As you likely already know, the stock market has a lot of algorithmic trading.
Complex mathematical models conduct millions of transactions, which is moving billions of dollars in and out of stocks every day. JPMorgan Chase recently estimated that computerized trading makes up 90% of the market's total trading volume.
Some fundamental analysts dismissively refer to these algorithms as "dumb money." The algorithms are based on numerical rules, which often miss out on qualitative aspects such as competitive advantages or visionary leadership teams.
But Sentient Technologies is taking a much more deliberate approach to algorithmic trading. Their evolutionary, artificially-intelligent platform monitors and learns from the trading taking place all across the globe, which gives them a much more sophisticated solution than the simple quantitative screens used by other trading desks.
In this video, Sentient's founder and CEO Babak Hodjat explains how his company's solution is adaptive to the world around it and is now being used to run an AI-based hedge fund. He also describes how the company's AI technology helps e-commerce companies optimize their websites to better appeal to users and improve conversion rates.
A full transcript follows the video.
10 stocks we like better than Walmart
When investing geniuses David and Tom Gardner have a stock tip, it can pay to listen. After all, the newsletter they have run for over a decade, the Motley Fool Stock Advisor, has tripled the market.*
David and Tom just revealed what they believe are the ten best stocks for investors to buy right now… and Walmart wasn't one of them! That's right -- they think these 10 stocks are even better buys.
Click here to learn about these picks!
*Stock Advisor returns as of May 8, 2018
The author(s) may have a position in any stocks mentioned.
Simon Erickson: I'm Motley Fool Explorer Lead Advisor Simon Erickson. You've heard a lot of buzzwords about artificial intelligence. This morning, we're going to be talking to somebody who's actually been developing it behind the scenes for the past decade.
My guest is Babak Hodjat, who is the CEO, co-founder, and Chief Science & Technology Officer of Sentient Technologies, whose goal is to commercialize a lot of artificial intelligence applications.
Babak, thanks so much for joining me this morning.
Babak Hodjat: My pleasure. Thank you for having me.
Erickson: Babak, you have a distinguished background already in AI. I know that your team was working a lot of the technology that went into the natural language interface that went into Siri for Apple. Just to start us off, can you talk a little bit about that research, how that actually got picked up into Siri, and how it influenced Sentient?
Hodjat: Yes. The story is I was working on distributed AI for my PhD back in Kyushu University in '96, '97. It was agent-oriented technologies. A friend of mine misunderstood what was meant by agents. He thought an agent is a representative of you to do something. He took that definition, and he thought that it'd be great to build a company that understands you and on your behalf, for example, changes the TV channel or operates your home entertainment system.
I explained to him that natural language is very, very hard, and he challenged me. I really thought about it long and hard and came up with a distributed AI way of processing natural language, which led into my first start-up Dejima, where we created a conversational system that was, what is called "active ontology". You actually create an ontology of domain of discourse, typically operating command and control. You lay a thin layer of language on top of that. It's language-independent, it's extensible, and so forth.
We were one of, I think, two or three commercial entities involved in the CALO project, which was a DARPA-funded five-year project to build an assistant, and we were the natural language interface to that. It was a project that was coordinated by SRI. Our VP of engineering, Adam Cheyer, a very, very good friend of mine, moved from Dejima to SRI to become the coordinator.
Then, at the end of the five years, when I had by then left doing other things and starting this current start-up, he called me and said, "I want to carry the torch and continue", and he started Siri. I wasn't officially a part of Siri, but the team moved on and built Siri, used the natural language technology, but of course, added a lot of very interesting features there and took attack. By then, with this feature, it should be where it was worked quite well as required by Apple.
Then, they left to do Viv, which I think is the next iteration of breakthroughs for conversational systems. They had a very, very strong vision. They got acquired by Samsung. I've not been doing conversational systems for about 10 years or so now, but that's where it all started.
Erickson: Then, you went on to found Sentient, which has been using a distributed AI, evolutionary AI as you describe it, for a couple of applications. You're using it for stock market trading and also for e-commerce. Can you talk a little bit about how that's being used right now?
Hodjat: Yeah, so evolutionary computation is inspired by natural evolution and survival of the fittest. It's naturally parallelizable. We call it embarrassingly parallelizable and at many different levels and makes for a very robust, distributed AI system that can become more powerful the more processes and capacity you throw at it, beyond a single data center. It doesn't have to be fully connected to compute. No, it's working together. It can be geographically dispersed, vocationally available, will all lend themselves to the power of evolutionary computation. That's what we liked about that particular technology. That paired with different representations, such as neural networks and deep learning, adds qualities to AI beyond just modeling robustly the world. It allows AI to become much more adaptive. As the problem statement changes, the system adapts to that. It also allows us to go beyond just modeling the world. It allows us to actually create new solutions. That's a property of evolutionary computation that is used in our core platform for these types of applications.
We started off with building an AI-based hedge fund, essentially. Coming up with trading strategies and not us coming up with trading strategies, allowing the AI actually to automatically offer the trading strategies by virtue of evolutionary computation. That hedge fund is now Sentient Investment Management. We spun that off and it's doing quite well.
Since that spin off, we've focused on digital marketing as another area of focus, where we believe being adaptive and creative and using these robust models allow us to disrupt digital marketing. We've had some very good traction there with our Ascend product.
Erickson: Okay. It's learning from the stock market, building strategies around what it sees. Fundamental analysis? Technical analysis? Depending on who's using it, I suppose?
Hodjat: Well, this is used by us. We're not providing a service or signals. We're actually using it in our own trading desks. Yeah, it's mainly technical but also fundamental.
Erickson: And e-commerce? Hoping to improve the conversion sales cycle for people coming to websites? Is it you're optimizing so that people will convert at a higher percentage?
Hodjat: That's right. It has all the properties that are a good match for our technology. It's very well-measured. You mentioned conversion rates. Average order value, dollars spent, hitting the thank you page, sign-ups. There are a lot of ways that we measure the performance of our website or mobile experience, and that measurement is online. We can get it at any point in time you ask an owner of a mobile app how well is their system doing. They can give you that measure.
That allows us to optimize around that measure, adaptive around that measure. Then, you have a large user base typically using these systems, which gives us the ability not just to optimize in an online fashion that gets to the user base but also to be able to segment in an online fashion that user base. Traditionally, the way that user segmentation happens is you take a lot of historical data, and because these systems need a lot of historical data, it usually ranges all the way back to irrelevence. You create these models.
Then, the model remains kind of static. The model gives you what the different segments of users are. Using evolutionary computation, evolving a mapping between what we know about the users and what we should show the user at any given step in time, allows us to optimize against the KPI of the business and have an online, and adaptive, segmentation of the users. That's the point of disruption that we're bringing to the table.