SCB Data X: Building a Culture of AI Innovation

In this episode, Dr. Shuki Idan, Chief Scientist, SCB Data X shares how they are using healthy competition to spark innovation, while ensuring AI remains a tool for human empowerment rather than replacement. From business unit battles to organization-wide KPIs, discover how one organization is turning AI adoption into a catalyst for growth.  

Transcript

AIBP: 0:00

Innovation often comes from unexpected approaches. For some organizations, the key to AI transformation lies in turning it into a shared mission through healthy competition.

Dr. Shuki Idan, SCB Data X: 0:09

So we are doing things like battles between business unit about ideation and about execution, and each company in the group, by the way got the KPI that certain amount of revenue should be should come from AI

AIBP: 0:26

True transformation isn't about replacing humans with technology. It's about empowering people to achieve more than they ever thought possible

Dr. Shuki Idan, SCB Data X: 0:32

To scale AI by itself has two aspects. The one that I think is the most important is really the for it to be accepted as a mindset

YY - AIBP: 0:47

The mindset

Dr. Shuki Idan, SCB Data X: 0:48

Yeah, and I'm not talking now about the, you know, computers and software. I'm talking about the the mindset that it and have a trust that this technology can do something for us.

AIBP: 1:00

Making this vision a reality requires more than just technology. It demands a fundamental shift in how organizations think about and trust in AI's capabilities.

Dr. Shuki Idan, SCB Data X: 1:09

Personally, I think that the contribution of AI today, even to enterprises, is more on a personal level, being able to do more, being able to be faster, being able to do things more independently

AIBP: 1:27

Transforming an organization through AI isn't about having the best technology. It's about empowering every person to be part of the journey. Welcome to the AIBP, ASEAN B2B Growth Podcast.

AIBP Intro: 1:43

The AIBP ASEAN B2B.Growth Podcast is a series of fireside chats with business leaders in Southeast Asia focused on growth in the region. Topics discussed include business strategy, sales and marketing, enterprise technology and innovation.

YY - AIBP: 2:06

Well, today, instead of us telling you about what's the trend in the market, we are very happy to have with us as part of our AIBP social and ask me anything segment Dr Shuki Edan, he's the chief scientist of SCB Data X, and he has a long journey in driving data and AI journeys for multiple organizations within, not just Asia itself, but globally. Well, tell us a little bit more about your journey, Dr Shuki.

Dr. Shuki Idan, SCB Data X: 2:37

So in fact, I'm, my first degree was in computers in which included, at that time, also hardware. So I'm an engineer i in my blood somewhere. And then, like many things in life, you know, things happen by mistake or by chance, or when I was on in my master of the degree, I walked into a seminar that talked about neural networks, and I was doing master in computer science. And I was, in fact, I was fascinated, on one hand, to see that, you know, people are thinking of imitating the computation, the very simple computation and happened in our mind and creates great things. This was one thing that fascinating. And the other thing was, in fact, that I saw it was a moment when a lot of there was this special moment. It's it was somewhere in 1984 1985

YY - AIBP: 3:43

You cannot reveal your age with this chat.

Dr. Shuki Idan, SCB Data X: 3:46

Yeah, I hope to go beyond the 24 but some somewhere here when I went to the seminar. The other thing that happened there was some discovery done at that time. And in fact, the person just got the Nobel Prize.

YY - AIBP: 4:03

Oh, Nobel Prize.

Dr. Shuki Idan, SCB Data X: 4:04

Yeah, but he got it two weeks ago. So for him, it was like 35 years, you know, to wait for the Nobel Prize. But the thing that was nice that people came from physics, came from chemistry, came from biology, came from psychology, came from different disciplines in trying to build this kind of that time. We didn't call it AI, but we talked about neural models. You know, because our brain is composed of small neurons, they do very simple things, but at the end, together, they create something which is performant, but what we can send, and this is my PhD in I decided to go to this area and to do at that time I did the PhD on on reading the zip codes of Yeah, and OCR in the case of handwritten recognition, because if you look at the things which are machine typed, you can easily understand, but when you look at people writing, it becomes a bit more complicated. But the fact was that there was not enough data. So at that time, I had to ask the postal office to give me some examples, and they made me go and work two weeks in the postal office in order to get the recognition that I can get this data. And the computation power was not there. But the key thing is data. In fact, if you can look pieces arrived somewhere at the 80s. So, you know, we started storing data. We started writing during excels. We started people started to store data. Companies were a bit behind, but it also so they started to collect data. But, you know, nobody in the post office thought that they should digitalize all the envelopes. This is some kind of exercise. And then, you know, there was this AI winter or whatever. But what happened to me in fact, I did research. I did theoretical research, and I did also applicative research. And the customers that we had at that time were more scientific, that had a problem in modeling something they could not solve with math. So, for example, I had a big tire producer in Europe, and they had problems of quality assurance. They at some point, this big tire, that kilometers of rubber that goes into their production started not to pass the qualities. Now, at that moment, they had to throw away all the kilometers of rubber and and this was a way so, but they couldn't. They couldn't model this production line and say, you know, when the temperature at this point goes above this, and when the elasticity goes out this, we have a problem. So they needed the model and this, this kind of thing was possible to do with the with the AI, again, it's a very scientific project. They had another scientific project with the army. They wanted to do autonomous vehicle, but not on roads, but in the mountains, on all kinds of terrain. And there is a problem to say, no, if I push the gas at that amount, this what will happen to the vehicle? It depends on the inclination, it depends on the terrain. It depends on so many factors. So again, the technique enable us to model it. But that was it. There was no, no other interest. Let's say around us what happened later. You know, I put here like four disruptions that, for me as a person, happened with the with computers and computation, you know, there is the PCs, then the internet and the mobile. So the mobile, at this moment, is just for calling an SMS thing. The real thing happened, you know, with the iPhone, with the smartphone. So it's not just calling, it's doing everything from the phone and and now AI is we can tell, tell someone to do something for us. And I think this, I will talk about it later. But what I also put on this graph is that that the technology, in many cases, exists. It's somewhere in the data center, something in the enterprises, but in order to be really life changing for us as consumers. It needs to be accessible. AI is there for many, many years, but just chatgpt that enabled us to communicate with that made it a disruption in our life. So going back here, when the internet happened, people started to do things online. Cloud appeared at some point, so it's more accessible for others to create the website, etc. But the online created another source of information. So here, in fact, this is the, for me, a disruption from the point of view, that we collect information. We can track orders, we can track what people are doing online. So this is a another point that contribute to data. And in fact, in my journey, if you look at Amdocs, it's a company that provides software for communication systems the organization. And started to think customer, consumer, trying to understand, and this is the place where our technology could analyze data of people. You know, the typical question was, at that point, with all the privatization that happened in the in the communication, the competition, the mobile and all that. The major question was, you know, who is the customer that is going to leave me next month? And it was all about retention, but looking at the data without tools was not a good thing to do. So this is the place where we again, we did machine learning. We couldn't explain why this guy is likely to live, and what this guy is likely to stay. So we have data, we have some algorithms to analyze, and then we can predict something. So this was about predicting things, mainly around marketing, who will churn, who will buy something, who will do something. The next thing that happened on my journey is, in fact, once they knew to how to predict, the problem was how to use it in the business process. Because if you predict who will leave the company, who will churn as a customer in one month, and to do the retention process will take you two months. In fact, you will be calling people that are not anymore there, you know, they're trying to retain them. So there was and still, you know, business processes, with all the sophistication, the business processes are critical in, you know, in executing, in operating. So this happened around this area, where people were looking to optimize processes. Then on my journey, things became more real time, so we looked into recommending, what is the next best action? What's the next best offer. I did five years ago things, a lot of things around bots at that time, you know, banks that wanted to provide the bots to do service, but really the breakthrough with Gen AI was that, at that time, the challenge was, you know, once you understand what the bot needs to you understand the customer, what he wants to do, he wants to transfer money, you can look at the account of the customer and see what needs to be done. Because there is not enough fund. He needs to put money or to sell stocks or whatever. How do you communicate it back? You know? So at that time, we had the template. We had you have to deposit space this amount of dollars, and we took from the system the number that he had to deposit in order to transfer. But that was it. The nice thing with generative AI today is that the discussion is natural. It's not that we are using templates. We are not putting numbers. And this is this is the disruption, at least for me and others. The other thing is that we can communicate with AI. We can ask him to do something else for us, not It's not that we have just about they can do money transfer. We can ask him to do anything that we want. letters in order to do a dispatch of letters. This was my subject. But I and trying to do it with techniques of learning, you know. So show machine a lot of examples of ones, two, zeros, etc. And at the end, you know, show it something, an envelope that you didn't see before, and it should classify that this is the number.

YY - AIBP: 13:03

So OCR in the 80s so the world's the limit right now, if I could kind of summarize what you've mentioned, right, the basis behind say, AI was always there, even back when you started that journey, when the Rubber Company came to you, where they had a problem statement, how do you then have equation to solve this problem? But perhaps there was not enough data, compute. Power was not there, so you couldn't really scale it to the next level.

Dr. Shuki Idan, SCB Data X: 13:36

Mainly data.

YY - AIBP: 13:37

So now there's a lot more data, I think Chatgpt or all the large language models have already been developed, and now, instead of perhaps having specific knowledge in order to do if x do Y, you can just use very natural language to query and the world's your oyster, right? You can kind of ask all the AI models to do anything for you is like a perfect assistant.

Dr. Shuki Idan, SCB Data X: 14:06

Yeah, the big question is, you know, how good is this assistant? Current studies show that, on one hand, AI is a wonderful coder. This is, and you know, if you go back and think about it, computer science coding, coding is simply something that the machine can learn. I mean, it's you have a certain set of instruction, you have a lot of examples out there. And so if you look at the numbers, when you have coding tests, AI get 93 point something on all the other fronts, it's average. So it's around the 70s. You know, there are people that even takes those AI to IQ tests. So they let them, you know, run IQ test, and you looked at the result, the average around the 70. So on one end, if I don't know anything about this domain, 70 is great. It's good enough, yeah, and it's not bad. But still, I think we are still in the seat. We are looking for new advancements, and there should be more disruption in, you know, going to the next phase. And we see funny things, that a lot of these models by different vendors, like open AI and tropic and others, they converge about the same. There is some someplace that we need. On one hand, what we have. We need to use it and to build more on it. On that hand, we should expect some other disruption in the architecture, in the algorithm. There should be something else happening

YY - AIBP: 15:52

It is a very big investment that many of these large tech companies are spending, right? But if I go back to your experience, because you have a wide range of experience across different sectors. Right, from what you've mentioned, from an academic approach, from a product approach, from consulting approach, how has that changed your view of data science? Because at the end day, there are two camps. Right? Some people like for internet. When internet first came out, you like, some people think that internet was just a fad. Now, some people say that AI is a hype cycle. It's going to go away. What are your thoughts around that?

Dr. Shuki Idan, SCB Data X: 16:31

So I think that the I think that it will not go away. I think that the big thing that happened is that, you know, forget about the benefit for now. But if you look, there is a hype also that happens in the number of AIs that come out. I think that now the number is around more than 10,000 say I are waiting out there in the web, and they are specialized in something. They are specialized in creating videos, pictures, analyzing, general purpose, whatever. So on one hand, I think we will see that, and people are talking now about agentic AI, so we will see AI that are specializing and again, here data becomes, again, the differentiator. If I have my data and it's something that others don't have, I can fine tune my AI and get something that the others cannot get. This is one thing, and the other thing is the creativity. I think that there is a lot of creativity, and people will continue to use it because it's good enough, let's say so. I believe that, you know, it will stay with us. Probably the mode of interactions will change. You know, I may ask my AI to plan my next week travel to India, and this AI will talk to another AI that can do bookings. We'll talk with another one, the find driver. There will be perfect person that existed. Yeah, there will be and there will be specialized AI and but, but I think it's here to stay as a tool, as a technology. I think it's here to stay.

YY - AIBP: 18:17

I think this is the perfect sideway to go into your role at SCB data x, right, maybe to our audience who may not know about SCB and its journey to become a technology company, SCB to SCB X to now datax being spun off. Tell us a little bit more about that journey.

Dr. Shuki Idan, SCB Data X: 18:38

Okay, so back in 21 in fact, there was a decision made before I joined the bank. At that time, SCB, that you know, looking at technology, but also looking at the FinTech industry, which is much more agile, which is much more innovative, that the bank and the management decided to become, instead of a bank, traditional bank, to make a switch and become a fintech group. So on one hand, there was a business side for it to spin off. Businesses to be independent, with their own PNL, more mobile in maybe also to avoid the tough regulation in some places. So the bank span of, for example, the credit card business to be a company by the name of card x, it spend the investment arm to be invested. So some startups, let's call them startups. Our business is spin off at the same time we start. We we started a lot of new companies that mainly try to approach the unbankable segment, which requires small loans with high risk. So several companies. Uh, looked at this area and but the other side of it was to really to create the technology company. So for example, there is a company by the name of tech x, which is a technology or IT company, and this company can help the small companies to go live instead of investing and hiring technical people, they can do the project for them and last but not least was the belief that data and AI is the future differentiator. So this was the the idea behind the creating data x and the mission was to leverage data from all those company to be able to share it, to share insights, to identify opportunities. You know, if I, if I paid to Robin Hood, which was another company with the premium credit card, you know, we understand that there is someone who is wealthy, and maybe we can do with other companies, can offer him other things, etc, so leverage the richness of information, and, of course, getting information from other places. So we are talking here about one and a half year before Chatgpt, but the thinking was there.

YY - AIBP: 21:17

Yeah, I think I was just sharing with you, Dr Shuki, that you are the second guest for Ask me anything. The first one was Dr David hardoon from Aboitiz, who it's also a visionary in the data and AI space. So one of the questions that we always talk about when we speak to Chief Data Officers, right? It's like, the importance of scaling data? Because we, at first, we talked about last time there was not enough data down, there's a lot of data, but the AI model, but the AI models, how do you scale that? Right? What are some of your thoughts around that? What are effective ways of scaling AI within organization?

Dr. Shuki Idan, SCB Data X: 21:52

So to scale AI by itself has two aspects. The one that I think is the most important is really the for it to be accepted as a mindset

YY - AIBP: 22:07

The mindset

Dr. Shuki Idan, SCB Data X: 22:08

Yeah, and I'm not talking now about the, you know, computers and software. I'm talking about the the mindset that it's and have a trust that this technology can do something for us. I'm not, I'm not saying that everything should be done by AI, but people should think about AI as a way to do something. And the second thing that comes into mind is the data. Do I have the data to really to do that? So thinking data is important, looking at numbers you know, not operating by hunch and

YY - AIBP: 22:42

By feeling instinct.

Dr. Shuki Idan, SCB Data X: 22:46

Being able to be more critical about what is going on. So I think this is the essence of it. Now, really, for to scale AI, you need to have the good use cases. Get good use cases, and then, you know, build on that and do that, but in most cases, it will require some changes in the way we work. So we have to keep that in mind, like in any new technology. In our case, you know, we looked at it as a challenge of education. So, in fact, every person in our 3000 people group needs to do AI foundation course.

YY - AIBP: 23:37

Oh, wow. How long is the AI foundation course?

Dr. Shuki Idan, SCB Data X: 23:40

It depends on you. I mean, I could go exactly immediately to the exam.

YY - AIBP: 23:45

There are five

Dr. Shuki Idan, SCB Data X: 23:46

There are five chapters. So if you, if you, if it's an online course, but if you don't pass the exams, I mean, you have each chapter, you can few hours of listening and with Q&A examples, and then you go to the test. But it goes from the teller to the CEO.

YY - AIBP: 24:04

Interesting.

Dr. Shuki Idan, SCB Data X: 24:05

So everyone should understand a bit about it and understand, you know, when it's good when it's bad. So we took it as challenge, and the same thing we are trying to create, you know, I don't know an activity around it. So we are doing things like battles between business unit, about ideation and about execution. And each company in the group, by the way, got the KPI that certain amount of revenue should be should come from Ai. So it's some kind of a crossover

YY - AIBP: 24:39

competition of sorts. So you make it into a game, and then, like the units become more incentivized.

Dr. Shuki Idan, SCB Data X: 24:44

It's a game. But also, there is also what numbers that they meet they need to meet

YY - AIBP: 24:48

Not just a play game. Well, today is Ask me anything. So we do take questions from the floor. I think one of the challenge, or should I say, the common questions that we get to speak to a bank is regulations. Right in each of the countries, there's different regulatory authorities, and in our discussion with Dr David, he talked a lot about that, because he used to wear the monetary authority Singapore hat. In this case, what is one of, maybe some of the key challenges that you face, or perhaps organization face, when it comes to managing the bank data, and how do you cope with it? Because in your scenario that you identify, right, you get everyone to be familiar with data, to be comfortable with playing with data, right? But how do you set up a governance structure to make sure that the data privacy and all the rules and regulations are still being managed.

Dr. Shuki Idan, SCB Data X: 25:43

So it's a headache.

Unknown: 25:46

But that's a very good way of summarizing it.

Dr. Shuki Idan, SCB Data X: 25:50

But you know, the real headache is around privacy, and privacy in our case, has several levels, because there is the privacy between what the customer enables you to do with this data and there, by the way, there is, it's not a big problem, because most of us, when we go to any application, there is some accept in the first

YY - AIBP: 26:20

Otherwise, I cannot access any part of like the application after that, right? So accept,

Dr. Shuki Idan, SCB Data X: 26:24

Yeah, we don't read it anymore what is written. So data sharing with the among people is solvable, and this is not such a big thing. The other thing is, in our case, is sharing information between the different companies.

YY - AIBP: 26:42

Oh, that's interesting

Dr. Shuki Idan, SCB Data X: 26:44

Because if the bank has an agreement with 7/11 to get data from 7/11 you know, we are not in this agreement. So we should look for some creative ways to use this data if you need it, by anonymization, by doing things like that. And there is a bunch of technologies that even look on building models, but not moving the data. So, you know, all kinds of distribution technologies to do to be okay with the law. And I want to use this data. Don't care that you know, you bought two bottles of of energy drink, but someone which is some, maybe you bought energy drink and maybe bought sandwiches, something so. So some of the things we can solve with the technology, the other data, side of things, beyond privacy, is AI brought to the attention of the regulators all kinds of things that exist since centuries but were never monitored. So all this thing about trustworthy AI, that our decision should not be biased, that we should not discriminate people, that the discussion should not include the, you know, hate or

YY - AIBP: 27:59

racial discrimination.

Dr. Shuki Idan, SCB Data X: 28:01

Yeah. So all kinds of things that existed there for many years

YY - AIBP: 28:06

Come to the regulator so they are meant to keep track of what the public is concerned about, maybe.

Dr. Shuki Idan, SCB Data X: 28:12

yeah. But I think, you know, I think we try to be one step before the regulator, the regulator, it's still a work in progress. And, you know, there are all kinds of standards that appear in different countries. But, you know, the culture, the sensitivities are different, all kinds of things. So, you know, we build our own capabilities. So we have, we are building capabilities to really, to secure the AIs. It's if somebody is picking things that should not be there, should be blocked and reported. If you know another thing, an employee sends some confidential information to an AI asking to do something, you know, we need to discover that. So all that we I think we even one step before the regulation and and many things came, you know, that didn't exist, no explainability. This was before we had mode. We had the credit models. They gave decision. Nobody asked, you know, explain me why didn't give a loan to this customer or that customer. So many things came to the phone, the regulators are working on that. And, you know, we want to have regulation in this case, you know, because we want to know where the limits.

YY - AIBP: 29:30

the limits, right? So when you have the limits, it's a lot clearer. Yeah, I think if I could just ask, this is one of the question from the audience itself, like we talked about AI, and just now, we started off by looking at AI. It's a it's a series of problems that you try to solve with, like equations, right? How important is AI in relation to data governance? We touch a little bit upon it just now.

Dr. Shuki Idan, SCB Data X: 29:50

So AI is helping us a lot in data governance. So we, from the first day, we implemented, you know. Were Data Governance Initiative and tools, and you know, it was also deploying them across companies, because it's not enough that on our most of our data comes from those companies. So it's we wanted to be able to track things until the source. But data governance as a work is, again, I'm using the work headache, because data keeps changing. And if you want to have and dictionary is an explanation, and you know, you want to know the lineage, this data comes from that source and on that moment, and it's a combination. So it's a tedious for a person to do it by itself. It's a tedious work. And, you know, I don't envy those people that need to be data stewards. But today, with AI, we can do most of it automatically, so the AI can track, look at the database and understand that this field comes from that place. And there was a change. And most of, by the way, most of the technology vendors today that you know, like data breaks, Informatica, Microsoft, offer those AIs. It was one of the areas that was a sweet spot for AIs to do this, let's say bureaucratic work of managing data.

YY - AIBP: 31:20

Those are, I think, technologies that we've been seeing quite a bit of, right? But I think one of the interesting thing is, when we go to the individual Southeast Asian country, they always say that, let's say if I'm using entropic or if I'm using ChatGPT, right? OpenAI. Much of it is not catered to local language. And perhaps, for example, I think we talked about Thai food, right? The local Thai language model tell us about this whole journey of data apps really taking the lead in trying to drive this. I know that perhaps there were a lot more people that was included in this journey. But what got you started in driving or putting an investment for typhoon?

Dr. Shuki Idan, SCB Data X: 32:12

Sorry about that.

YY - AIBP: 32:13

No problem.

Dr. Shuki Idan, SCB Data X: 32:17

So I'm not the typhoon is not my project, and I'm a bit feel a bit uncomfortable to talk about it. But Thai, for example, is a difficult language. Because Thai, as a foreigner, I can but if you look at the text, there are no spaces, so you have a series of characters, and you know you do the segmentation, because you know to recognize the syllables or the words in it or whatever. So when we when we first started to work on llms, and we used what was available in the market, like open AI, we found out that they don't know how to break the this set of characters into llms. What they do, they create tokens in English. A token will be a word, the equivalent of a word. And you pay by token, by processing of a token. And when we took characters in Thai, we got, instead of talking per word, we got two tokens per character.

YY - AIBP: 33:31

For one character

Dr. Shuki Idan, SCB Data X: 33:32

One character, because he couldn't decide, you know how to split it. So many of the use cases become impossible from an ROI perspective, you know?

YY - AIBP: 33:45

So you had to develop it yourself, because there was an incentive to do so.

Dr. Shuki Idan, SCB Data X: 33:49

so this is where we embarked on on doing things on ourself. By the way, the main, let's say, innovation is happening on the token what is called the tokenizer, the thing that breaks and with some heuristics, you can do it better and then improve the understanding of Thai. The second thing is the to have enough text to train, because those things they call all the vendors, they collect text, and they train it on different languages with time, they improved greatly. Today, Typhoon is the core engine is Lama, three point, something which is open source. And so the engine itself today, they are doing enough. They are good enough in doing that some, some work is done outside the model. And another thing is, is collecting data in order to really to specialize and find unique to what you want to do with it. KBank, which is another bank they have their own. They developed by FinTech, one again, taking relevant information and which is related to their business and trying to improve the capabilities of the chat bot.

YY - AIBP: 35:03

Thank you for sharing that, because I think you have a lot of experience right in seeing the journey. So we thought that it would be good for you to shed some light on that. But going back to the core, right, when we first started, we talked about how AI is meant to solve problems, right? And I think there is sometimes the feeling that people are fearful of AI, sometimes, and if I could look into that, ideally, AI should be seamlessly integrated into workflows where, like you talked about how the different teams are competing with each other, they are able to find how AI works best for them in their particular use case. How do you set an assistant to ensure that AI actually does that so it can bring up the most value to the humans that's utilizing AI in your organization?

Dr. Shuki Idan, SCB Data X: 35:57

So personally, I think that the contribution of AI today, even to enterprises, is more on a personal level, being able to do more, being able to be faster, being able to do things more independently. You cannot. You don't find like end to end things which are based on AI, okay, we still don't see it. We in the back, we have the traditional models that create predictions, scores, things like that. And this is embedded in some process I don't see today, any you know, end to end process in a company that is AI based. So this is one thing, the other thing, as in any process, when you use it, you need, you need to measure things. And AI adds to the set of metrics, few more metrics that you know people need to understand and be understand that it is statistical model. You know that it makes mistakes. What is, what is the cost of mistake? You know, so we need to be aware of it at the same time. There are many things that block. I don't know if to say block, but don't let us go quickly with AI, because, you know, there are all kinds of processes that, when we have a model, needs to go to some committee. They do model risk management, and you have to do simulation and show them something, and they need to see that, you know, if, if the model do mistakes, you know, we are still safe, etc. So this

YY - AIBP: 37:40

Traditional risk management

Dr. Shuki Idan, SCB Data X: 37:41

Yeah, we shouldn't. We need this risk management. But this risk management need to change if you want to be very quick and you know,

YY - AIBP: 37:49

Way of looking at risk in the AI space

Dr. Shuki Idan, SCB Data X: 37:51

Yeah, and a new way of measuring, and probably, you know, being more on the production than you know once before we go live, but really monitor, and we have today the tools to monitor, to see that something is drifting, you know, and things drift, data drift, you know, economy change. But since this system also learn, you know, it's not that it if a good AI will always learn. So we cannot go every day to model risk management committee and ask, you know, should we accept what he learned yesterday?

YY - AIBP: 38:23

So there are things that some margin that you have to work with.

Dr. Shuki Idan, SCB Data X: 38:29

Margin or other ways of thinking, you know, you know, maybe let it, let it go. And just when it go, goes out of boundary, you know, change it over, send few of them and let them compete. You know.

YY - AIBP: 38:44

Kind of like a child, right? You cannot really set very minute instructions, like you need to give it like some boundaries, and you work within those boundaries and let them I want to take you up on that. So you talk about learn.

Dr. Shuki Idan, SCB Data X: 38:55

But another thing is, you know, the good thing with AIs is that it's a machine and it doesn't get tired. So, you know, send 10 models out there and let them compete, let them and, you know, see how it goes. And I think this is a useful thing, if you talk about the benefit of AI the way to try many things. You know, I have an idea how to do setting out 10 AI and then seeing what comes back, right? something, but there might be 10 ways in doing that. If I let the AI, you know, crunch it a bit and come back, you know what will happen if and then take the decision. Any specific use case that was very surprising, that brought about quite a bit of benefit for the organization. So, by the way, all those new AIs, they do that, they they do some, some coalition of

YY - AIBP: 39:56

A/B testing. Now is A to Z testing.

Dr. Shuki Idan, SCB Data X: 39:58

Yeah, so all the improvement happen now it's, there is some mechanism of many AIs trying to do this thing, and something needs to decide which one is the better. So this is one comment. The other comment is, you know, I was involved in some strategy thinking of doing something, you know, should we increase the price? Should we do something to send two products at the same time? And we did some crunching with with AI. It's not a huge, huge case, but it's very anecdotal, but it's an example.

YY - AIBP: 40:35

Are there Because we always say that AI being used for the business, often it's two cases, right? You increase the revenue or you bring down costs, right? Any examples that you can highlight about how, perhaps your AI projects have brought about some of these outcomes for the bank

Dr. Shuki Idan, SCB Data X: 40:59

Currently, not, at least for the Gen AI, I don't have concrete examples because, I mean, I have examples, but it's not something that I will

YY - AIBP: 41:08

List as mind blowing

Dr. Shuki Idan, SCB Data X: 41:11

On the next analyst call talk about it. But most of the use of that we see now is about reducing cost and what the other side of it is, having people in the loop. Because, again, it's,there is a question of trust. And so, you know, we have internal applications. You can, if you have questions about HR, you know, how many leave days I have left, or something that you can, you can ask a bot. So, you know, before we had few people that this was the daily job to get questions and to answer a bot can do that. So to you, it's but it's a playground. It's not, you know, for me, a transformation investment agent. They have an assistant that tells them, you know, looking at the customer, looking in the market, looking at the analyst, reports he can, he can give you advice. Still, we don't give it directly to the to the customer, because,

YY - AIBP: 42:17

if not, you're liable for it.

Dr. Shuki Idan, SCB Data X: 42:18

Yeah, liabilities is a everything. But you know, on the other hand, we do things for regulation, for example. So we have people that agents that give loans and they talk with other peoples. The regulator came and he said, you know, you have to record it, and you have to check that in each call, six, seven things were said, you know about this amount, you will return like this. So this is, you could put some guy that will listen to every call, and we'll say yes, yes, yes. But, you know, AI can do that. So we are taking the audio, making transcription the AI goes and, you know, flags it.

YY - AIBP: 43:00

So taking into the fact that AI doesn't need to sleep, right? So you can actually use it to do a lot more in terms of workflows and automating those processes. Well, we've talked a little bit about this, and this is a question from the floor with regards to, like, privacy, right? Which brings me to the point of what you spoke previously, responsible Ai. It's a big topic right now all across Southeast Asia, because people are scared of like biases, right? What if you put in wrong data? What if this is not gender neutral? All of this? How do you balance innovation? Because you need to constantly push the envelope, still with the need for Responsible AI?

Dr. Shuki Idan, SCB Data X: 43:46

you have to introduce tools that will help you to do that. So, so we are implementing, for example, tools that are in the category explainability.

YY - AIBP: 44:00

Explainable AI

Dr. Shuki Idan, SCB Data X: 44:02

Yeah, so it's Explainable Ai, but it also looks at the data, for example, and identifies there is bias in the data. Or when a data scientist build the model, it verifies that the results of the model are still fair. So it enables the explainability. Goes, you know, from the data scientist to his manager, to the MRM guy, to the business owner, and then this is on the development cycle. But then, you know, on production again, it's operation people that needs to see, it's the agent. It's sometimes even the customer needs to understand why he didn't get something or what. So explainable AI is very, very bored, and it covers many stakeholders and many use cases. So yeah, we have to do it in order to deal with this new requirements, security, again, it's another, it's another place where, you know, we want to verify that people are not, you know, the board does not discuss about family consulting or worse. So all that we are equipping ourselves with new tools, new capabilities. It requires a lot of research to find what is, what is meaningful, and then how to customize this to our place, to our language, to how again, we meet all those questions about typhoon and others. You know, I can monitor discussion in English, but, you know, Can I do it in Thai? And what I need to do in order to be able to do it in the right way, the all kinds of specialities. You know, in Thailand, you should not, for example, say bad things about the King. I'm sure that it doesn't exist in all the

YY - AIBP: 45:49

Other countries.

Dr. Shuki Idan, SCB Data X: 45:51

No, no, in all the off the shelf solutions

YY - AIBP: 45:55

Off the shelf solutions open AI probably has not included that.

Dr. Shuki Idan, SCB Data X: 45:59

Yeah, so we have to respect that. But in order to do that, we have, really, we are doing a lot of research to use, but also to find, you know, new solutions that can help us keep the pace.

YY - AIBP: 46:16

One last question, we talk about responsible AI, there's also been talk about the energy usage of AI to power all those GPUs takes a lot of power. Any thoughts around that?

Dr. Shuki Idan, SCB Data X: 46:33

So, you know, I remember in the past, I remember I knew the number that not even when you Google, you make a query on Google how many miles you could drive with the car to make the same carbon print. But I don't have a solution for that, but my hope is, in fact, that with AI, we can also find the remedy for the other side, you know. So, you know, there are now systems. One of the things that happened with the fact of the with the carbon print is that we have the weather got wild. You know, we have floods. We have fires. So there are solutions now. They can predict floods. So we are trying to survive by the use of AI. We are killing ourselves on one side and then trying, my hope that there will be some, some, some change, you know, even, you know from other domains, maybe eventually we'll have gold fusion, we'll be able to create the power sources

YY - AIBP: 47:49

Efficient

Dr. Shuki Idan, SCB Data X: 47:50

Yeah, but it should happen, and hopefully it will happen, but

YY - AIBP: 47:58

Those are things that we have to consider. But thank you very much Dr Shuki for joining us today, for sharing your experience in this. For all of you who are watching this, when you have any other queries about the tool that Dr Shuki could be using, you can reach out to him on LinkedIn and find out more from Dr Shuki himself, or be part of our community, and we'll invite Dr Shuki to share more.

AIBP Intro: 48:28

We hope you've enjoyed the episode. For more information about business growth in the ASEAN region, please visit our website at www.aibp.sg