AI Is Changing Financial Markets – Hear From This Expert That Has Been In The Field Since The 90s
Kaiju ETF Advisors builds, trains, and employs robust artificial intelligence (AI) and machine learning technologies designed to improve fund management decision-making. By empowering these innovative technologies to curate and provide direct management of our ETF, we’re striving to go places no Registered Investment Advisor has gone before.
Transcript
00:00 (upbeat music)
00:02 - Welcome, it's Michael Murray with Benzinga.
00:04 I am joined here today by Nicholas Subryan,
00:06 Director of Artificial Intelligence at Kaiju.
00:09 Nicholas, it's an absolute pleasure to have you with us.
00:11 How are you?
00:12 - Hi, Michael, I'm great.
00:14 How are you?
00:15 - Doing great, Nicholas.
00:16 It's great to have you here.
00:17 Always happy to be talking with the Kaiju team,
00:20 and we're very excited to hear from you today.
00:22 Nicholas, you've been involved in AI since the mid '90s.
00:25 That's a long time.
00:26 Can you talk to us a bit about your involvement
00:28 in the AI space and give us a sense
00:30 of the history in the field?
00:32 - Sure, it'd be my pleasure.
00:34 So recently, as we all know,
00:37 AI seems to be in the news nearly every day,
00:40 maybe for the past year or so.
00:42 And I think a lot of people have had a chance
00:44 to hear about or maybe experiment
00:48 with some of the new large language models
00:50 that have emerged recently,
00:52 and people are seeing the opportunities
00:53 for maybe efficiency, productivity, using this technology.
00:58 And if you weren't following the field,
01:01 it may have seemed that this technology
01:03 just exploded from nowhere, right?
01:06 But in fact, AI is not new,
01:08 as you alluded to in your question.
01:10 AI goes back to the mid '50s
01:14 and has been progressing since then.
01:17 We've had periods of significant enhancement
01:21 and capability over that time.
01:23 And I think recently in the last couple of years,
01:26 we've had another period of significant capability building
01:30 with the large language models.
01:32 When I was in research in AI,
01:36 everyone was researching neural nets with machine learning.
01:41 However, although mathematically sound
01:44 and conceptually interesting and solid,
01:49 they never really showed any progress
01:52 in real world settings in industry.
01:55 And the reason for that was
01:57 they were missing some dependencies.
01:59 Essentially, they needed a lot more data to train on
02:04 and a lot more computational power.
02:06 And those things weren't available
02:08 until after the internet came into being around 2000 and after.
02:13 So when I moved into industry in the mid '90s,
02:19 the systems in AI that were used
02:23 and were extremely successful were called expert systems.
02:28 And these systems are still around today
02:30 and they have been wildly successful
02:34 in specific applications.
02:37 These systems were, in a sense,
02:41 what we did is we took experts,
02:44 we codified their knowledge,
02:46 the manuals and the regulations,
02:49 and built systems that reacted to
02:51 and made the same decisions as a human would do.
02:55 Think autopilot as an example.
02:58 Still in use today and extremely successful.
03:00 After the 2000s, what we saw as a shift to modern AI,
03:06 let's say, machine learning based on neural networks,
03:10 loosely speaking, based on the human brain,
03:13 neurons with connections in between them.
03:16 And that's where we've seen
03:18 the vast amount of economic benefit
03:21 that's been derived from AI.
03:24 It's changed industries, advertising, social media, news,
03:29 the way that we distribute content.
03:31 And that's the technology that we have been developing
03:37 and that's been in the background
03:38 in many of our applications and in the home
03:40 and in businesses for the last 20 years.
03:42 So smart assistants or news feeds and social media feeds.
03:47 So although it may seem to have just sort of
03:49 emerged from nowhere,
03:50 it's been in development for a long time
03:52 and it's been in the background of our lives.
03:54 - Amazing.
03:55 So, Nicolas, that brings us right up to date
03:57 to modern machine learning.
03:59 Can you tell us what it is that makes ML so powerful?
04:03 - Sure.
04:04 There's a lot of depth and nuance here
04:06 and we could spend a lot of time talking about that,
04:09 but I think that there are two things.
04:12 One, data crunching capability,
04:15 and two, pattern recognition.
04:17 So if I just break that down, let's see.
04:20 The process of machine learning on neural networks
04:26 is able to very efficiently take in data
04:31 and compress it in some sense
04:34 into some sort of knowledge representation
04:36 that's order of magnitude smaller than the initial data set.
04:41 So it's very good at compressing data
04:45 into a very condensed form of knowledge.
04:48 And during that process,
04:50 what the machine learning algorithms are able to do
04:54 is find patterns in that data,
04:56 either with guidance from humans or without guidance.
05:01 And what that does is when presented with new data,
05:06 new samples of data,
05:08 are able to answer questions on that.
05:11 And that's the basis for all of the modern applications
05:16 of AI and all of the economic benefits that we have seen.
05:22 - Absolutely.
05:23 Now, shifting gears for a little bit,
05:25 how do you see this technology playing out
05:27 in the investment space specifically?
05:29 - Sure.
05:31 So once again, if you think about the paradigm of data,
05:35 I think we've all heard things like,
05:38 of all of the data that humans have ever created,
05:40 half of it has been created
05:43 in the last 18 months or something.
05:44 There's just more and more data available.
05:47 Human investors have always used information
05:50 as a tool to make good investment decisions.
05:53 If you had access to information before everyone else
05:56 or early, that's how good investments decisions were made.
06:01 And with the advent of more and more data
06:06 being easily available to investors,
06:08 it's difficult for humans to consume that
06:13 and make decisions at scale.
06:16 And that's where artificial intelligence
06:18 and computers can do that.
06:20 So you may have heard stories of investors
06:24 using satellite imagery or imagery of warehousing
06:29 and ports to make decisions on investments.
06:35 You may have heard of investors using sentiment analysis
06:40 coming off of the internet
06:41 about new products in the market.
06:44 These are all examples of new data
06:47 being available for investors,
06:48 but to consume it and make sense of it
06:50 is extremely difficult for humans at scale,
06:54 easy for AI to do that.
06:56 More specifically on financial data,
06:58 whether it's fundamental analysis from reports
07:03 of companies, for instance,
07:05 the humans can use AI to condense that information quickly
07:10 and use that in their normal process of investments.
07:19 So as a step in crunching that data,
07:23 or in terms of technical analysis,
07:27 investors or traders have always used analysis
07:30 on market data, for instance,
07:34 let's say momentum trading or moving average,
07:38 but where a human may be able to do a handful
07:40 of those analysis in a day,
07:44 AI is able to do thousands at scale.
07:47 So I think that's where AI is showing its real benefit
07:52 in scaling up investment decisions quickly.
07:56 And I think that's what we're seeing now.
07:59 And I think that's gonna be part of the future.
08:03 - Understood.
08:04 Now, Nicolas, as we wind up here today,
08:06 AI is of course moving very, very quickly.
08:08 How do you think about retaining control
08:10 and being responsible with this technology?
08:13 - Sure.
08:13 So I think in some sense, AI has its nuances.
08:17 It's an emerging technology,
08:19 but with all the new technologies,
08:21 we've had to ask this question and answer it.
08:23 And AI is no different from that perspective.
08:28 In terms of retaining control,
08:30 I think economic imperative is one, right?
08:34 When people invest money to build these AI systems,
08:38 it's expensive, it's time, it's computational power,
08:41 and that's all really costly.
08:43 And so most people do this for economic return.
08:48 And so they want good products that people want.
08:50 If you were looking at your social media feed
08:54 or your newsfeed, and suddenly you've got some sort
08:57 of random illicit or illegal information or content,
09:02 you would be like, well, I'm not going back to that source,
09:06 and then they lose their ad revenue, right?
09:09 So I think economic imperative builds systems
09:12 that do something good that customers want.
09:16 And I think that's probably the biggest imperative
09:19 or the biggest sort of driver here.
09:22 I think in future, legislation will matter
09:26 and will happen as with all technologies.
09:29 So that's another way that we'll find control.
09:33 But I think also just from technically, as we develop AI,
09:37 we have to, in a sense, build controls in.
09:40 So even if a model has the opportunity to give an answer,
09:45 somewhere between 80 and 20, we build hard limits in.
09:50 We have to build hard limits in
09:53 to the answers that it outputs.
09:56 It's so, it's technically, I think, easy and necessary
10:00 to build constraints in and controls in.
10:03 And we have to do that for a whole bunch of reasons,
10:05 regulatory or economic.
10:09 In terms of responsible or responsibility,
10:13 I think that in all of the areas that I've worked
10:16 in the industry in my career,
10:19 we've taken this extremely seriously.
10:22 From bias to make good decisions
10:25 and not under-represent a specific population, for instance.
10:30 Or in terms of the data that we consume,
10:34 making sure that it's good quality data
10:37 and that we have the right to have access to that data,
10:41 for instance.
10:44 So I think responsibility comes with the territory.
10:50 And I think it's a natural part
10:54 of any technology development.
10:57 - Amazing.
10:58 Nicolas, we of course want to thank you.
10:59 And we of course want to thank Kaiju.
11:01 It's an absolute pleasure speaking with all of you.
11:03 Every time the team over there is phenomenal.
11:05 Nicolas Sabryan,
11:06 Director of Artificial Intelligence at Kaiju.
11:09 Thank you so much for joining us today
11:10 and talking more about AI and the effects in the world.
11:13 - Thank you.
11:14 It's been a blast.
11:15 (upbeat music)
11:17 (upbeat music)
11:21 ♪ Hey ♪
11:22 [BLANK_AUDIO]