In this episode, we dive deep into the current state of AI… getting to its transformative impact on the alternative investments landscape. Our guest, Mohammad Rasouli, is a renowned researcher at Stanford University and the founder and CEO of AIx2, a leading AI solutions provider for private equity, venture capital, and hedge funds.
Mohammad shares his extensive expertise on how AI is revolutionizing the way alternative investment firms operate, from streamlining due diligence and deal sourcing to enhancing portfolio monitoring and reporting. We delve into the key challenges and best practices in AI adoption, as well as the potential implications for the future of finance jobs and the broader economy.
Whether you’re a seasoned alternative investment professional or simply curious about the intersection of AI and finance, this episode packs an artificial punch. So get use to “sit back and relax” because the rapidly evolving landscape of intelligent investing will be doing more than ever – SEND IT!
_________________________________
_________________________________
Check out the complete Transcript from this week’s podcast below:
The State of AI, and AI in Alternative Investments with Mohammad Rasouli of AIx2
Mohammad Rasouli 00:06
The core business of investment is finding these opportunities, right? The rest of it is basically performance around it. The second thing is that if you look at AI and you match it, what it can do for investment, it’s perfectly matched to this use case asset, because what is ai ai is, is gathering a lot of information, finding a pattern, then running the pattern of success and finding that into another set of data, and then find this way, find a needle in the haystack.
Jeff Malec 00:35
Welcome to The Derivative by RCM Alternatives, Send it!
Mohammad Rasouli 00:38
Hi. My name is Mohammad Rasouli, and I’m a researcher at Stanford and CEO and founder of AIx2. I’m here to talk about AI in finance and what that might mean for you coming up here on The Derivative.
Jeff Malec 01:04
All right. Mohammed, how are you?
Mohammad Rasouli 01:07
Thanks, Jeff.
Jeff Malec 01:09
It’s two degrees here in Chicago. We’re recording on the 20th. It looks nice and sunny there, although I’m assuming that’s a virtual background.
Mohammad Rasouli 01:18
It is, but we have good weather here in Palo Alto Yeah,
Jeff Malec 01:21
it’s the same. What am I looking at at Stanford?
Mohammad Rasouli 01:25
It’s the Stanford campus, yes.
Jeff Malec 01:27
So at Michigan, was undergrad at Michigan, post grad at Stanford, or
Mohammad Rasouli 01:31
I was doing my PhD in Michigan, my PhD was dual degree between Electrical and Computer Engineering and economics. I worked with a thesis committee of Stanford, Harvard, MIT and University of Michigan, and including Nobel Prize winners. And my topic of research was how AI can help reduce friction in the marketplaces, including the investment market, where there’s information frictions and operational frictions, especially when it gets to private market, when these frictions are more significant. And the question was, how AI and algorithms can reduce these frictions result in better allocation of resources, more prosperity, including increasing the social welfare. And then I moved in Stanford and continued that research in management science and engineering, and I stayed as a visiting scholar researcher in Stanford, and I worked in McKinsey, as I said, in New York office, where I was managing a transformation for financial institutes, particularly private equity, and did due diligence projects with them. And I started AIX two here in Bay Area, which is AI for alternative investment. So hence AI multiplied by two. And the idea was that to use the real new power of AI to address the needs in the industry for investment, including finding good opportunities to invest on as well as having complete understanding of the deal due diligence all the way to the final reports, and reducing that friction in the market. That’s what we have been doing in AIX two with our set of clients. And there is so much happening with AI and so much happening with AI in this space.
Jeff Malec 03:22
I’ll go backwards for a minute. So you’re working for McKinsey in New York. What? What is that like? I don’t know how many former McKinsey people we’ve had on the pod. I asked all the Goldman Sachs alums if it really is the giant squid sucking the life out of the financial industry, which was Matt talie Taib coined that term. So McKinsey’s kind of got what would you say? Sorry if I’m bashing your former employer, but right, it’s got a bit of a negative connotation out there. What was it from the inside? Did you like your time? Now, I
Mohammad Rasouli 03:53
really enjoyed my time in McKinsey. I love that firm. Sometimes people ask me why I did it. Because I was in a data science background from an engineering perspective, and I worked in Microsoft, and I was set for being in the AI research labs. And I consciously decided to go to McKinsey because I wanted to have that exposure to that world and that experience, which was super interesting for me, for the short times I wanted to invest there, and I got to work on topics that I wanted. I worked in these topics across US, serving from the headquarter in New York, but also I went and served European clients and Middle Eastern clients as well. So for me, it was a perfect ride. It’s a huge, fairly diverse set of groups topics, and it really depends. The experience is really dependent on what you work on and what you do and what you like. It’s definitely a lot of fast paced working, a lot of work that should be done if you want to be successful there. And I appreciated that. But what I would say is that the set of people there are really interesting. So. Of people, and that’s what really matters. Come from very different backgrounds. I have, I don’t know any other firm. I have been in academia, I have been in tech industry. I have been in a startup scene. I have been in McKinsey, and I have been in finance. I think McKinsey uniquely. What they have is not only global presence, but also very diverse background of people coming from engineering, MBAs, law, military, art and everything, and they work together to solve interesting problems.
Jeff Malec 05:25
And how, how does that not just McKinsey, but any consulting firm, especially in AI, right? Hey, pay us couple million dollars and tell you what the AI landscape looks like six months later. It could be a totally different landscape, right? Like, how did, how did that work out? Do you think what you put forward in some of these reports was, how long was it valid for? Like, what’s the shelf life of that? So
Mohammad Rasouli 05:48
that’s a very good question. So the underlying assumption you rolled out here is very valid. Like, AI is a changing thing, right? And anyone who says, I can predict AI for the next five years should have access to a crystal ball of future prediction, which I don’t think anyone has. So it’s kind of unfair ask for anyone to predict for a long even like two three years ahead, what AI can look like, right and just look what happened to us in the last two years, like how many wrong bits existed among the experts in the field right now. This doesn’t mean that someone that wants to start now shouldn’t talk to experts, right? You want to take your bets, right? It’s a speculative to some extent, because we don’t know where the technology goes, and it can go all the way down to what is causing that black box approach to AI. But that’s the fundamentals that it’s to some extent, it doesn’t mean that you should not talk to experts. Let’s just we are in finance, right? There is a speculation that we are all used to probability and statistics. We talk to we understand the fundamentals and we talk to experts. Now, two things here. Number one, when you talk to experts, depending on the use case, where you are and what you want to do, that expert should have the capability to answer that and what I can say as a common factor for AI experts you go to no matter what in which industry you are to or what level, make sure they are number one expert in the engineering aspect of AI. Now you may want to talk to people who are also experts in finance domain or organizational transformation topics or other topics, or HR topics, but don’t miss out on the fundamental requirement of being expert in AI deeply at the engineering level. What do you mean
Jeff Malec 07:29
by that? On the engineering level, the actual chip set and all that?
Mohammad Rasouli 07:33
I mean the algorithms, the AI algorithms, right? That’s that’s the table stick like people who have written codes, done data science, understand neural networks, understand fundamentals of this technology, because that intuition is helping them predict what can happen. What are the use cases? What’s the value? What is the data required? What is What are the trends in the future? Now, no one can, as I said, predict perfectly, but that’s the table estate for being able to talk about AI as an expert, and I’m glad that the time I was back to your question in McKinsey, I was glad that on the project I worked on, which is AI for AI transformation for private equity funds, at least I came from PhD in engineering, so I had that aspect of that, and it showed that, showed It itself in how we designed the framework for AI transformation for private equity with an angle and understanding of the engineering part, and that results in more reliable and longer impact reports, as opposed to people I see, especially sometimes in consulting industry, who talk about AI and they have never really done anything like concrete that you are reading a code or done research, or others like so. But
Jeff Malec 08:49
I can see that both ways, right? I don’t need to understand, we could argue this, I don’t need to understand the self driving algorithm in Tesla to want to buy Tesla stock or something like that, right? So, or I don’t need to understand the scheduling assistant algorithm to use that tech. So it seems like that, right? It’s and I’ve struggled with this personally. Do I go down a deep hole with AI and learn all these tools and learn how to code some of them and do it, or just wait for Microsoft’s copilot or Google’s new tools, right? Do I just wait for that to integrate into my life seamlessly, instead of learning it myself so I can, I can see both sides of that, so I agree
Mohammad Rasouli 09:31
with you, but there’s two differences here. First, when I say someone should do, should know about the I mean, at the expert level you go to ask your questions, right? That Express No Ai, right? And second, what i ai use case I’m talking about is not about investment in AI companies and which one is going up or down. That’s some level of understanding. I’m talking about a deeper understanding on using AI for your organization, like building versus buying, what use cases are deliverable, but the keynotes I give. Conferences. I gave a keynote in finova fault, for example, or super return and other. One framework that they update every six months is, where are we in a scope of AI deliverables, how, which framework you can take and compare use cases against that? Is it going to be a good use case right now, at high quality, high accuracy, without a lot of human in the loop. Or no, it’s a use case that yet AI cannot address it at major performance without human in the loop, right? For example, the video analysis, the audio chat bots, the writing documents analysis, financials, inside extraction, these are different functions that we should understand where AI is and what use cases can do better job for each of them. If you want to go into that, that deeper question of, how should I use AI in my daily routine, especially in organizations? Should I invest in buying versus building? How much data costs will impose so that expert to respond to respond to those question requests to understand AI. But at different levels, you may probably not. You don’t need to really, like be all the way coding and stuff if you want to answer, like investment in store orders and
Jeff Malec 11:21
what, what what were the private equity firms? Or what are they doing now? Are they saying, I want to use AI to identify new investments. I want to use AI to streamline my paperwork and borrowing docs and whatnot, all the documentation, all the above, what? What were they and are they looking at to improve their operations,
Mohammad Rasouli 11:41
yes, yes. Let’s take a look back in the history for the last 10 years, and it’s a very interesting journey that finance, especially private investment, has done in this so the idea of using AI or algorithms for finance and trading is not new. It’s existed for the last 20 years, or even more, right? At least it
Jeff Malec 11:59
was trend following managed futures. Fans on this podcast know that well, right? They were doing it since the 80s. Yes,
Mohammad Rasouli 12:06
exactly. And as a job description, like a solid job role in Wall Street as a quantitative researcher, it’s been there for 20 plus years. And yeah, it’s a very mature field. My friends from PhD who go to Wall Street and work for hedge funds and others. So that trend existed. And I would say, like eight years ago, eight to 10 years ago, people in the private investment space like private equity, thought about, okay, maybe we can use AI in the same way that hedge funds use to predict Scott stocks. We can predict value of an asset to do an MNA or to do to invest on and
Jeff Malec 12:41
just back up. Sorry to interrupt you, when you I keep calling it private equity, you kind of refer to it as private investment. So you’re including, like, private equity VC, private bank investments, kind of all these different things that are outside the exchange traded markets,
Mohammad Rasouli 12:56
exactly, exactly that, even real estate and others, right? Yeah, okay. So the idea came that, okay, can we predict the value of an asset with AI across all of these different asset classes that you mentioned now? And there were some good results. There were a lot of research going on. There were some bigger firms who were using these things, and almost all the mega funds were using this before chat GPT was hot like EQT Mother Brain and KKR general partners, and all of them are using particularly for the single use case that I mentioned. And why that? Because if you look at that single case, it’s first of all, the bread and butter and core business to the to this, to these funds, the core business of investment is finding these opportunities right? The rest of it is basically performance around it. The second thing is that if you look at AI and you match it to what it can do for investment, it’s perfectly matched to this use case asset, because what is Ai, ai is gathering a lot of information, finding a pattern, then running the pattern of success and finding that into another set of data, and then this way, find a needle in the head stack. That’s exactly what an investor does. Investor looks at the history or whatever forms a thesis, forms a pattern. This is how I’m going to invest. Then he goes out, he collects a lot of data about different assets that exist. He pattern matches what he thought is important for investment or successful to those new data and says, Okay, these are the top five, top 10 I want to do the audience or invest on. And this is again, finding the need on the head stack for him. So if you look at these two AI, what AI does at the core and what an investor does at the core just perfectly match to each other, right? So that’s why it was very natural to have that use case of finding good opportunities as we make this would
Jeff Malec 14:42
you say we talk a lot with different managers on here? Of to me, that’s more. I’m replacing 10,000 workers that are scouring documents and looking for these opportunities, right? So it’s not necessarily I’m relying on the AI for its crazy insight that I wouldn’t have thought of. It’s. More of doing a lot of work in a short period of time, they will get there. We’ll get there. Okay,
Mohammad Rasouli 15:06
history we are now getting so it was up until chat GPT became a U turn in the industry. Right before chat GPT, it was mostly focused on finding opportunities. But once chat GPT and open AI commercialized the set of natural language processing algorithms at Mass and made it available to everyone at low cost. Suddenly there was a U turn, because people realized exactly what you said, Jeff, that hey, I can have this beautiful device, this beautiful technology, to process my documents, find out any issues in a contract, write a new investment memo, write a marketing email, write my LP query, monitor my performance of my portfolio companies to the documents they sent to me. A lot of things suddenly became available at a low cost that were not at the core. And nlps before chatgpt Were one of more complicated, like complex and expensive algorithms, right? So commercializing that was definitely U turn. So what happened after that was not only the big funds, big firms that used to have aI teams for using predictive AI for finding assets, now suddenly also tried to use generative AI for for this day to day work improvement, but also a lot of smaller funds who didn’t have the luxury of having AI teams and using AI found it accessible and affordable for them to use these tools for generative AI tools based on chat GPT, directly by chat GPT or derivatives of chat GPT to help them with this. Document writing and many use cases from that that I mentioned a few. It’s like due diligence, market research, portfolio monitoring, exits, preparation, document writing, LP, query, LP, reporting, investment memo, all of these things. And you can see it’s present across the investment cycle all the way from forming a thesis to executing that, existing that and reporting 12 is everything can be impacted with this. So you’re kind of separating
Jeff Malec 17:09
between predictive AI and generative AI. So was predictive, was kind of that, hey, I’m replacing a room of 200 people, and this is just going to do the work more quickly. Versus generative was I’m coming up with new ideas based on it. I would
Mohammad Rasouli 17:24
say it’s predictive and generative. The predictive one is more on alpha. Finding is finding opportunities, finding good assets to invest on predicting the performance of an asset, predicting a chance of fundraising from an LP, predicting the chance of hiring someone, finding someone who can join your portfolio company, finding something and predicting its performance, being an asset. An LP, an individual, is predictive, doing things day to day better, like writing reports, analyzing reports, analyzing unstructured data, reading, providing market research is generative AI is power of chat, GPT and generative AI. So it’s these two most backwards. It’s beautiful that these two categories of use cases match perfectly to these two technologies of AI.
Jeff Malec 18:14
And then I almost want to say before that, right? So you said in the last 10 years, 20 years ago, they were already using predictive analysis and computers to based on numbers, on just prices and data, right? And so the the leap was to use natural language processing, like, now we can. We’ve done it with numbers for years now. We’re going to use it, do it with words. Basically, yes, that’s an unstructured data,
Mohammad Rasouli 18:38
yes. And you know, like, it’s very like fundamental. This is an interesting observation. Now we are relying on a power of language, and language as a model of communication and everything we can do with language, but there is a lot of research going on that maybe the fundamental AGI we want to make, the great AI agent we want to make it should not be built on language. Maybe it should be built on audio or video sensories, right? You know, like 3d there’s a good amount of research going on in Stanford that they think, and there’s companies that started off of that, that they think they’re going to replace chat GPT, because they have agents, fundamental agents based on a different modality, which is more powerful and more complete for addressing like for example, language is limited for analytics and math and analysis, right? You can do function calling and this way and that way to add these functions to chat, GPT, but at the core, language models don’t have that. If you ask, what is two by two? It doesn’t do a math in the background by chat GPT, unless it does function calling it just looks in the in the history, like how many times it has hairs,
Jeff Malec 19:49
two by two equals four, right, right? And it might say some, yeah, and
Mohammad Rasouli 19:53
it might say something strange or hallucinate. So, yeah, observation, that’s real quick, because
Jeff Malec 19:59
you mentioned. Hallucinate. When I use some of the tools, I say, Don’t hallucinate. That doesn’t seem to work right. It’s still hallucinating. So it’ll be like, well, this the market was down because the Fed surprised cut rates or something. I’m like, No, they didn’t. What are you talking
Mohammad Rasouli 20:13
about? There’s a I encourage people to always get the source from the experts, as I said, that there’s a good amount of interviews, especially people in anthropic are very open and vocal about the type of trainings they do with LLM like, right from the top before they release cloud. And do watch their podcasts and others like, they talk about it actively, and they say these things, like, at the top level, they should really turn the knob on how the machine turns to be polite and accepting versus to turn to be rude. So it’s literally a knob that is turned on the language models, right? Think better and better on at
Jeff Malec 20:50
their high level, at that high level, and see, isn’t that that scares me, right? That they have knobs they’re controlling at the high level, which is by the time it gets down to the people using the tools which leads into the whole like, how do we control this stuff? Let’s get there later. But interesting, you said, so they’re going to use different modalities, visual, right? We can know, right? If I’m reading something, I’m reading a text versus I’m telling it to you over video, there’s a whole different interaction, right? You can see my facial expressions, you might think, Oh, he’s lying, or he’s he’s joking, right? Of like, oh, the text might miss the humor or something. So that’s interesting, that the video could, in theory, pick up on those video and audio, pick up on those intonations and and all of those non text clues, right?
Mohammad Rasouli 21:38
No, perfect. Exactly that, exactly that. And the language is a construct of many, many years of human development, right? We have come to this protocols of communication, language called language, and it has a lot of structure around it, and it allows for pattern matching, pattern recognition, that is the fundamental like, if you go back to the history of chat, GPT, how they started and what with what they did, like, open AI. I mean open AI, fundamentally. And do watch some of some open like earlier comments and podcast interviews that talks about earlier history, it was not clear the language is the model to focus on, necessarily, for the goal of making that super intelligence agent, AI agents, right? So they had to navigate a bit worker work around reinforcement learning was one idea. There were other ideas. And eventually the language picked up, because it’s this modality and this protocol that humans have worked many, many years on to to perfect that that’s that ways of communication, plus there’s a lot of data around it, like in all everything in Internet, and you can train machine on that massive amount of data. So the thesis was, if you have a lot of data, like more data, more compute, more bigger training, bigger deeper and deeper neural networks, bigger and bigger models, if you combine them together, you can crack the highly intelligent AI solution that was the fundamentals. Now, where to find it? It could be language, right? Because language has all of these availabilities, and it’s to some extent, the structure so perhaps easier to decode and predict. So
Jeff Malec 23:26
you’ve mentioned a few times this super agent. What are you calling it? The mother of all, AI agent. Is that? I mean, is that a goal, or is that a fear?
Mohammad Rasouli 23:35
That’s the topic of AGI. I’m sure that people,
Jeff Malec 23:40
yeah, maybe define that quickly for my friend George,
Mohammad Rasouli 23:42
yeah. So the general proposed AI, like, what we have it like, AGI is something that open AI actually talks about. And do we get to that or not? Like, there’s a lot of even like, debate around what is AGI in that such as or not, like, and I don’t want to get that, that debate like, we can always debate about, what is it? Is it like, clear milestone is like consciousness of the machine. Is it what it is? But without getting to all those extra conversation, the idea have is, is roughly, to have an AI that can do many, many tasks, right, that can be at almost at the level of advanced human right? That’s the idea of AGI, right?
Jeff Malec 24:23
How did you get AGI? Is artificial general into artificial intelligence, right? Why is the G in
Mohammad Rasouli 24:28
general? The general part is basically, like, the idea is that, right now the chat GPT is still not very general, like, it’s still a lot of limits in how to work with it and others. It’s like, not a human, advanced human that can do a lot of tasks, right? So getting to that general AI is the goal set for the AI community by open AI and other companies right now,
Jeff Malec 24:51
where one bot could be tell me how to do a million different things. Exactly, right. Look. Versus now it seems like, yeah, I need different bots that are kind of pre programmed with different parameters to do specific things. It limited
Mohammad Rasouli 25:06
the scope, like you can have an AI, very scope it and train it for just writing your investment memo, just crediting the price of a stock. Just do this, and you cannot give them high level tasks. Hey, should I invest in this company? Or no? Go find all the information, all the sources, write the documents, manage the process, right? The idea of getting to that high level AI like an advanced human who can do everything, rather than when I give it to finance people, I love to use this analogy, right? If you hire an analyst, right off the bat from undergrad, he’s not an a general proposed expert yet in your field, but stays in your field for 20 years and become like very advanced in thinking and has seen other patterns now with general purpose. You can think about investment. You can work with LPs. You can write documents. You can predict patterns. So that’s the transformation you are seeing in AR. If you, if you look at the letter that was released for all one GPT, oh, one. The main argument, the beautiful analogy and argument, was that the previous AI, previous chat gpts Were almost an undergrad who was able to collect data and read the o1, is a PhD. That’s the what’s the difference? PhD can analyze and think critical, critically and think creatively and right have a thesis and reason fundamentally reason about a complex task. Right now we have got to that level, and it was a different technology approach to previous chat GPT improvement, so reinforcement learning and a lot of chain of thoughts and others. So that is the progress, and that progress continues right now, being a more and more advanced person at the level of, let’s say, Nobel Prize winners for every field. Can we have an AI at that level who can think at that level and bring innovation, even to science and research and technology? You tell that AI agent, that Nobel Prize winner, AI agent, go find a new cure for, let’s say, cancer, and he designs experiments, and he designs everything, and he thinks about it, and he reads the results, and that’s the AGI. That’s what AGI means. But
Jeff Malec 27:10
how do they make that jump right? Because if you’re learning on the past, how can you create the future right? Like it seems, in fact, come back to just I’m training on a data set of prices and crude oil, and I create a model that’s going to predict crude oil prices, and then the next 10 years look nothing like the my data set that I trained on right. And I have issues that I didn’t expect whatever they found oil in Antarctica. And we have more than we ever knew we needed, whatever the case might be, right? I don’t have the full future picture. What’s the AI community’s thoughts on that? Yeah,
Mohammad Rasouli 27:46
that’s a very good question. The reflection of that for finance is exactly what you said, like events that are less frequent in the history. Let’s say if there’s a suddenly, a revolution somewhere in the world, right? Yeah, it’s not. And the impact of that in the stock market, right? Can the AI predict that and others? The short answer is exactly what you said. AI is the technology of predicting within the data, within the patterns it has seen. So it’s not for counterfactual analysis, right? Having said that, there are disciplines for counterfactual analysis and understanding, right, through utility based modeling, or through model based AI or others, that we can go out of the sample set and have a broader view of the world. And there are a lot of techniques in this, including reinforcement learning and synthetic data. The idea of reinforcement learning, like people probably have heard about like that broke the news through the AlphaGo, right? What was all the AlphaGo was, hey, we have an AI machine that is trained on play Go, and if there’s a lot of data, but the real breakthrough was when they said, You know what it did something different. Let’s, let’s let the machine play against itself. And every time he plays against itself, it finds a new world that have never had existed before, a new data, right? A new plane go, a new strategy, a new
Jeff Malec 29:10
position that’s creating synthetic data, that’s
Mohammad Rasouli 29:12
that’s that machine created synthetic data and reinforcement learning and feed it back, and this way, explored many, many wars that didn’t even exist and became so powerful. So I’m saying that, yes, your comment is correct, but AI community is also take trying different techniques that I mentioned to try to address that. And that’s an interesting observation, because there was this new reps conference this year. New York is the number one conference in AI, like, it’s a blood data to get through. Like, publish, I publish my papers that I know, like, it’s so, so competitive to get published papers there people don’t sleep for months to get their paper, like competitive and submit to that conference and get accepted and one more. Thank you. One main, one main conversation this year was that. Are we running out of data as the fuel for AI progress? Yeah. Look at the last 20 years on AI, the main driver of AI has been the single thesis of more data into bigger and bigger neural networks. That’s been the major prediction of AI, that’s been the major fuel for AI and major approach for AI development, like what is open AI, like massive amount of language. It on large, larger scale models for and just tuning and training. Now the conversation was, are we running out of data? Because the data in this world is limited, right? Have we trained on everything that we could find, and the Internet and Wikipedia and everything? Is there any legal, legal
Jeff Malec 30:39
and other ones, right? So, right, that’s another discussion, probably for another podcast that who owns that data to train on? Yeah? So yeah,
Mohammad Rasouli 30:47
that’s, that’s a good conversation. I tried to stay away from this one particularly but, but the conversation was that, are we out of data in this world, right? But then there’s this good arguments that that that? Yes, the AI community addresses this solution, this problem, and we have done it before, and there’s so many ways around this synthetic data, reinforcement learning, other types of AI solutions. So I don’t think AI is going to stop because lack of data, but I think it’s a major question for AI scientists to think about and part of that is what you said about counterfactual data and the words that we have not seen. Yeah,
Jeff Malec 31:25
that breaks my brain thinking about it. How can you come up? But how could anyone, how could a human come up with something new with only their past experiences too? So let’s talk a little bit about, what do you see in finance, in particular, alternative investments, private investments, in terms of, are we going to get rid of all the junior employees? Is there just one big machine to run everything, or is it just add on? It’s just going to help the tools? Like, is it going to affect the jobs in finance?
Mohammad Rasouli 32:01
So I think finance, in this question you posed, is there’s nothing like particular to finance that separates it from other broader question about like labor market and how AI will impact labor market. I do think that this technology is just transformative the way we work, right? And what it means is just just like previous technology shifts, and the technology shifts has been so fast in the recent years that all of us have lived through this technology shifts, if not multiple of them, right, from PCs like personal computers to cell phones to Internet being available at Broad broadband everywhere. All of these technology shifts, we have lived through them, right? We have kind of in our short life spans, we have seen that in a longer lifespan of humanity, we have the invention of fire, wheel, stimulating electricity. What
Jeff Malec 32:58
was that? Just Jimmy Carter at his friend. They were saying he was the first president born in a hospital. I was like, That was crazy. The first president born in a hospital was still alive here today, where we have generative AI. So
Mohammad Rasouli 33:09
I think that all of us have a good sense of this question by looking at our history of humanity and our direct experiences that we have had, and it’s just going to be transformative in the way that we work. I don’t think it’s gonna replace humans fully, but I think it’s a massive change the way humans work. I was giving a lecture last quarter in the Wharton and for their MBA class, and another one in Columbia Business School. And I gave a I taught those class, and I gave a lecture in the FinTech club in eastern as well. And I especially in those classes you talk to, like, analysts, associates and VPs of those funds who are now in the MBA programs, and you they all ask this question, like, what should I do to make sure I’m like, ready for this change, and this is gonna replace me? Am I gonna be, like, out of Java? Because if you’re more senior person, you expect AI to take time to get your role, you know, if you’re a senior, yeah, CIO, but, and the answer is that you definitely need a skill shifting skill. You definitely need to ramp up, and you definitely need to think about how you can work. And there will be eventually solutions for finance, right, be it AIX to be it others, but eventually we will figure out that solution, right? And you should be able to work with it, right? So
Jeff Malec 34:24
does that change your mindset? Of you don’t need to know how to do something. You need to know how to ask queries. Essentially, you didn’t
Mohammad Rasouli 34:31
know how to use the tool. It’s asking queries. It can be managing the work stream in the platform, right? It’s just imagine when Excel didn’t exist for finance, and now suddenly the Excel exists. Of course, no Excel, because in every single interview, right? Do you know Excel? That’s a question, right? Yeah. So that’s the same thing with AI. And understand the fundamentals of AI, because I think people should not be shy away from technicalities of AI. I don’t think it’s actually super complex. Data to learn those technicalities at the abstract level, like the conversation we had earlier, and it helps you to understand the buzzwords and and just especially younger generation they are in in a world that they’re going to see massive shifts in technology, not one like multiple of them. So better understand the fundamentals of this. Now back to your question. I don’t think it’s going to replace the humans. There is going to be new ways of work for humans. Like, yes, we have had all of the inventions in the history of technology, from fire and wheel all the way to cell phones and internet, but we are all busier and busier with more work, right? So, and it’s more creative work. Rather than going to the farms and like just working on the land these days, we can think about investments and opportunities and resource allocation and frictions in the market and how to use AI, right? So it’s there’s going to be more room for humanity to be more creative, solve bigger problems and more interesting problems,
Jeff Malec 35:52
and then at the firm level, I’ll ask this a little different way, what? What are you seeing? Well, I’ll do two things, what? What are you seeing? Is the adoption right in terms of across the let’s call it generally, the hedge fund space or the private market space, like, what is the adoption of AI? That’s a difficult question, because some might just be using small pieces of it. Some are using it for big things, or so I can you can answer that one. I’ll give you a choice or answer. What are most of them that are trying to use it getting wrong that one seems more fun. Okay, can I take both? Sure, take both. So
Mohammad Rasouli 36:30
these are intertwined questions, right? Like, yeah, responses, both of them. So the the finance industry, the finance industry is relatively doing good in AI, and I’m impressed every time I give these keynotes in the conferences or the articles are right, and the feedback I hear like the community is moving forward, like I remember a year ago when I was giving talks in Super return versus, like, the recent one that they gave the king of the first last year was all the fear of AI and hallucination and security and danger, and now it was all about, Hey, how can I scale my experiments? I like, like, what is next? Tell me next. How can I scale this? Like, beautiful things I did with AI.
Jeff Malec 37:21
You think that was all basically chat GPT, like, put it in the hands of millions of people, and it becomes less scary.
Mohammad Rasouli 37:27
I think multiple things happen. First of all, chat GPT dramatically, like, significantly improved in its quality of work. Like, it was really nice to see that the fundamental the operating system providers, which are chatgpt, anthropic, Mistral, long one others are doing great job in addressing some shortcomings, especially the reasoning, especially the multimodal reading, especially the control of hallucination and alignment. And there’s a way to go yet. So it passed the bar for adoption in the industry, but also it was a lot of work by not only chat GPT, but the derivative companies that use that operating system for providing a solution, right? What does it mean? Like chat GPT is the operating system is, if you want to make an analogy, like it’s like a translator, it’s like a dictionary for a translation and translator. So it’s not a translator itself, right? Chat, GPT, if you want to translate V dictionary, it takes it. You can do it, but it’s painful and takes a lot of time. But if someone takes a dictionary as an operating system and makes a translator on top of that, suddenly there is a solution that can really work for translation, and the adoption goes high, right? So for the finance industry, I think multiple things is happening. Number one, finance is at the focus for AI, innovators, researchers, entrepreneurs, because finance has a lot of data, structured data, unstructured data, use cases, smart people, so it’s naturally a place for a lot of AI researchers, practitioners, entrepreneurs, to start from, right? And you see that Microsoft copilot has Microsoft copilot finance. That’s the only like domain specific copilot that Microsoft released. Yeah, right. The other thing is that the finance people have the idea of roughly the idea of using AI and data and getting this so they were waiting for it somehow. So I wouldn’t say finance is necessarily the best and former foremost adopter of technology generally, but I think in AI space, they have done a good job, and it’s going to continue like the firms are experimenting. I don’t think there’s a consensus in the industry, and what are the main use cases, and what are the main solutions, and what are the best practices? Right? Like, if you go to this industry and ask about data analytics, Excel is the word you hear, right? If you ask about I don’t know, like if you go out and ask about sales, maybe Salesforce is the solution you hear, right? So it’s not yet there with AI. The game is not settled, and it’s everyone is exploring. Now, what it means for practitioners is that if you want to start with AI, you have to do a little of homework for yourself, because there’s not one solution for all. Yet you have to do a little homework. Understand the use case. Understand what is good for you. Decide your own way of working on that. Provide a solution you want, build it if you want, like going through that experiment, that’s why, back to your question about consultancy, like when I say, work with experts, we can help you get there at this point that expertise is not only on AI engineering and research, but also in transformation of organizations with AI, and have seen the patterns that gets your second question, what has gone wrong for the many other folks in the industry, similar CEOs and CIOs and CFOs and CEOs in hedge funds and private equity? What did go wrong for them? Like, what can I learn from their tracks? And there’s a lot of things that are not necessarily the core engineering of AI like how to define use case, how to design the solution, how to roll it out, how to manage the expectation, how to manage the cultural change and behavioral change, how to scale it up. So many times people come to me and CEOs, CEOs, and they say, Hey, we bought this solution and that solution and we paid and no one uses what’s wrong, right? Yeah,
Jeff Malec 41:19
Salesforce, I joke all times, the most successful, hated company of all time, right? Nobody wants to or likes to use it. These are the
Mohammad Rasouli 41:27
type of questions that are maybe at the fundamental there they should understand the technology of AI. But there are also other aspects of that, that the behavioral change, the cultural change, the organization change, and that’s the type of expertise required for successful AI. And I would say that we see that like, day in, day out, and we work at AIX two does, like, one thing you’re proud of, like this. We have this logo in our website that one week to impact. And if you think, what happens in one week to impact, what is it? I mean, it’s a clip of subscription to a software. What about one week? Like, it’s one second to subscribe to the software. Why do I need one week that? One week is exactly designed to cross that barrier, like from subscription to a software to significant scaled impact in your organization, and how do it? I would say there’s a lot of nuance that has gone through, like designing that, a lot of experiment, observation, systematic surveys of the industry, working with our clients that get us to that. Now that’s the state of the industry. Maybe in five years, it’s so streamlined, so like, public knowledge, common information, but you don’t need that. Like, it’s so clear, oh, you’re gonna say Salesforce, let’s say for sales. And this is how it works. And there are so many online YouTube videos about how to use Salesforce. They don’t but at this point, it’s not yet there for AI, but the impact and adoption is just growing. Every conference I go to I see more and more. No,
Jeff Malec 42:41
it seems real quick. It seems there’s a bit of a split, right? And no, it was all these quants, as we call them. Like you said, quantitative finance guy has been around forever. They’re using AI to generate new models and do testing and all that stuff. But they’ve kind of been doing that for a while. This, I think, just speeds that up, maybe for idea generation. I don’t hear a ton about that, of like, for new trade ideas or some models, but then it seems like you’re everything. You’re just saying, hey, there’s the whole business side that can be vastly improved. Which hedge fund people generally hate the business side, right? They just be like, Hey, I know how to make money. I want to make money for people. Oh, I have to deal with clients. I have to do paperwork. I hate all that side of it. So, yeah, AI is a huge, huge unlock for that side of the business of, Hey, you don’t have to worry about that as much as perhaps you used to
Mohammad Rasouli 43:28
exactly Jeff, exactly like you said, and exactly like you said. There’s this alpha generating side, or alpha predicting side, and then there’s this operational efficiency or business side, if you want to call it. And AI has just opened its impact and its way to the into the operational efficiency of the funds, hedge funds, private equities, VCs and others. And it requires a lot of these internal processes, document writing, and a lot of hedge funds come and talk to us about the same sets of operational efficiency, how they do it with generative AI. And to your point about the traditional, like, more, older use case of like, predictive AI for alpha, alpha prediction and alpha finding even that, like, people come and say, this, trading this, we want to have summary of all this, some all the, all the data online, like, all the sentiment analysis, for example, reviews, all the alternative data, all that reviews in social media, all the news reports about an asset, so you can fit into our alpha finding models. So it’s also getting there, like thesis generation, I would say less is very high level question to ask an AI to provide a thesis for you at this point. But if you break it down to a smaller task, AI can solve, but for data feed in to those Alpha finding models, that’s been something that because the power of this, there’s one power of generative chat GPT that people for. It’s not only a chat spot. There’s a huge. Powerful machine for turning unstructured data into structured data, something that would take huge, significant manual efforts,
Jeff Malec 45:09
yeah, which I always said, like, Okay, you’re using AI to generate models and do, like, why isn’t it told you to buy Vancouver real estate and whatnot, right? So there’s little things like that, of like, Okay, but why isn’t it grabbing any data it can and saying, Hey, here’s, here’s something that fits with what you’re trying to do with your model. Here’s a trade I did, exactly.
Mohammad Rasouli 45:28
So, yes, so nicely, not you put it nicely that there are two domains, business and alpha generation, and each of them can be impacted by generative AI and chat GPT. There’s a obviously, on the operations side, there’s the whole new domain and whole new way of thinking, but even in the Alpha finding, there’s a lot of use cases that people come and talk to us. And now, what has gone wrong? The other question that you have, I chose both of them, so I have the responsibility respond now a lot of time that what goes wrong is first choosing wrong use case. People have wrong expectation of what AI can do. They think AI can come through and find a risk analysis for all the risk in their in their assets. That’s a high level of thinking that AI cannot do yet, right? Second,
Jeff Malec 46:26
if you ask it like, what am I missing from a risk standpoint, yes, that’s
Mohammad Rasouli 46:29
like, a huge question. Yeah, I don’t know. Break it down. Break down the risk into the 100 elements even, like, for them, get clear inputs, clear prompts, and ask the machine to synthesize right? Unfortunately, the machine is not able to do that yet. Maybe in the future, we have this AGIS or advanced AI agents that we discuss, and they can do the high level thinking, right?
Jeff Malec 46:50
Does it ever drive you crazy, like, the completely unrealistic expectations of, like, why can’t it do this? Yeah, what’s wrong with this stuff you’re like, do you understand how much it’s already doing, yeah, it’s
Mohammad Rasouli 47:01
not only at the level of like finance. I see some startups in Bay area that is formed by people come from business background, rather than they have this idea that AI is going to solve the risk prediction. For this simple example, I gave risk prediction for investments, and let’s make any startup around it, and they hire some data scientists, and they want them to build that machine for them, because the founder was businesses. He doesn’t even understand this is not yet there. You can invest the resource and everything, but it’s not yet ready for that. You should start from another use case, and eventually is going to be able to address that. So these are some examples and wrong expectations, and that goes back to not understanding the fundamentals of technology and where it is, but also the other domain that goes wrong is in adoption and skill in the organization, which is a lot cultural and behavioral, right, many things should come together to inform organization to have a behavioral change and start using a new platform, right? And part of that is the communication lines, description of the roles, addressing the concerns of the employee, for replacement and others, proper evangelists and advocates in the organization. There’s so many ways in AI in organizational transformation, McKinsey, we had a playbook for these kind of things and all the techniques. So some of them are really like working for AI, is it top down, bottom up, how to design it like that’s something that I see goes wrong a lot of time. That’s not because the technology is there. It’s because it’s not the organization has not defined a good path for making this change in the organization, right,
Jeff Malec 48:38
right? And it could be a significant change, right? It is a significant
Mohammad Rasouli 48:41
change, like and it’s kind of a thing that it requires some level of learning. It’s not like happening in moment. I always give you this example that people say, Hey, I’m going to wait for AI to get better. Then I start, I’ll say, hey, AI is not like a Tesla machine that you don’t buy now you buy in a year, and suddenly you have Tesla. No, you start with AI, and there’s a learning curve and experience in especially with organizations to learn and build that culture and and it takes some time for people to get there, right. So the sooner you start with the solutions, the sooner you are on board and you learn, and people get the ideas of AI, and then you can go with this journey with the organization successfully start. So this is the other area that I see goes around around. So that’s why I think, for a finance specific the type of people who work on this should be a mix of like people who are really deep in engineering and research, not only engineering research, because AI is changing quickly, and researcher has the capability to see beyond just the current engineering and APIs and solutions to where the research is going, plus someone who understands the fundamentals of like investment organizations and this behavior of people there and how technology can come through a type of consultancies that we talked about, like a mix of these two is, I think, is minimum required for a. People in the finance industry, being hedge funds, public or private, private equity VCs and others to use AI effectively.
Jeff Malec 50:12
Close it out with what could go wrong. What’s going to stop this freight train from right? I was talking with a guy who sold his AI startup. He thinks, if any like fortune 500 company comes out and says, Hey, we’re pulling back on AI investment, like Nvidia stock will go down 50% so not to have you make market predictions here, but just overall in the whole ecosystem, what can you see that slows this down? If anything.
Mohammad Rasouli 50:41
So I’m not going to comment on if AI is hype or not, and what happens if, like, a fortune 500 like, yeah, it’s a good conversation. Like, where is AI, how much it has delivered, how much revenue it has generated versus investment. And if it’s hype or not, you can go to that conversation. What
Jeff Malec 51:04
seems now, if you’re a CEO, you you like Play Books, obvious. You need to say you’re investing in AI. You need to be doing things to stay relevant, whether it’s actually adding anything to the bottom line seems secondary. Um,
Mohammad Rasouli 51:18
and the question comes down to how much AI has really delivered versus how much is expectations for the future. So we are bidding in the future versus right now like and there’s a good amount of work that should go with real providing value from the operating systems of AI. I think that example, I said, from dictionary translator, there’s a lot of work should happen to get to the translator level. But I take your question from the angle of, what can stop this train from moving on? Right? What can stop it? I think one clear AB test and experiment we have is the Europe versus us. Right? In US you see AI is rapidly growing because of supportive regulations, right? Regulations really matter, like the regulations in the I can say now, we can say last administrative, yeah, because it’s, it’s happening now. So the regulations were limiting, and to some extent, even in Bay area here for startups and people in the scene that honestly don’t care much about politics. That was probably the first time I saw these people really caring about politics, because the regulations was really impacting their startup life. What kind of data you can use, what kind of AI you can use, and you can see that AB tests in a bigger perspective between Europe and US, and what’s the growth of AI innovation between the both and in the last 10 years, and what’s the gap now. So it’s, I think, an interesting observation the upcoming the current administration, I would say, it’s fair to say they seem to be very pro AI and reducing regulations and providing support for this kind of technology shift. So it’s going to go even faster. The competition is another angle. I hope the competition exists. I’m very happy to see that it’s not only chat, GPT anymore. It’s not only open AI anymore, this anthropic miss, troll x ai and others. And I hope the competition keeps going, which is a driver of success and innovation and better outcome. I guess. What can the other things that can stop it is compute and data, because these two are like fuel for growth, as
Jeff Malec 53:27
I said, and compute tied in with energy usage, also perhaps
Mohammad Rasouli 53:31
energy as well as like the real compute, like GPUs that we need and what Nvidia provides for The world, right? I wouldn’t say energy is a major bottleneck now, but compute, like the demand for Nvidia, and if it can get to all the researchers in the field, and everyone can use it, and the price comes down enough, I think generally, all in all, we are there’s not a signal of, like, a major slowdown or bottle road block here, not for data, not for compute, not for energy, but these are areas that we should actively work on as we grow to ensure that we can, we can see the impact of this technology into day to day life. And you’re
Jeff Malec 54:12
supposed to say like, as soon as the robots take over, and there’s a world war three with the AI versus the humans,
Mohammad Rasouli 54:21
that’s a good thought. I remember back in my PhD, is like four things happening, and we were working on AI. And even then there was this ideas of AI, what’s the view? I was working on a grant, NSF grant, and security, cyber security, and AI. And there was this question of AI is going to take over and the robots are going to come and extend the humans. I think these are, at this point, very much sci fi,
Jeff Malec 54:45
right? Because what would that even look like? Who knows, with that’s a whole nother podcast as well. But so you’re fading that that’s not a real threat,
Mohammad Rasouli 54:52
at least not the closest threat that you’re facing. Yeah,
Jeff Malec 54:57
it seems to me the bigger problem you have. You. The AI is flying the planes and driving the cars, and there’s mass like problem that something happens, there’s bunch of people die or something, and it gets
Mohammad Rasouli 55:08
hacking AI and AI, not alignment of AI is super important, because if you cannot predict some of the behavior of an AI machine and suddenly does things that is not expected to do, and examples, especially is like high stake control, like driving or flying or something, but
Jeff Malec 55:23
from a public right, if, like, okay, like 2000 people died when these planes collided or something like, and then some congress people come out and say, hey, put the brakes on.
Mohammad Rasouli 55:35
Yeah. But I think AI community, all in all, has taken a very more conscious approach to the topic of alignment is probably the biggest, hottest topic of research here in Stanford, and AI alignment and safety, I think, has been taken more seriously,
Jeff Malec 55:52
especially in alignment. Is alignment with us humans, aligning
Mohammad Rasouli 55:55
the tasks that you do, like if you’re supposed to drive a car and not collapse with others, like just not do that, right? Think that. So I think especially with the anthropic push and the nice competition we see on the field, there’s more and more importance for AI alignments with open source AI, there’s access to the models by everyone, like the law, mall models and others. So it helps the community to work on that more actively. I think they’re all good pulses that we see in the community. So I’m relatively optimistic about where AI will go.
Jeff Malec 56:33
Awesome. I love it. I think we’ll leave it there. We’ll put the we’ll link to your website in the show notes, and people can go find out more, any last thoughts before we wrap it up.
Mohammad Rasouli 56:46
I’m available. Also in LinkedIn, people can reach out. I’m always happy to talk about, talk with people who are excited about AI, research, engineering, finance and the mix of them, or creative ideas in this space. And I would say that we are definitely going to see this change. Finance is the beautiful area that has the forefront of AI focus, a lot of innovation in research and in business and building solutions. Happening here is beautiful field. There’s a lot of interesting problems for everyone, from a research standpoint to engineering to business. So if you’re excited about them. I’m very happy to work with people and talk to people in the same area, and thanks for having me on this on this podcast
This transcript was compiled automatically via Otter.AI and as such may include typos and errors the artificial intelligence did not pick up correctly.
