So, first of all, thank you very much for being here and Richard nice evening. So it's always hard like after the lunch breaks or after an extended coffee break to start the first presentation. So I hope I make it up a little bit likely and cheer up a little bit about artificial intelligence and how it can improve governments, risk and compliance activities. When we talk about this, we first think about how the challenges, when we talk about compliance, when we have numerous challenges in compliance, and I've been in compliance for the past 25 years as I was in AI for the past 25 years.
So what are the first, what's the first one, as we have heard in many presentations this afternoon, of course, the huge number of data that we have to deal with identify what are the key messages in the data? How can we make the right predictions out of it and very important, how can we make the results easier understanding this next one is, as we have also heard in various various presentations this afternoon, or in this morning, we are flooded.
It's not dribbling in, it's a flood of new regulations.
And the question is, is this tech invasion by Greg tech, FinTech, whatever is it really here to risk us? And it's AI, the holy grail that can help us to solve the problem we are in another crisis we have is the cost. The estimations differ a little bit between 10 billion. Like what is 10 billion in the world? Last year? The financial service industry has spent between 70 to 80 billion just in being compliant. And it's estimated that the number will raise 220 million and then I'll only the cost is increasing significantly.
The fines imposed by regulators are just sky working high, not talk about the huge number of people you have to hire to work in compliance is compliance really helpful. Can it be done by AI?
Yes, it can. That's one example. Example of us, an example of JP Morgan chase.
Over the last year, they have introduced a quite easy solution identifying keywords and key phrases in contracts in, in this case with contract and they claim that just this pre identification, what is the key content in the contract saves us 360,000 of lawyer hours every year just calculated by $50 an hour. That's a huge saving. That's why I clearly say if AI is implemented right in the right context, it will increase efficiency in significantly.
Reduce the cost and having been in compliance for 25 years, we've always been blamed. You are just cost money and bring nothing. And suddenly we can go back, Hey, and we can improve and save money. That's a complete new game. This is awesome. Is AI the holy gray everybody talks about? So what is AI? Is
It any actually Mr.
Touring, who in the fifties touring test, that's not him. That's by the way, from the motion picture, the invitation game, or is it even going back to other lovely, the lady, a famous lady in the UK over enrolled in 1844, the first kind of algorithm of linear system. Now coming to your question, how intelligent are we? Where are we right now? We are the very beginning of the food chain of AI. We can only call it narrow AI. We talked about go this morning, we talked about chess playing. That's what they can do best and perfect.
They can sift through thousands of documents, hundreds of thousands of documents, better than you can do it. But if you want to change a little bit to the context they fail, that's exactly where we are right now. If we talk, just reaching a little bit human level, we talk about another 20 years of research.
And if you wanna get even to super intelligence, the two of us won't really get it anymore because it's 40 to 50 years down the road. So this is really, really far, far away. So we are only talking about the narrow intelligence. That's what the systems are currently categor.
So we have to know where are the limitations? And that's a funny one talking about face recognition and some other examples that are really funny while this is still a funny example. The next one is a problematic one because, and mark this morning from Botanic stated, AI systems are written by Westerners. So I have to add are written by male white Westerners because they use biased data and intentionally, or in intentionally, they are including a kind of a biased algorithm into their systems. And while this picture's two years old, that's a reason research from February.
This year of MRG systems are still failing, identifying women and college people. So there are sex system racist because the data used to train these system is still biased. An example from the healthcare industry, is it really helpful?
Yes, for example is see these hundreds of thousand people running around with their kind of motion trackers and everything. And they communicate with their health insurance and exchange the data. Last year in November, the FDA, the food, the federal truck authority in the us approve the first digital pill. You swallow it. And then it controls how you take medication to your own Medicaid, etcetera. In Switzerland, there is a, Biovotion constantly tracking all your activities. Perfect. We can get the right therapies we can do early diagnosis. People likely find by 75% that the data is shared.
If the data is secure is it's secure, their blockchain might come in, but what can we do with it?
So for example, generally the product completely, when you constantly communicate your health data and your living healthy, your premium on your health insurance is reduced. It could also cause an issue last, you had a little scandal in Switzerland when the head of innovation of Swiss stated, if you don't share your private data, you don't get health insurance. This could be the tricky part of it where AI can get us soon.
Now, everybody talks about AI. If it's there, where is it? I unlike the statement from Mr.
McCarthy, that he made in the sixties already, if it's there, we won't call it AI anymore, which is logic. So there are already many, many AI solutions in the market. You gearbox is AI as far controlled. So there many examples of AI wouldn't call it AI anymore. So as soon as it's there in the market, we wouldn't call it AI.
And then we are waiting for, of course, the real intelligence of AI and the systems and that the companies are just popping up amazingly right now, like our company founded three years ago, this is just a universe in Europe.
So we talk about four companies in Europe, alone, almost 40 billion invested in AI research and development. And about 20 billion closely just on AI venture capital. I'm not talking about blockchain only on venture capital for artificial intelligence companies, looking at what the market says. For example, McKinsey, they see financial services in the hijack industry as the number one industries applying AI, amazingly enough, transportation only comes third. We all read these tickets about autonomous driving or company board for 30 billion for supporting autonomous driving.
But the key market right now are seen in a bunch of services and the communication industry, some examples, and I will go through this little bit faster.
So for cybersecurity, we have, for example, on stock exchange, they're controlling all the activities Doche bank they're controlling their it monitoring completely then Aetna. They develop their biometric identification system to improve eBanking then a, B, B, B a, and SK bank. They do constant customer screening and monitoring to improve their service recommendation while it's Fargo bank.
Funny enough being caught by little bit with accounts, they are have implemented the system that constantly captures all the data across all the scale systems throughout America, so that you can really track down what is the right data for client insurance, tech, underwriting, actually modeling claims handling or the German health insurance company AK or in 2012, they started to constantly analyze the data
Of 24 million patients every year with all their treatments, with all the hospital treatments, they have done to identify more risky patients, what a risky models, what a risky patterns.
And of course back take, we talked about JP Martin chase, but even the regulators are getting AI finger. For example, finger is screening 50 billion events every day, FCA just last week called up for input. How can we improve the regulatory reporting to the regulators, cetera, et cetera, the secs checking constantly in filings that companies have doing in the us. Do we identify risky patterns and findings by words used in the records. Now talking about AI, is there, what are the obstacles that have been identified in the market that project project have failed?
The biggest one is you do know this kind of statement for Carl Carl, who, which I said AI, which if you know, don't know precisely what you exactly want, you are lost in this huge tree of options of AI and bear in mind, each of these keywords on top is just the tip of the iceberg. There's a huge iceberg underneath. So when I wrote my PhD 25 years speaking about expert systems, we use newer networks as well. So this is the biggest issue that the market is not really aware. What is AI? What can you do? And where are the issues with AI and what are the different tools that you're doing for me?
Next one. And this has been said this day as well today as well. How to train the system? What kind of data do I have available to train my algorithms? Me personally, I'm not a fan of complete unsupervised machine learning. So the so-called new next deep learning approaches or 25 years making my PhDs. You don't know how they work. If you don't have the right data garbage in garbage out, I'm a perfect absolute fan of algorithms that allow you to use data. You have and combine it with the expert knowledge, which is, which is around. So why don't you use the expert knowledge?
If you want to use data, you make have to make sure that data is right labeled. And is there any buyers in the data and how do I get up another threat, which is made by the press mainly that everybody's afraid, oh, if AI is around, we are all unemployed. I want view absolute nonsense. I come back to this in a second, because what people are interested in is how can I best use AI to improve my work? There will be some jobs disappearing, but many, many others will actually pop up that we even don't know right now.
As I said, already, 25 years back, I criticized that these normal systems are all black boxes. What people want is open head blocks. I want to see what's inside. And how is it working? They don't really interest how the algorithm works, but they want to know how did the system come up with a result as it presents to me.
And this comes to the completing point completing point, what is needed to build successful AI solution for your company? Number one, and I like the presentation from IBM, make a corporate strategy. You have to notice certain extent. What is AI?
Not in detail, but at least a fair understanding. And then you have to design, okay, what's the corporate strategic approach to implement AI solutions in my business. Then you can say, fine. How do I bring AI to the business?
Now, like I said, I like the statement from the IBM. He said, don't do AI. Just because of AI. It will fail full stop, do AI to improve product as service. So business driven and bring it to business means two things bring to the product you want to support. You want to improve and bring it to the people. If you introduce it right to the people, they will accept it.
They will see, okay.
I have to, to adjust a certain extent, but I'm not losing my job. I'm improving my job. I'm improving my work and I'm getting more efficient and I can guarantee you. And I've done several researches on that one. And we've done many kind of pilot projects. It will significantly increase efficiency and significantly reduce cost because already last year, McKinsey was an article with title. If AI is around, where is it? And they have identified three key items that people normally do wrong when they implement AI solutions.
Number one, they don't really define, as we have learned, AI is a narrow AI right now. So it's perfect in chess. It's perfect in go. If you wanted to something else, it fails. It only does precisely what it was trained for nothing else. So we have to know that and have to precise, define, I want to do precisely this first step.
Second step. I have to know what kind of data do I have and then select the right method based on that data. And very important example of our solution. The output has to be nice and beautiful because people want to have transparency and very important.
People want to cooperate with the system. That's an example of cancer identification already two years old, but back then we see it's only by the error rate with the human will little bit higher than with the it system. But in the combination, the two had almost zero errors. It just illustrates. If you set it right to the people using it, they don't like it. How can the system be built? I like to keep them simple.
So we just build a three layer application saying, okay, we have to have an underlying layer gathering all the data, internal, external, analyzing it with a variety of tools, because it's always a mixture of methods you're using to come to the final result.
Centrally next layer. All the data needs to be categorized structured in the same way throughout the corporation. And then you can use the right prediction models to come up with the conclusions in forward looking results. AI can produce very important. The energy user will only work in the presentation layer.
They want to have the systems easy, understandable, and the results easy digestable in front of it. And this will allow you to say five constantly gather data, build your knowledge base.
As I said, I'd like the approach combining historical information and expert knowledge in your knowledge representation, use the learn knowledge to analyze the data, maybe build prediction models and present the results respectively. And then the service starts are over again. And of course the motor space is retrained on a regular base. If you needed with this approach, it allows you to build a system without any coding. It allows you to define what kind of external information you need. How do I analyze it? How do I get the key assets out of it?
How do I build the prediction models only ization? What kind of reports are helpful to present the results? And in the end, I just set up the system to run the continuously.
And this is exactly how I have experienced in the past 25 years, how AI works, how we can implement successful solutions and what the market is actually looking for. Any questions. I think I'm so time, two minutes left. I think I thank you very much.
