be more potent, as you are looking to sort of manipulate the way people think. So yeah, I think in practical terms, yes. I don't see China as being too different than any other country. So I know that there is a healthy skepticism of big tech and big governments, as there should be. China is just able to get away with it. I think most- wherever you have a concentration of power - and again, those who are in charge, you know, call it political, elected officials or otherwise - there is this natural tendency in humans to want to consolidate that power. And using AI, machine learning technology and so on and so forth can be used for those, and very effectively. So the only real difference in- between China and the other countries is less that they have necessarily different values or different thought processes, as at least when it comes to the political elite that they have making decisions, it's more a case of they are able to get away with it, just because of the authoritarian regime. But the danger is in five, ten, twenty years, unless we as sort of citizens hold these elected officials in sort of countries that we reside in accountable, then there's a very clear trajectory and path towards us slowly, blindly perhaps, walking into these authoritarian states or environments, where we are- basically, we don't have any say. And because of the pace of innovation being so fast, these governments can sometimes even utilize Chinese software to suppress and oppress their own populations. So that is why we all need to be vigilant and careful, and always question whenever it comes to civil liberties, and all the other values and principles that we don't just hold dear, but that are just basically fundamental to why it's special to live in the places that we live in.

Paul: Great. Great stuff. Great to get your insight there. I'm going to leave- ask you one final question before we wrap up, Husayn. So for Onfido, going forward then with- to serve AI technologies, where else do you think you're going to be successful at in supporting the need to clamp down on fraudulent activity or criminal activity, and keeping certain privacy over clients' data? Do you see that there are other areas that you're going to be able to support and help in the future?

Husayn: My hope is yes. So as an industry, we celebrate- or as a fraud and security industry, we celebrate ourselves all the time. We give each other awards. I was just the recipient of an award last night. And, you know...

Paul: Congratulations!

Husayn: ... everyone- thank you! We all give each other a virtual high five. The fact remains that, according to the United Nations, up to 5% of the world GDP is laundered money. And that's almost two trillion dollars. That's using human trafficking, drug trafficking and terrorist financing. So there is a- and 99% of that is successful. So just to repeat that: The United Nations says that up to 5% of the world GDP, which is almost two trillion dollars, is laundered money. And 99% of that is successful and goes undetected. So as much as, essentially, it's disappointing. I don't know if you watched last week's BBC Panorama documentary around large banks are being- essentially, there are question marks around their practices and them enabling significant money laundering happen.

Paul: Yeah, I... sorry! I just- I do remember the Panorama documentary, but I just read something regarding certain banks and laundering cash for criminals, yeah. I do.

Husayn: Alright. So in what your question is or what our hope is: So we're helping the neobanks, fintechs and now increasingly the mainstream banks stop small-scale, you know, fraud, so trying to launder five thousand pounds, twenty thousand pounds here and there, and gaming, gambling companies and so on. The large ambition, the longer one, is to feed through to the whole channels and make sure we're also stopping these much larger money laundering tickets as well.

Paul: Fantastic! That's great stuff! So Husayn, thanks a lot for bringing yourself on the show, along with your company, Onfido. It's been great having you, and having some insight, certainly, about artificial intelligence. The good and the bad. And I wish you all the best with your organization! Thank you very much for joining the show!

Husayn: (inconclusive) It's been a pleasure.

Paul: Thanks a lot, Husayn! And that is all for today's episode of "under CTRL". You can find links to all our social platforms and to our guest in the episode description. If you like the show, make sure you subscribe and leave a review. Join me again in two weeks' time for the next episode.