Inside the Global Financial AI Race

[Music] thanks everyone for joining us and to this open opener discussion on finance today you know here at the abfw auditorium stage uh this is a very interesting discussion we have a lot to cover so I'll get right to it with our esteemed guests and to really understand you know as Khal was saying around AI what is it and where we'll be discussing about the finance space you know um it can range from simple automation to generative AI so you know especially in an industry which uh security along with Healthcare uh you're dealing with a lot of private information sensitive information uh this is uh something really interesting to explore so I'd like to put that question out there first when we're speaking about AI in the finance space what are we talking about what are the benefits currently uh being used uh in this Finance space uh I'll put that out to Casey to begin sure um as a ex bankr Goldman Sachs I have to say that my old old days as a banker my daily job has very little to do with technology in AI to be honest and I saw my team spend a lot of times processing Manu work mechanically most of the time and I think the world of Finance has really changed a lot over the past year I think um the generative AI brought a lot of open mindset to to the people on the ground in terms of appications but if we talking about the general applications of of AI I think it's quite broad so it can be value added to each aspect of of the operations of the institution so starting from B generation very importantly in this termal time everyone wants to get the top line revenue right so AI leveraging deep learning the knowledge graph technology can help for example FAS to identify high quality companies fast growing companies before they become unicorn and be be surrounded by various FAS and they're also the technology that can help corporate Bankers to identify the high quality companies for Lending and these are are the applications that are very powerful not just getting you the name but telling you the shortest C the shortest path to to reach these people so that's very important and also I think um the other way AI that can add value is customer engagement from simple applications like using robots to provide customer service Q&A to some more provan um more advanced applications that we provide to for example wealth advisers that they use the technology solution to serve their end users doing automatic portfolio assessment health checks as well as the rebalancing of the portfolio recommendations based on the market movements based on their goals risk tolerance Etc so and I think the most critical part is when it comes to productivity enhancement and that's where I feel most passionate about in particular if the solution is curated and embedded into the workflow of different roles where you play in the financial ecosystem so we have solutions for um research analyst so there there's technology that can help help you to pull out the latest quartery earning results in a single clicks and also help you to update your valuation model based on the latest macroeconomic industry um other company level assumptions and inputs and also the sales size solution for investment bankers allows you to pull out the pitchbook as well as company profiles in the simple CS rather than having analyst to working on it the day and night as well as if you a Trader for example you might need to report to your boss about your why did you lose this money on this trade and in what happened in the day so AI can really help you to pull all these stories together this is the trade the performance over the day this is The Benchmark and then what happened in the day throughout the course um positively negatively that drove those movements and last but least I think um compliance operation management is at the core of financial institutions these days and then I think it's it's in the past as well so a lot of value AI can help to to for example help with automatic due diligence um ekyc identify fraud and transactions and related party transactions you these Knowledge Graph if we have the this information of penetrating the macroeconomic the industry value chain the company level information intervene everything together this knowledge graph is really powerful um when it comes to these operation and risk management that's a great answer and I think that was very comprehensive and identifying that it's very integrated on all facets from the customer side you to the be uh behind the scenes it's uh it's some AI isn't something you can say this is AI it's across the board and um Ron I'd like to extend this to you in terms of how is traditional Finance base receptive to this um this is always something where the finance bases and along with Healthcare and others and security are maybe slower to adopt you know these types of things because the risk might be higher there are a lot of unknowns can you talk a bit about that at the current stage maybe to what extent is AI being used when because um Casey had mentioned you know a broad spectrum of things to what extent are we actually using and how receptive is the traditional Finance space to it sure S I think Casey's already given a great overview of all the different use cases so I won't go through those again um maybe I'll zoom out and say without being boring um we need to Define our terms um I used to be an analyst for many years so what do we mean by AI because a lot of what banks Enterprise government does is machine learning regression analysis sometimes very high level data analysis but isn't generative AI now there's a lot of projects Pilots work streams underway on geni and over the next year two years there will be a lot of Genna used inside Financial Services why information what's the raw material of our industry in finance it's data its numbers its words and what's our output yes we provide as an intermediary financial services but again the output tends to be words and numbers we don't make you know objects we help these objects get made so in a way our industry is ideal for both discriminative Ai and gen Ai and any sort of variation of that but if you zoom in and ask any major financial institution not how many AI project s do you have but how many gen AI projects are in production across the firm the numbers are very small if you said how many AI ml projects do you have typically in a large organization you're looking at several hundred depending on how you define a model or a project you're typically looking at hundreds um again these numbers can be put into thousands but again depends on definition um where we see going forward the biggest and best use case uh in the next one to two years I think Casey used the word productivity we see significant benefit in productivity both for our technology staff our developer staff and for the non-technology staff and maybe just two examples um there's about in major Global Banks there'll be anywhere between 30 and 50,000 developers or technology staff depending on how you define it um obviously for PR purposes we use bigger numbers uh so we put all our Tech staff into developer so public number is about 40,000 now those 40,000 developers are going to be do a lot more in terms of basic coding code generation code checking code op this this this is this is going to lead to significant boost in productivity anywhere between 30 to 50% on basic tasks in coding 10 to 20% in more advanced tasks so you're going to see significant benefits in that space in terms of non-technology stuff um it'll be initially internal because none of these models are 100% foolproof so we're not going to be using it that much for external high value contact but internal analysis internal just kind of going through your existing content there's whether you're a bank or a consulting firm or any large organization there's a huge amount of content usually outside the reach of most employees so information retrieval analysis and management is going to be the major other use case for non-technology Stuff wonderful and I think that speaks to some data that I've seen online that around 70% of companies uh are aware that you know AI is playing you know going to play a major role in finance uh it percent it said it was the future again this was from various studies so I'm just giving very general numbers but then also the R&D phase is is U where a lot of attention with AI to find those use cases and as you said internally more than externally you know dealing directly you know with customers or with clients so I think it's really interesting that to identify the stage where it's a it reminds me of the crypto approach where it's a bit more cautious approach but using internally to understand it first you know before um using it uh publicly or you know with clients directly and uh Professor mwan can you give an example of what that looks like in terms of um you have like a you know from your research like have you seen like case studies you know where to understand what problem AI is solving in the finance space okay so I'm less in the finance sector than in the generative AI space and basically of course I mean I can give you my experience where in the uee uh we built when I was Chief researcher at the Technology Innovation Institute a big large language model called f Falcon LM which is now basically one of of the biggest open source Foundation model these Foundation models are quite General and in general what you need to do in what we're seeing with the various I would say verticals is the need of course to build what we call on premise type of generative a meaning bringing basically the model within the company and then try basically to fine tune it for specific use cases related to that on those cases of course what we're seeing is of course the the difficulty in in getting basically reliable good answer so I can give you for example in the finance sector um the examples which are given I think are very illustrative of things that you can do but a lot of these things also could be done related to machine learning and techniques and they're being used already in the industry but when you start looking at what gener is doing is mixing numbers being able to calculate Mi some calculus at that stage we're still at that level so having basically gener VI that can uh beat mathematicians be able to do very good calculus is a bit complicated so what we're seeing is people wanting to use the result and then checking it and then adjusting the results because you still need whenever there's numbers going on these these things are still not very well in terms of behavior giving the results that that you need the second thing also of course is everything about you know the memory of the company and I think this is very important one one thing which is very important about gener AI is that of course it gathers all the knowhow of the company because you know there's a turnover in these in these in these compan compies finan is one where people leave and come and you'd like to have some information about what happened in the last three years related to that and who was dealing with that all that information can be gathered in that model so even if there's a turnover of employees all the knowhow and you can get information about what happened in the last 10 years on the financial status within the documents but also in terms of all the exchanges of information how current is it is it only like uh for example uh I think with chat gpts up until 2021 I forget the exact here but how how current can you uh yeah you can go further because what you do is that you take these models for example if you take Falcon then you bringing in your own company and then you will be starting to what we call find units based on the on the knowledge so you'll increase basically the kind of of knowledge and up toate so this is something that can be continuously trained within the company so there's no like question of of saying you're going to stop at a certain moment because once you start giving it the data within your company you can train it up to the last things and you can continuously train in this thing so that's fine and what's the adoption rate at the moment of you know such technology so it's just a start I have to admit it's just a start I think like what my colleague was saying I think in the next two years you'll see a lot of of big things of course a lot of people come to us and saying okay I want to predict the market this is not the kind the generative AI will not be able to predict the market so we're not going to tell you because things like the war in Ukraine is something that never happened and it's unexpected so if you're trying to use it for Investments this is not typically the thing however all the do doents all these cases are the the trend that we're seeing but it's just starting at the moment yeah exactly and this is the time like where you have to get the accuracy High Earth to be used in again I keep repeating it's such a sensitive space I think Casey you have at windfall an example of you know using uh generative AI like uh like being able to ask questions especially with financial language which is much more specific um you know than using uh chat GT can you talk a bit about you know how this has actually benefited you uh in your type of work yeah sure I think really Echo what what Ronnie just said um so so there's two big issues for for for General AI to be applied in financial sector number one is hallucination you really need to address those hallucination issues to order in order for for get to external usage right accurate information real time updated information that you cannot continue to train the model every week the cost doesn't allow that so we have been spending many times in this in this area and we think the only way to address this is to combine a financial large language model as a base with a huge database a information database like for example wind and other information database it can be your internal database as well but there is also complications with that you may have experience of chatting with GPT and asking a question it gives you the number immediately but that is only a friction of numbers and apis that we as Financial professionals dealing with every day the more powerful the information database is the more compli complicated it is to the right apis and functionalities for example way process PB level information Global Financial information Global macro industry company level information and the ability to interact allow the model to interact with informations to find the right thing is very critical the Second Challenge is to to deal with numerical tasks right unless the model can do the numerical tasks precisely um there is no way that the financial professionals will use it for these critical um transactional kind of kind kind of activities and the way that we solve that is it's not just through the Master model we believe there needs to be some expert model that working conjunctively with the master model so we leverage our past three decades of experience of dealing with data data analytics to train a specialized model this model has the capability of processing thousands of lines of excel functions and Excel informations with with many many different indicators pick the right needle for you from the oceans and and the accuracy rate as well as the ability to trst information is very important because if you look at the global leading general applications of that they may only get you 70% accuracy rate but that is far from enough for financial usage because Financial people has literally zero margin of error right so through that I just want I mean I think you covered it well and especially that it's at the stage it still needs human oversight to especially incorporate context Nuance things like that confirm numerical uh calculations and we have two minutes left I just wanted to put uh you know to the audience if you have any questions this is a very vast topic and you know start starting off the day but if you have you know any uh questions from what we've heard at the moment uh please raise your hand and we'll bring the mic to you okay great I do I see me hands up no great please continue then and um I wanted to ask you like say in the next I'll put this out you know to everyone in the next like five years how do you see um AI you know particular I mean because it seems like it's a uh a peak discussion around generative AI how do you see this evolving in the next five years do you see adoption do you see that this is something that's still ongoing the same way that crypto and regulation stuff like that is yeah where do you see this you know moving five years forward um I think I think it's a race between good use cases and bad use cases um and also the other side of the triangle is probably the regulator what's permitted because by Financial Services in most economies if not all economies are a highly regulated industry good use cases we've talked about mainly productivity Information Management retrieval analysis bad use cases is um we can just create loads of more content I mean and think about our lives 30 40 years ago I mean the amount of content we would consume we read a newspaper in the morning or turn on a TV and get a couple of channels now we're getting constantly spammed 247 if I can create content like this is that a good thing or a bad thing and am I creating quality content um Casey talked about the analysis of um an earnings report I'm I think many of us have seen Microsoft co-pilot in action you drop in the Nvidia earnings release that just came out that would take me as a young analyst half a day to read probably if not longer it can do it in a minute or two and they can create a PowerPoint presentation that would have taken me a whole day work 25 years ago that can be done in a couple of minutes is that a good thing or a bad thing seems like a good thing right but the bad thing is I can just Spam all of you with loads more content and so will we actually get a productivity boost um second thing is um def fakes and frauds and the question of trust um I could have you know if I'm on the trading floor or if I'm a CFO I could have my CEO call me video call me and say send the money now it's urgent I'm getting on my plane do it now when I land I want the money or you know you're a parent and you get called by your kid at school saying you know this is a very safe country I live in this country but you know many other parts of the world it's not so safe you get a call from your kid saying Mom Dad send the money now I've been kidnapped I mean you panic at that point right there's so many bad use cases so I know I shouldn't be saying this this early in the morning but uh there are good things we can do and bad things and it's still to be seen how much of this productivity boost will actually play out no definitely and I think this is what's expected it's it's always double- edged sword with technology right like you have new Solutions but at the same time you have new problems it reminds me of social media you know in terms of like mental health things like that you know we don't even know how to identify these things yet and that's what makes the job of regulators you know a bit more difficult and even central banks you need to be able to monitor the finance space so yeah I I'll leave you know final thoughts uh to you professor mwan and especially you know when speaking about Advanced Ai and you know as as R was talking about you know it's going to change the way that we operate in our day-to-day what are some things that you see as possible um you know the good and the bad in terms of things that we can't even fathom just yet but that we're going to see in our lives again good or bad yeah so so around the good things of course the first one is what everybody's working on which is the capability of the generative AI of having what we call reasoning and planning capabilities today we're not still there there but everybody's working on that and I think in the next five years we'll see some kind of emergence of capability of reasoning of these models you give an example of like what what it would Sol Yeah so basically you give them a task uh with a very like task which is at a high level description and it's able basically to give you that information and solve the problem like decision making decision making exactly so you're not just behind and I think this is one thing and and of course we're already seeing at the moment some of these models with additions of reinforcement learning techniques are trying already to achieve some results that we're seeing in the case of reasoning and planning so it plans things the second thing also which which is important in finance is the speed at which these systems are working so whenever you're so there's two part which is like the the training and I think once it's trained it's okay for you but then it's the inference part getting the results quite fast and getting the results quite fast is very important because at the moment it's a human which is behind a machine but after it's a machine trying to get the inference from that machine quite fast to do these transactions that we're talking about and the third thing that we thing which is very important which is going to come next year is what we call multi-agents meaning basically that well you'll specify to an agent what needs to be done and the agent is going to discuss with another agent which has it's a multigen Ai agine and they will do the transactions start talking and stop so it's like a chat GPT talking to another chat GPT today you're talking to the chat GPT but the future is about Intelligence being connected together and they're talking and if you have so many they can of course collectively do something and this is the next big thing next year A lot of people are going to be bringing more and more what we call AI agents which are generative AI based and I think this is going to be a big hit it begs the question as well I mean where do humans fit into that if they can you have discussions by themselves and make decisions and manage so yeah but I mean I I kind of know the answer to that I think we've all read about AI in terms of it's not going to replace humans if anything it's going to need humans to adopt different roles have different bodies you know in companies that are uh you know providing even human the human aspect you know to technology so again there's so much more that we can talk about with this but uh we are out of time and I just want to thank our guests for this insightful conversation and please enjoy the rest of your [Applause] day