Latest / ContraMinds Podcast - Unlocking Personal Growth and Professional Excellence / #050 Prof. Vasant Dhar on The Evolving Role of Humans in the Age of AI

Transcript

#050 Prof. Vasant Dhar on The Evolving Role of Humans in the Age of AI · ContraMinds Podcast - Unlocking Personal Growth and Professional Excellence

• 0:00
  SPEAKER_05: This is a big debate at the moment, right?
• 0:02
  SPEAKER_05: Like, should it be open source, for example?
• 0:05
  SPEAKER_05: Right.
• 0:05
  SPEAKER_05: So my my colleague Jan Le Kun, who is head of AI research at Meta, believes that uh you know the that that everything should be open source.
• 0:16
  SPEAKER_05: Um and there's certainly to be something to be said for that, because you know, the whole community gets to participate, it's transparent, like we know what the risks are, you know, we can design policy around it.
• 0:26
  SPEAKER_05: So it has all those advantages.
• 0:29
  SPEAKER_05: You know, but what scares people is you know, I guess Kim Jong il.
• 0:36
  SPEAKER_05: So, you know, you mean like everyone has access to it?
• 0:39
  SPEAKER_05: You know, that's a shame, right?
• 0:42
  SPEAKER_05: So so so that's the you know, that's a downside, is like, you know, it it gives it gives um you know provides ammunition to everyone, the power and the ammunition.
• 0:53
  SPEAKER_05: And so those are some of the uh concerns uh with with open source.
• 0:59
  SPEAKER_05: Yeah, I think there's you know that there's um good reason to go the open source way, but that also brings with it all kinds of other requirements for transparency and use.
• 1:13
  SPEAKER_05: Um and so we may end up in a situation where we have open source, but everything is heavily tracked.
• 1:19
  SPEAKER_05: You know, that that that that that ultimately the responsibility has to lie with someone, and those are some of the most sort of open and vexing issues around AI.
• 1:31
  SPEAKER_05: You saw that there was a Taylor Swift deep fake.
• 1:34
  SPEAKER_05: Yeah, you know, last week that got a lot of attention, but they've been but but this has been going on for a long time, you know, it's been going on for years, it just hasn't gotten a lot of attention because it wasn't Taylor Swift.
• 1:45
  SPEAKER_05: Correct.
• 1:46
  SPEAKER_05: Now that it was, it's gotten a lot of attention, and people say, Oh, you know, it happened to Taylor Swift, it could happen to anyone.
• 1:51
  SPEAKER_05: Well, it has been happening to lots of people already, you know, these deep fakes.
• 1:56
  SPEAKER_00: Today's episode is proudly brought to you by Effortless.
• 2:00
  SPEAKER_00: Effortless is the ultimate solution for startups and SMBs seeking seamless money management and compliance handling, e-invoicing and e viable generation with just a couple of clicks, slashing collection efforts by a staggering 90% through the magic of automated follow-ups, snapping a quick photo of your bill to instantly log expenses.
• 2:23
  SPEAKER_00: With Effortless, you get the power of a consolidated bank account view, seeing both business and personal accounts seamlessly integrated.
• 2:31
  SPEAKER_00: With Effortless, making UPI-like payments is a breeze.
• 2:35
  SPEAKER_00: Effortless keeps you on track, ensuring your finances and compliance are always in top shape.
• 2:41
  SPEAKER_00: Effortless wasn't just created by anyone, it was crafted by chartered accountants and entrepreneurs who understand the pulse of startups and SMBs.
• 2:50
  SPEAKER_00: Visit www.goeffortless.ai and embark on a journey that will redefine the way you do business.
• 2:58
  SPEAKER_00: Effortless, because managing a business should be anything but a hassle.
• 3:28
  SPEAKER_00: This is a very special episode as you have achieved an important milestone of 50 episodes at Contraminds.
• 3:35
  SPEAKER_00: I would like to thank all our fans, our phenomenal guests who believed in us and our story, and of course, our sponsors.
• 3:43
  SPEAKER_00: Today we have Professor Vasandar, who is an artificial intelligence researcher, a data scientist, and the host of the podcast Brave New World.
• 3:53
  SPEAKER_00: Brave New World explores how technology and virtualization in the post-COVID era is transforming humanity.
• 4:02
  SPEAKER_00: He's also the professor of the Stern School of Business in New York and the editor-in-chief of the journal Big Data.
• 4:11
  SPEAKER_00: I spoke to Professor Vasandar on a range of topics on AI, which included the paradigm shifts AI has gone through over the years.
• 4:20
  SPEAKER_00: How open should AI be?
• 4:22
  SPEAKER_00: Should we revisit loss with AI becoming mainstream?
• 4:27
  SPEAKER_00: And the importance of the knowledge of psychology and neuroscience to be a world-class AI professional among many other ideas.
• 4:38
  SPEAKER_00: I thoroughly enjoyed this conversation and the answers that Professor Vasandar gave, which were laser focused and straight from the gut.
• 4:48
  SPEAKER_00: Over to my conversation with Professor Vasandar.
• 4:51
  SPEAKER_00: Hello, Professor Vasandar.
• 4:53
  SPEAKER_00: Thanks a lot.
• 4:53
  SPEAKER_00: I'm really delighted to have you as a special guest in the Contra Minds podcast.
• 4:58
  SPEAKER_00: Thanks for uh accepting my requests.
• 5:01
  SPEAKER_00: Lovely talking to you.
• 5:02
  SPEAKER_05: Thank you, Swami.
• 5:04
  SPEAKER_05: I am delighted to be here.
• 5:07
  SPEAKER_00: So let me uh you know, first uh I just wanted to talk to you about your own journey into how did you get into AI and uh you know how is really the journey from where you started, and what does the Center for Data Science uh do today?
• 5:25
  SPEAKER_05: Uh the journey.
• 5:28
  SPEAKER_05: Uh, you know, as many things in life uh, this was pure serendipity.
• 5:32
  SPEAKER_05: Uh I, you know, I had no idea what I wanted to do in life.
• 5:35
  SPEAKER_05: Uh you know, I was like 22, 23 years old.
• 5:38
  SPEAKER_05: That you know, I was in graduate school in Pittsburgh, and I was uh in the computing center writing up a little program for one of my professors to implement some decision-making algorithm he'd come up with called the analytic hierarchy process.
• 5:53
  SPEAKER_05: And this senior PhD student comes to me and says, Hey, there's this professor who's like uh expert in artificial intelligence, he's built the world's first medical diagnostic system.
• 6:03
  SPEAKER_05: Uh, I'd like him to go offer a doctoral seminar.
• 6:07
  SPEAKER_05: Can you come along?
• 6:08
  SPEAKER_05: I need to show some strength in numbers.
• 6:10
  SPEAKER_05: And I said, like, sure, like, what is AI?
• 6:13
  SPEAKER_05: And he says, you know, it's about making computers smart.
• 6:15
  SPEAKER_05: And I said, Oh, that sounds great.
• 6:16
  SPEAKER_05: Let's go.
• 6:17
  SPEAKER_05: So we did, and we went to the medical school.
• 6:20
  SPEAKER_05: There was a lab up on the 13th floor, I remember, of the medical center at the University of Pittsburgh.
• 6:26
  SPEAKER_05: And um there was a long room, and in the middle of the room, there was a terminal connected to a computer at Stanford.
• 6:32
  SPEAKER_05: You're talking this was like 1979, and so there was no PCs, it was a days of time sharing, and there was a physician uh puffing a cigar talking to quote unquote talking to this system via his assistant who was typing for him.
• 6:47
  SPEAKER_05: And he entered some symptoms about a case, and the machine, the the the program was called Internist.
• 6:54
  SPEAKER_05: It came back and asked him some more questions.
• 6:57
  SPEAKER_05: And at one point, you know, during the dialogue, he said, Why are you asking me this question?
• 7:03
  SPEAKER_05: And internist said, Because the evidence you've given me so far is consistent with the following hypotheses, and this will help me discriminate between the top two.
• 7:12
  SPEAKER_05: And I was just blown away by that.
• 7:14
  SPEAKER_05: It just completely blew my mind to see a computer behaving in that way.
• 7:21
  SPEAKER_05: You know, my previous uh interactions with computers had been, you know, writing Fortran programs to solve you know complex differential equations, you know, iteratively, you know, as part of my chemical engineering days in IIT.
• 7:34
  SPEAKER_05: So this was just like blew my mind to see a machine having a dialogue with a human like this and discussing a case.
• 7:40
  SPEAKER_05: And it was clear that it wasn't following some sort of pre-specified logic, but it was actually sensitive to context, you know, and so that's what like you know, totally blew me away.
• 7:51
  SPEAKER_05: And uh, you know, and then we were waiting for the prof to get off the phone, you know, he was in his office, and then we went in and he was like just so sweet, you know, uh to us and said he'd be delighted to offer the course.
• 8:03
  SPEAKER_05: And you know, he ended up becoming one of my main mentors at graduate school and my thesis advisor.
• 8:08
  SPEAKER_05: So if that's how I sort of you know landed in AI, it was actually through medical diagnosis.
• 8:16
  SPEAKER_05: And and he introduced me to another, you know, pioneer in AI called Herbert Simon, who was actually a Nobel laureate who had won uh the Nobel Prize of Economics, and he was one of the people who coined the term artificial intelligence.
• 8:30
  SPEAKER_05: So he introduced me to Simon, and that sort of set me along this path that has, you know, um, yeah, that that that's what set me going.
• 8:39
  SPEAKER_05: And then so when I was graduating, I was looking for a job, and uh, you know, I walked down Washington Square south along the park for my job interview, and I thought, hey, this would be kind of a cool place to live.
• 8:52
  SPEAKER_05: And uh as uh luck would have it, you know, my talk went well.
• 8:58
  SPEAKER_05: I really got along with my colleagues, and and I ended up in NYU.
• 9:01
  SPEAKER_05: And so I've been there ever since, except for you know, a few years on Wall Street uh in the in the mid-90s where I brought machine learning to Wall Street.
• 9:10
  SPEAKER_05: And um and so I've been fortunate to have been, I guess, what some people call a prakedemic.
• 9:16
  SPEAKER_05: You know, I am I'm an academic, I'm sort of a researcher, but I get to actually try my ideas out and make them work in practice.
• 9:25
  SPEAKER_05: And in this case, the ideas really had to do with you know the fact I was in New York and in the financial industry, and and so uh market prediction seemed to be sort of uh you know the obvious thing to try out, you know, given that data was becoming available.
• 9:42
  SPEAKER_05: And so that's what I got interested in the um mid-90s.
• 9:46
  SPEAKER_05: Uh went to Wall Street, took a bunch of my PhD students along, and uh, you know, set up the first uh machine learning-based hedge fund.
• 9:54
  SPEAKER_05: So I've been very fortunate to have had the opportunities to have, you know, uh to have met great people who set me along a path that I've really enjoyed, I have to say.
• 10:06
  SPEAKER_05: It's been a lot of fun.
• 10:07
  SPEAKER_00: You talk about the uh you know the Kun Net you know framework of a whole set of paradigm shifts that AI has gone through, right?
• 10:19
  SPEAKER_00: Right, right.
• 10:20
  SPEAKER_00: Uh so uh what are the paradigm shifts that AI has gone through and what impact uh has each of the shifts happened for the business and the society at large?
• 10:31
  SPEAKER_00: So, what are these shifts that you have seen and can you talk to us about it?
• 10:36
  SPEAKER_05: Yeah, that's a great question.
• 10:37
  SPEAKER_05: And uh, you know, uh there's this article I wrote called The Paradigm Shifts in Artificial Intelligence that I posted on the archive last year, um, which I initially wrote as sort of an initial opening article for a PhD seminar that I was poteaching, you know, just to give students kind of a grounding about the history of AI and what got us here and why does it matter.
• 11:04
  SPEAKER_05: And you know, so I I start the sort of story of the or rather the history of AI, you know, at the end of sort of what I call the game playing era.
• 11:15
  SPEAKER_05: You know, in the 60s there was a lot of experimentation with sort of game playing algorithms uh for searching graphs, uh, you know, which sort of came out of the areas of sort of optimization and and search.
• 11:29
  SPEAKER_05: Um but in the 70s, I'd say there was a whole new era of what I called expert systems that got going where people felt that for specific domains like medicine, which is the poster child for AI, you know, and engineering and you know, problems with a certain amount of structure in it, you should be able to talk to human experts and extract the expertise in a way that was representable in a machine that could then be used for reasoning about you know that domain.
• 12:04
  SPEAKER_05: So medical diagnosis uh you know being a uh you know a prime example where the system that I referred to, internist, had been built over 10 years of interaction between um you know computer scientists and specialists, you know, diagnosticians in medicine.
• 12:22
  SPEAKER_05: And the knowledge has been had been extracted case by case, carefully assembled, represented, you know, in the forms of hierarchies, rules, causal relationships, all of that kind of stuff.
• 12:32
  SPEAKER_05: You know, the knowledge was extracted and declared, and then there was a way to sort of use it in the context of a new case.
• 12:39
  SPEAKER_05: And that's what I was observing.
• 12:40
  SPEAKER_05: You know, my first interaction with AI was this expert systems paradigm and a physician, expert physician discussing a case with internists.
• 12:49
  SPEAKER_05: And I'll come full circle to this, by the way.
• 12:52
  SPEAKER_05: So that was the expert systems era, but it was you know very hard to extract knowledge reliably from experts and to represent things like uncertainty, right?
• 13:05
  SPEAKER_05: Because at that time people were thinking about knowledge in terms of rules, but then how do you attach uncertainty and context and all that?
• 13:12
  SPEAKER_05: So it became really difficult to do that in a top-down way.
• 13:17
  SPEAKER_05: And you know, the machine learning era sort of started emerging in the 80s and 90s, you know, where people were saying, hey, you know, why you know, it's sort of like trying to specify these rules, why can't we learn them from the data?
• 13:28
  SPEAKER_05: Let's just look at lots of cases about medical diagnosis and from those abstract out you know the knowledge which can then be applied, right?
• 13:36
  SPEAKER_05: So that was sort of the next era of machine learning where you could take data, you could, you know, do what was called feature engineering, curate it properly, and then you know, use a machine learning algorithm to extract patterns from data, you know, to say that if I see someone has pallor, then maybe they have excess bilirubin, and so maybe there's something causing that elevated level of bilirubin in the blood, right?
• 14:01
  SPEAKER_05: So uh so so that was the uh the you know the context.
• 14:05
  SPEAKER_05: So now you could actually derive those rules instead of having an expert specify them.
• 14:10
  SPEAKER_05: Um, but it was still really cumbersome to do feature engineering, right?
• 14:14
  SPEAKER_05: So if you had to, let's say, re um uh use an X-ray image or an MRI as part of diagnosis, the physician had to translate that into a vocabulary and and and say that to the machine, right?
• 14:27
  SPEAKER_05: So so that wasn't great.
• 14:29
  SPEAKER_05: And so what the sort of new sort of deep learning paradigm did was that it solved perception, right?
• 14:35
  SPEAKER_05: That now computers could actually see the world and um listen to it and read about it in the way that humans do, and so they could ingest the input directly from the environment.
• 14:46
  SPEAKER_05: So that was a big deal, was solving perception.
• 14:48
  SPEAKER_05: Yeah, so intelligence sort of went more upstream.
• 14:51
  SPEAKER_05: Okay.
• 14:52
  SPEAKER_05: And and then the latest paradigm shift, which I call you know the shift to general intelligence, is this emergence of these pre-trained models like large language models that are just pre-trained with you know lots of knowledge that is available on the internet, um, but they're not pre-trained with any specific purpose necessarily, you know, other than predict the next word.
• 15:13
  SPEAKER_05: But it turns out that they learn a lot about the world just by doing you know, prediction of the next word, and they're usable for all kinds of things, like general question-answering systems, writing code and all that kind of stuff, right?
• 15:24
  SPEAKER_05: Now I'll end by saying this.
• 15:26
  SPEAKER_05: I took some of the dialogues that I had from internist days, yeah, and I tried working with ChatGPT.
• 15:33
  SPEAKER_05: I fed ChatGPT the exact same symptoms about a case, and I said, you know, what's going on in this case?
• 15:40
  SPEAKER_05: And ChatGPT actually was amazing, right?
• 15:44
  SPEAKER_05: Because it's very conversational.
• 15:45
  SPEAKER_05: It said, Oh, you know, yeah, I see what's going on here.
• 15:49
  SPEAKER_05: Maybe there's blah, blah, blah.
• 15:50
  SPEAKER_05: And then it just stops, you know, and that's it.
• 15:54
  SPEAKER_05: You know, and then I'm like, but tell me more.
• 15:56
  SPEAKER_05: And it's like, well, what do you mean?
• 15:58
  SPEAKER_05: You know, and so then I have to sort of feed it another symptom, like, you know, uh, you know, which internalists had sort of asked for, you know, like, oh, was there a pain in the lower left quadrant?
• 16:09
  SPEAKER_05: Right.
• 16:09
  SPEAKER_05: So then I'm telling ChatGPT, by the way, there was no pain in the, you know, you know, you know, after the patient had a big meal.
• 16:17
  SPEAKER_05: And so ChatGPT says, Oh, okay, that's good evidence in that case, you know, and then it chugs along a little bit more and gives me all kinds of stuff.
• 16:25
  SPEAKER_05: Okay, it, you know, it's so it hasn't gotten to the point where it gives me the kind of dialogue that I could have with internist, you know, 50 years ago.
• 16:32
  SPEAKER_05: So we've come a long way, you know, from the days of specifying knowledge to the machine just learning everything on its own, which is where we are now.
• 16:40
  SPEAKER_05: But we still have a very long ways to go in terms of like re you know reaching you know human levels of intelligence.
• 16:47
  SPEAKER_05: But I think we will get there, and I think the machine will even surpass us, you know, in the future.
• 16:53
  SPEAKER_00: Brilliant.
• 16:54
  SPEAKER_00: I think uh your uh way of uh summarizing the four eras was just quite amazing and very, very uh you know, I would say very vivid is what I would call it.
• 17:05
  SPEAKER_00: Um, I had one question around uh this debate around you know the connectionist models, right?
• 17:12
  SPEAKER_00: And how these connectionist models, there's a lot of debate around saying, hey, neural networks initially uh you know during the Minsky era was you know, it you know, they stopped working saying hey neural networks can't do reasoning too much.
• 17:26
  SPEAKER_00: And uh now when you look at how it has accelerated uh the AI uh you know development and the kind of stuff that has got done, do you believe uh you know there is a that do you believe there's a lot of uh you know uh what I would call as uh uh you know reasons to believe that the connectionist models are the most important when you want to drive this momentum towards what you call as artificial general intelligence?
• 17:56
  SPEAKER_05: I think that it will certainly be a major part of the future of AI going forward for a number of reasons.
• 18:07
  SPEAKER_05: One is that uh we've we've sort of looked inside the brain and we find that it's actually a neural network, right?
• 18:14
  SPEAKER_05: So there's good reason to believe that that this biological architecture that's motivated the artificial neural net architecture is a reasonable one for many kinds of um, you know, for for capturing types of intelligence, especially when it comes to perception, but but even cognition, right?
• 18:36
  SPEAKER_05: But when it comes to perception, this architecture seems to have some pretty major advantages, right, in being able to do object recognition, language understanding, and so on.
• 18:47
  SPEAKER_05: So it'll certainly be a big part of you know AI going forward.
• 18:52
  SPEAKER_05: Um whether it accomplishes everything really remains to be seen, right?
• 18:57
  SPEAKER_05: Because while we made huge advances in perception and you know all kinds of sort of general intelligence kinds of capabilities, uh we've also regressed in some ways, you know, because these these uh models are so complex that it's very hard to look inside them and see what's really going on.
• 19:16
  SPEAKER_05: So relative to sort of the earliest paradigm, you know, the the expert system one where everything was explicit and you could see why the machine was doing what it was doing, it's become impossibly difficult to do that now with these large, sort of complex models.
• 19:30
  SPEAKER_05: And so that's not surprisingly, that's where a lot of action is to get some more transparency into like what's really going on under the hood, how do they represent knowledge, yeah, you know, what what what do they really know, what do they really understand?
• 19:43
  SPEAKER_05: Uh so in a nutshell, I'd say, yeah, it's going to be an important part of AI going forward, but I'm not sure it'll solve everything that we need to solve in order to get to you know human-level intelligence and then super intelligence beyond that.
• 19:58
  SPEAKER_00: Okay.
• 19:58
  SPEAKER_00: So uh which which uh you know gives me the next question.
• 20:02
  SPEAKER_00: I think uh you've kind of got me into uh the next question that I wanted to ask.
• 20:06
  SPEAKER_00: How open should be uh you know, how open should AI be?
• 20:13
  SPEAKER_05: This is a big debate at the moment, right?
• 20:16
  SPEAKER_05: Like, should it be open source, for example?
• 20:19
  SPEAKER_05: Right?
• 20:19
  SPEAKER_05: So my my colleague Jan Le Kun, who is head of AI research at Meta, believes that uh you know the that everything should be open source.
• 20:30
  SPEAKER_05: Um and there's certainly to be something to be said for that because you know the whole community gets to participate, it's transparent, like we know what the risks are, you know, we can design policy around it.
• 20:40
  SPEAKER_05: So it has all those advantages.
• 20:43
  SPEAKER_05: Um, you know, but what scares people is you know, I guess Kim Jong il.
• 20:50
  SPEAKER_05: So, you know, you mean like everyone has access to it, you know?
• 20:54
  SPEAKER_05: Yeah, that's a shame, right?
• 20:56
  SPEAKER_05: So so so that's the you know that's the downside, is like you know, it it gives it gives um you know provides ammunition to everyone, power and the ammunition.
• 21:07
  SPEAKER_05: And so those are some of the uh concerns uh with with open source.
• 21:13
  SPEAKER_05: But if you were to ask me, you know, um you know what my position would be, I'd say, yeah, I think there's you know that there's um good reason to go the open source way, but that also brings with it all kinds of other requirements for transparency and use.
• 21:33
  SPEAKER_05: Um and so we may end up in a situation where we have open source, but everything is heavily tracked, you know.
• 21:40
  SPEAKER_05: That that that that that ultimately the the responsibility has to lie with someone, and those are some of the most sort of open and vexing issues around AI.
• 21:51
  SPEAKER_05: You saw that it was a Taylor Swift deep fake, yeah.
• 21:55
  SPEAKER_05: You know, last week that got a lot of attention, but they've been but but this has been going on for A long time, you know, it's been going on for years.
• 22:02
  SPEAKER_05: It just hasn't gotten a lot of attention because it wasn't Taylor Swift.
• 22:05
  unknown: Okay.
• 22:06
  SPEAKER_05: Now that it was, it's gotten a lot of attention, and people say, Oh, you know, it happened to Taylor Swift, it could happen to anyone.
• 22:11
  SPEAKER_05: Well, it has been happening to lots of people already.
• 22:14
  SPEAKER_05: You know, these deep fakes and you know, people producing, you know, pornographic films using the likeness of people.
• 22:20
  SPEAKER_05: Uh, and you know, and and the trouble is that there aren't even enough uh laws at the moment that you know that that say that a crime has been committed, let alone you know who's liable and how to address it.
• 22:32
  SPEAKER_00: Yeah.
• 22:33
  SPEAKER_00: So I think uh you talk about uh that a lot, which is about the risks that uh you know these algorithms really uh provide, and therefore how do you mitigate them?
• 22:44
  SPEAKER_00: So there are no uh written laws today.
• 22:47
  SPEAKER_00: So, how how do you think this would evolve?
• 22:49
  SPEAKER_00: Because you know, uh I I think the way I see it is uh this is not being done in the past, right?
• 22:55
  SPEAKER_00: Like what uh you write about in one of your blogs, you talk about how electricity changed manufacturing, internet changed knowledge work, right?
• 23:04
  SPEAKER_00: And now AI is changing a whole host of professions, a whole host of industries, and that is going to really need a whole set of rules.
• 23:15
  SPEAKER_00: That's I think something that you talk about in terms of transparency, in terms of risk mitigation.
• 23:20
  SPEAKER_00: How does one start?
• 23:22
  SPEAKER_00: And in your opinion, is there some kind of a set of frameworks that people need to look at?
• 23:29
  SPEAKER_05: Yeah, that's a great question, and I you know, I mean, I've written about the fact that this issue tends to be uh uh you know one that is being addressed differently in different parts of the world.
• 23:46
  SPEAKER_05: Right?
• 23:47
  SPEAKER_05: In the US, you know, not surprisingly, the culture is you know, let a thousand flowers bloom and we'll we'll deal with these problems, you know, as and when they arise, right?
• 23:57
  SPEAKER_05: Let the market decide what's best and and let's not interfere.
• 24:03
  SPEAKER_05: So that's like on the one extreme.
• 24:05
  SPEAKER_05: Um on the other extreme, you have China, which is observing everything, they own all the data, they have the right to do anything with it, right?
• 24:13
  SPEAKER_05: So that's like a totalitarian sort of uh sort of or rather an authoritarian kind of a model, yeah.
• 24:20
  SPEAKER_05: Um, you know, and then you have Europe, which is sort of concerned about risks of data misuse, you know, with the whole GDPR regulation coming in, um you know, and and those kinds of concerns.
• 24:34
  SPEAKER_05: And then India's got sort of another interesting kind of model, you know, with this the emergence of you know biometric identification.
• 24:42
  SPEAKER_05: And now traveling in India has become, I mean, you know, electronics are everywhere, right?
• 24:46
  SPEAKER_05: It's just right, and in a sense, it's become easy, right?
• 24:50
  SPEAKER_05: Yeah, you know, you can right, you can use these IDs, and a lot of things become easier, so it makes life easier in many ways for lots of people, like having uh you know an identifier where you can authenticate yourself, which by the way doesn't exist in most other parts of the world, you know, strangely enough, right?
• 25:08
  SPEAKER_05: I mean in the US, you can't actually prove who you are, you know just through an identity like you can in India.
• 25:15
  SPEAKER_05: Yeah, um uh but you know there's there are concerns in India that you know the institutions maybe are not uh strong enough to you know resist government overreach of data and you know things like that.
• 25:28
  SPEAKER_05: So there's there's there are those kinds of concerns uh in India around privacy and misuse of these kinds of systems.
• 25:35
  SPEAKER_05: So you know, there isn't any one answer, it's almost like it's being dealt at you know uh in different parts of the world differently.
• 25:43
  SPEAKER_05: Yeah.
• 25:44
  SPEAKER_05: Um and uh yeah, well, and and and I and I guess it'll just continue to evolve, or I was gonna say stumble along like that, right?
• 25:54
  SPEAKER_05: You get things that you know, mishaps that happen, things don't work.
• 25:58
  SPEAKER_03: Yeah.
• 25:58
  SPEAKER_05: Uh you know, in the US we'll see lawsuits, you know, we're already seeing that uh you know, around you know, Section 230 and platform liability, but we'll probably see more of those as the technology comes along, right?
• 26:09
  SPEAKER_05: So the US sort of deals with it on a you know on a you know ongoing basis, you know, with very light touch.
• 26:17
  SPEAKER_05: Yeah.
• 26:18
  SPEAKER_05: Um yeah, remains to be seen how other parts of the world will deal with it.
• 26:23
  SPEAKER_00: But uh, don't you think, Professor, that uh when you look at the big tech firms today, the fact that uh you know uh they were allowed to blossom and uh now they have become so large and almost monopolistic, right?
• 26:40
  SPEAKER_00: In terms of size, in that in terms of the data that they hold, and therefore there's a bit of worry that hey, these these four or five big tech firms typically control all data, and uh they have been allowed to sprout, they've become so large today to entangle and get them out, untangle them is also getting difficult.
• 27:07
  SPEAKER_05: It is indeed.
• 27:08
  SPEAKER_05: I mean, once once uh you know, once these platforms get established and dominant, it becomes very difficult to uh to displace them.
• 27:17
  SPEAKER_05: You know, I mean that's one of the lessons that we've learned from the internet is uh you know when these platforms become large, it's almost like a winner-take-all outcome.
• 27:27
  SPEAKER_05: Yeah.
• 27:29
  SPEAKER_05: So uh and that's what we see here.
• 27:31
  SPEAKER_05: So so you're right, but you know, but at the same time, I think society has become more aware of you know you know this imbalance of power.
• 27:42
  SPEAKER_05: And so you know, I would expect to see you know, possibly even you know, people working with the government to control um you know the private sector.
• 27:54
  SPEAKER_05: I mean that's so the government and the people might become allies uh in in in resisting this yeah uh I guess what I'd call data hegemony.
• 28:09
  SPEAKER_00: Yeah, absolutely.
• 28:11
  SPEAKER_00: Uh the uh other thing that I wanted to dwell into was this one article that you had written about uh how conversational AI versus search, right?
• 28:23
  SPEAKER_00: This whole hawkistic growth that uh you know uh how search is getting disrupted with conversational AI.
• 28:30
  SPEAKER_00: Uh, do you believe uh in the next uh decade or so search will completely get transformed with conversational AI and how would the world look like with conversational AI dominating?
• 28:44
  SPEAKER_00: In fact, you've written there saying it's almost giving sleepless nights to Google that Microsoft has got there, right?
• 28:51
  SPEAKER_00: So therefore, yeah, yeah, yeah.
• 28:54
  SPEAKER_05: So so this was something I I wrote about in uh I guess just a little more than a year ago, after the emergence of ChatGPT, where I said that you know, I I point to the fact that sometimes uh technologies get disrupted, right?
• 29:10
  SPEAKER_05: And that search had been improving steadily, you know, slowly and steadily over a long period of time, you know, and you know, Google was the dominant player in search, you know, Bing was like a distant number two, right?
• 29:26
  SPEAKER_05: So um and what happens in in markets like that, you know, in the and so search is sort of becoming mature, but you know, there's a framework called the innovator's dilemma by by Clay Christensen, right?
• 29:40
  SPEAKER_05: Which describes this really well, which is that incumbents get comfortable and and that they're catering to a market and they forget to see a threat that is emerging that might be a lot inferior, but it appeals to some people, and then it suddenly starts appealing to the entire market.
• 29:57
  SPEAKER_05: Yeah.
• 29:58
  SPEAKER_05: Um, and that's the way conversational AI sort of emerged.
• 30:01
  SPEAKER_05: Emerged as a tool, right?
• 30:02
  SPEAKER_05: It got people's attention and it it captured people's imagination, and suddenly that was the buzzword.
• 30:09
  SPEAKER_05: And so, why do you really need search?
• 30:11
  SPEAKER_05: Yeah, you know, boring plain search when you've got conversational AI.
• 30:16
  SPEAKER_05: So, yeah, I mean, and the ironic thing is that the architecture, this whole transformer architecture that underlies large language models was you know developed at Google.
• 30:26
  SPEAKER_05: Uh but Google was like too slow to actually capitalize on it, whereas OpenAI being this you know much more you know, much smaller uh you know company with no liability was able to sort of release it into the wild and actually have people play with it.
• 30:43
  SPEAKER_05: And it was brilliant for Microsoft to have taken a large stake in this because Microsoft couldn't have done this by themselves, right?
• 30:50
  SPEAKER_05: I mean their previous attempt at the chat port was a disaster.
• 30:54
  SPEAKER_05: Yeah, but they're not that kind of company, whereas you know, through OpenAI, they could do it without incurring any kind of risk.
• 31:02
  SPEAKER_05: Uh but to come back and sort of answer your question and bring it full circle, uh, yeah, I think going forward we'll probably see search, traditional search more or less disappear and be replaced by you know a much better intelligence in the form of sort of conversational AI.
• 31:20
  SPEAKER_00: Okay.
• 31:21
  SPEAKER_00: So which means that uh uh the business models around how conversational AI will then get monetized, uh unlike how search is monetized today, which is largely through ad networks, right?
• 31:36
  SPEAKER_00: Yeah, uh and therefore there'll be a whole new model that one needs to think, which could be a subscription model, like the way OpenAI AI Pro has got, which is really you need to subscribe to some uh you know some services around there.
• 31:50
  SPEAKER_00: Would would there be new business models if you were to look at yeah, yeah, absolutely, yeah.
• 31:56
  SPEAKER_05: I mean, and you can already see this, right?
• 31:58
  SPEAKER_05: I mean, open AI is essentially like a freemium model, right?
• 32:01
  SPEAKER_05: Yeah, you know, they make they make 3.5 available for free.
• 32:05
  SPEAKER_05: If you want Chat GPT 4, you've got to pay 20 bucks a month.
• 32:09
  SPEAKER_05: Yeah, and then if you want more, and and and but but you're limited still, right?
• 32:13
  SPEAKER_05: You can't have like a huge transcript, you can put it in GPT-4, it complains.
• 32:18
  SPEAKER_05: And so if you need to do something more sophisticated, you know, for example, I wanted to get GPT to take my podcast with Aswat Tamodharan and summarize it and give me the key five points, right?
• 32:32
  SPEAKER_05: And so I uh you know basically had to get OpenAI's program called Whisper to take my MP3 file, translate you know, translate that into a text file, and then you know, use an API to say, okay, whatever text comes, summarize the five key points, right?
• 32:49
  SPEAKER_05: And it cost me you know 22 cents for the uh transcription to text and 23 cents uh for that API call to actually give me the five points, yeah.
• 32:59
  SPEAKER_05: Which the free version could have given me, but not for the entire transcript, right?
• 33:03
  SPEAKER_05: So we're already seeing this emerging.
• 33:05
  SPEAKER_05: So uh I I think there's a lot of money to be made here, uh both at the retail and business level.
• 33:12
  SPEAKER_00: So so therefore, if you if you if you do a you know blue sky thinking around how businesses can take advantage of something like this, uh you had written about this uh in your real estate example.
• 33:25
  SPEAKER_00: I saw in one of your newsletters, uh, how you applied it in the real estate world.
• 33:31
  SPEAKER_00: Uh, how do you think uh you know businesses can take advantage if I was an automotive company or if I were a retail company?
• 33:38
  SPEAKER_00: And I because there's a lot of discussion around, but if I were a retail company, how do I take advantage of this?
• 33:46
  SPEAKER_05: And that's a great question.
• 33:47
  SPEAKER_05: And the answer goes back to sort of what I was saying earlier, which is that you know deep learning solves perception, right?
• 33:53
  SPEAKER_05: A lot of human work involves perception, right?
• 33:57
  SPEAKER_05: It it involves going looking at something, it involves reading something, right?
• 34:01
  SPEAKER_05: Uh it won it involves listening to something, to people, right?
• 34:05
  SPEAKER_05: So a lot of the input that humans get is perceptual.
• 34:09
  SPEAKER_05: And you know, traditionally that had to be transformed into something for the machine to process.
• 34:14
  SPEAKER_05: Now imagine, you know, instead of in the real estate example you referred to, you know, you know, you know you don't need to have like an inspector to send to a house to look at its condition and take pictures and stuff like that, right?
• 34:27
  SPEAKER_05: So in the example I mentioned, I just use Google Street View images to uh predict the condition of a house.
• 34:34
  SPEAKER_05: So I trained, so I got a bunch of humans to say, hey, what's the condition of this house?
• 34:38
  SPEAKER_05: Got Google Street View to look at it, and the results were you know pretty much very similar to human results, right?
• 34:46
  SPEAKER_05: So if you can actually eliminate a lot of that grunt work, the very inefficient work, right?
• 34:52
  SPEAKER_05: Someone has to go to a place, take pictures, and then translate those into language and put it into a report, right?
• 35:01
  SPEAKER_05: There's a lot of human work, a lot of inefficiency involved in that, right?
• 35:05
  SPEAKER_05: And you're paying people hundreds of thousands of dollars to do that, right?
• 35:10
  SPEAKER_05: Uh and now you can do that with a machine that says, hey, you know, take these, send a drone over there, take a bunch of pictures.
• 35:16
  SPEAKER_05: You can even send a drone inside a house to take pictures, right?
• 35:19
  SPEAKER_05: Uh you can, you know, process unstructured data.
• 35:23
  SPEAKER_05: So a lot of this sort of general intelligence becomes available at really low cost.
• 35:32
  SPEAKER_05: Right.
• 35:32
  SPEAKER_05: And and this is what I find like incredibly uh exciting about what's coming down the road is that a lot of the sort of grunt work that humans do can really get streamlined and done equally equally well or even better by the machine.
• 35:48
  SPEAKER_05: So it has huge uh potential both at the retail level, just you and me using the machine, or businesses trying to make processes more efficient or better through AI.
• 36:01
  SPEAKER_00: So it's almost like uh saying that uh so the internet uh typically got the supply chain optimization uh you know out of out of the way.
• 36:12
  SPEAKER_00: And now you're talking about uh the grunt work, which is really the effort and information processing, uh becoming far more efficient, and therefore there's a whole new information supply chain that might get optimized because of AI.
• 36:26
  SPEAKER_05: Exactly, yeah.
• 36:28
  SPEAKER_05: You put it better than I did for me.
• 36:33
  SPEAKER_00: Uh so so clearly, if I were uh uh you know a retail CEO or an automotive uh uh you know CEO, or if I were uh running a large enterprise uh uh say a bank, uh if I were to look at how do I when do I start, what would you suggest saying okay, what kind of experiments should I be doing?
• 36:54
  SPEAKER_00: Because there's a lot of talk currently saying, Oh, here is Gen AI, you start here, you start here.
• 36:59
  SPEAKER_00: So, is there a way in which I could uh you know say that hey, here is a roadmap that you need to try with all these uh uh you know LLMs.
• 37:09
  SPEAKER_05: So I don't think there's a roadmap.
• 37:11
  SPEAKER_05: That would be nice to actually have a roadmap, right?
• 37:14
  SPEAKER_05: To go into a business CEO and say, hey, here's the roadmap for you for AI, go do it.
• 37:18
  SPEAKER_05: Uh that doesn't exist.
• 37:20
  SPEAKER_05: Uh in fact, I see a lot of confusion uh and concern in in you know amongst senior management.
• 37:25
  SPEAKER_05: Like we need to do something, but what do we do?
• 37:27
  SPEAKER_05: Right, so there's a lot of a lot of FOMO, right?
• 37:30
  SPEAKER_05: Doing something because if we don't do anything, bad things will happen.
• 37:34
  SPEAKER_00: Or my board is asking me some questions, so I have to answer them.
• 37:37
  SPEAKER_00: So let me put up some couple of slides and take it up there, right?
• 37:41
  SPEAKER_05: Yeah, exactly.
• 37:41
  SPEAKER_05: Yeah, and and this is and this is sort of I think where sort of senior management's creativity and imagination, you know, really comes into the picture.
• 37:50
  SPEAKER_05: I mean, and to me, this is what I think leadership in the age of AI is really all about, like figuring this thing out, because there is no roadmap, right?
• 37:57
  SPEAKER_05: No one's gonna tell you what to do.
• 37:59
  SPEAKER_05: I mean, I had this great conversation with Piyush Gupta, who's the CEO of DBS, yeah, correct.
• 38:04
  SPEAKER_05: Uh, last year, about you know, what are you thinking about with AI?
• 38:09
  SPEAKER_05: Um, you know, and he mentioned a couple of uh examples like use cases, you know, where he said, you know, like we'd like to be able to predict when someone isn't is likely to leave, right?
• 38:19
  SPEAKER_05: Because talent retention is a big part of you know effective management, right?
• 38:26
  SPEAKER_05: Because when talent leaves the door, uh that that can be debilitating, especially like really important, you know, you know, people who are important to the business.
• 38:35
  SPEAKER_05: So we'd like to be able to retain talent and so use AI to actually tell us when you know someone's at risk, which I thought was an interesting application.
• 38:44
  SPEAKER_05: And then I asked him another question, which is how about like using AI to actually capture the knowledge of people who may leave?
• 38:53
  SPEAKER_05: You know, so you can say, you know, Swami was like an amazing Python programmer, right?
• 38:57
  SPEAKER_05: I mean, he just killed it with all those implementations, but now he's you know he's gone and he's working for Citibank.
• 39:04
  SPEAKER_05: You know, how do we get uh virtual Swami to you know write some code for us in his style?
• 39:12
  SPEAKER_05: I mean, imagine something like that, right?
• 39:14
  SPEAKER_05: Where you actually have the ability to now retain knowledge in the organization that previously would just sort of bleed out, right?
• 39:23
  SPEAKER_05: To me, this seems like a huge uh opportunity to to sort of retain knowledge and just be effective uh you know and just literally become a learning organization, right?
• 39:36
  SPEAKER_05: It's it's like you you just learn and retain and can reuse information over time.
• 39:41
  SPEAKER_05: So lots of possibilities uh for how to uh you know to leverage AI, but no roadmap, but lots of interesting ideas.
• 39:52
  SPEAKER_00: So which means that uh like what you said, should therefore as a leader, uh me as a leader, should I should be doing a lot of experiments with AI and the ability to creatively come up with use cases uh is really what the leader's role in the age of AI is going to be.
• 40:11
  SPEAKER_00: Is that how you would define it?
• 40:13
  SPEAKER_05: Yeah, it it really is, you know, because one of the you know, I I think one of the, you know, just sort of uh you know, building on this uh DBS bank case, I think one of the things that they did early on was things like, you know, let's let let's really sort of get a better sense of how well our apps are working, you know, like how happy are people, but without them having to tell us this, right?
• 40:38
  SPEAKER_05: So let's just sort of um maintain touch points, and if we find that uh people aren't happy, we can actually step in and improve the experience, right?
• 40:49
  SPEAKER_05: So you know, being sort of more connected or using data to be more connected and more intelligent about the way you interact with your customers, right?
• 40:59
  SPEAKER_05: So that was you know, so those are the initiatives they took, like, you know, I guess 10-ish years ago, like five to ten years ago, that have really worked out for them.
• 41:08
  SPEAKER_05: And now the question is like, what's the next level?
• 41:11
  SPEAKER_05: You know, what's what's the next frontier?
• 41:13
  SPEAKER_00: Yeah, okay.
• 41:15
  SPEAKER_00: Uh the uh which brings me to a very important point that you talk about, and uh, you know, which is uh AI with decision, decision making with AI.
• 41:24
  SPEAKER_00: Uh, the real challenge of taking decisions with AI is we still have to get about the nuanced introspection that humans uh go through, or uh the fact that uh there's a bit of transparency and there's a there's a depth of consciousness that you talk about, right?
• 41:46
  SPEAKER_00: Which is really something that as humans we go through.
• 41:49
  SPEAKER_00: So that's never that kind of data never gets captured in an organization, and everything is around uh how the data is captured in a structured form in some systems, but not in the way probably the way the internet captured it, and therefore you were able to build models.
• 42:07
  SPEAKER_00: So, if I were an organization, would you think uh the way I should be looking at AI is how do I capture the context as you call it?
• 42:19
  SPEAKER_00: Because context capturing requires a new data model, right?
• 42:25
  SPEAKER_05: Yeah, so I mean the you know this this is a very interesting question about um AI for decision making.
• 42:34
  SPEAKER_05: So let me use a slightly different example, right?
• 42:39
  SPEAKER_05: Uh and and I I actually uh uh you know was at a a company recently where this came up, you know, where I was talking to uh some of the managers about the what how are you thinking about using AI and generative AI in your business?
• 42:58
  SPEAKER_05: And you know, this gentleman happened to mention, you know, we're trying to address our pain points.
• 43:03
  SPEAKER_05: And I said, Well, what are those pain points?
• 43:06
  SPEAKER_05: And he said, Well, for example, uh, you know, reviewing uh a complex thousand-page legal contract, you know, that that's uh that that's complicated.
• 43:16
  SPEAKER_05: It takes a human a long time.
• 43:19
  SPEAKER_05: And I thought that was a really interesting example of uh interesting use case of AI because the the question really is the following which is if you get an AI to process that really complex legal contract, are you willing to trust it?
• 43:32
  SPEAKER_05: Right?
• 43:33
  SPEAKER_05: Are you willing to trust its decisions about that it's okay or or or not?
• 43:39
  SPEAKER_05: Um the not is probably a little bit easier because it'll flag it for you and you can tell whether you agree with it or not.
• 43:45
  SPEAKER_05: But what if about all those things where it said it was okay and it didn't even flag them for you?
• 43:50
  SPEAKER_05: Yeah, would you trust it completely?
• 43:53
  SPEAKER_05: And at the moment we don't, right?
• 43:55
  SPEAKER_05: Because we just have this sort of lingering feeling that it might have missed it, it might have missed some important plan.
• 44:00
  SPEAKER_05: And so we'd have a human go through the thing again anyway, right?
• 44:03
  SPEAKER_05: And then so the question is like, so what did I gain by using the AI if a human had to go through it anyway?
• 44:09
  SPEAKER_05: Yeah.
• 44:10
  SPEAKER_05: Well, maybe I gained something in efficiency, I don't know.
• 44:13
  SPEAKER_05: Maybe I gained something in terms of accuracy, I don't know.
• 44:18
  SPEAKER_05: But that's the problem right now, is we just don't know because there haven't been enough evaluations done of the technology like in use in the wild, right?
• 44:30
  SPEAKER_05: So at the moment, there's a lot of excitement when, hey, look what I can do.
• 44:34
  SPEAKER_05: Look, ma, no hands.
• 44:35
  SPEAKER_05: I can do this, I can do this, I can do this.
• 44:37
  SPEAKER_05: Yeah, you can do that, but is it going to make things better?
• 44:42
  SPEAKER_05: And how?
• 44:43
  SPEAKER_05: Right?
• 44:43
  SPEAKER_05: And those are the biggest sort of unanswered questions.
• 44:46
  SPEAKER_05: Now, it's easy for me to say this, but it's very difficult in practice for organizations to do it because you know, businesses are not in the business of doing science, they're in the business of making money.
• 44:56
  SPEAKER_05: Yeah.
• 44:56
  SPEAKER_05: And so there's only so much uh scientific experimentation they can do, you know, or A B tests, right?
• 45:04
  SPEAKER_05: So they have to like pick these tests very carefully, um, right, because they involve cost.
• 45:11
  SPEAKER_05: Yeah.
• 45:11
  SPEAKER_05: So so so evaluating AI is not a costless exercise, but it's necessary in order to be able to understand in a way you can use it.
• 45:20
  SPEAKER_05: So which brings me so so so so so.
• 45:23
  SPEAKER_05: I guess to sort of bring it full circle, I'd say ideally we want AI to be such that we just use it without regard to its mistakes, right?
• 45:32
  SPEAKER_05: That we know it'll make mistakes, but we say that's still okay, right?
• 45:35
  SPEAKER_05: Just like we do with humans, right?
• 45:37
  SPEAKER_05: We use humans knowing that humans will make mistakes, but we said that's okay, right?
• 45:40
  SPEAKER_05: You will make mistakes, right?
• 45:42
  SPEAKER_05: So that's the equivalent scenario we need in AI is we have a machine, you have a human.
• 45:47
  SPEAKER_05: You know what?
• 45:48
  SPEAKER_05: We're better off with using the machine even with the mistakes it makes.
• 45:52
  SPEAKER_05: Yeah, that's the ideal scenario.
• 45:54
  SPEAKER_00: So it's almost like a world where there's a human-AI collaboration, right?
• 45:59
  SPEAKER_00: That's really how the world would change slowly.
• 46:03
  SPEAKER_00: Where I have my assistant, uh, some directions are being given by my AI assistant, I then take those decisions, but then I apply my context on top of whatever knowledge that I have, and therefore, this whole ability to work with AI and humans and collaborate uh intuitively is something that we need to get ready for.
• 46:26
  SPEAKER_05: Exactly.
• 46:27
  SPEAKER_05: And and it'll probably happen in phases where let's say you you tell the machine, hey, generate me a marketing plan for this new client.
• 46:35
  SPEAKER_05: Here are the specifics of the client, use prior engagement information and generate me a business plan or a proposal, right?
• 46:43
  SPEAKER_05: And the machine will do it for you, but you as the uh specialist now still need to go and tweak it and look at it and everything, right?
• 46:50
  SPEAKER_05: That's fine.
• 46:50
  SPEAKER_05: I imagine you get to a point where you don't even need to do that, right?
• 46:53
  SPEAKER_05: That it gets so good that you that you just trust it, right?
• 46:57
  SPEAKER_05: Or you've got to do like very little work, right?
• 47:00
  SPEAKER_05: And then you get to the point where it just is so good that you just trust it anyway.
• 47:05
  SPEAKER_05: Uh so we'll we'll probably see that trajectory of adoption.
• 47:10
  SPEAKER_05: You know, it's what I call sort of the automation frontier, where you know things go from being purely manual to being you know mostly automated, but you still need the human in the loop, and then it becomes so good that you know it makes fewer mistakes and they tend to be not that serious.
• 47:26
  SPEAKER_05: So you say, well, let the let the machine just do everything.
• 47:29
  SPEAKER_00: Which really brings me, uh Professor Asander, a question where when we move from an industrialization era, okay, to the internet era, to now maybe the AI era as we call it, the industry industrialization era almost made humans into machines, right?
• 47:49
  SPEAKER_00: Uh, here is a blueprint, just get me this work done, and we produce mass products and we then ship it out.
• 47:57
  SPEAKER_00: And now, and before the industrialized era, there were craftsmen who knew their uh you know craft.
• 48:05
  SPEAKER_00: Now, when you come back to the AI era, uh the when I need an AI human collaboration, it has to the craftsmanship has to come back because I basically have a machine helping me, but I need to know my uh you know uh my my skill has to be extremely good.
• 48:24
  SPEAKER_00: I must be an expert.
• 48:26
  SPEAKER_00: I need to know what to prompt that machine, which means that my skill and knowledge and competency is becoming more and more important vis-a-vis what it was in the industrial era or the internet era.
• 48:37
  SPEAKER_00: Would you agree with me on that?
• 48:38
  SPEAKER_05: So you raise a really interesting question, right?
• 48:41
  SPEAKER_05: And um, I I I wrote this article called The Day I Chased My Dog, where I, you know, I had an incident where I gashed my knee because of some accident.
• 48:53
  SPEAKER_05: I had to go to the emergency room.
• 48:55
  SPEAKER_05: And I talk about how at the moment we have the worst of both humans and machines.
• 49:01
  SPEAKER_05: You know, you talked about the fact that the industrial era made robots out of humans, right?
• 49:06
  SPEAKER_05: Humans started doing repetitive work.
• 49:09
  SPEAKER_05: Ironically, it's not that much different now in healthcare.
• 49:14
  SPEAKER_05: You know, the way I felt like I'd been through an assembly line, you know, that six people saw me, they treated me like an object, they asked me a bunch of standard questions, many of them were repetitive, and then by the end of it, by the time I'd seen the sixth person, I finally saw the specialist, the doctor.
• 49:31
  SPEAKER_05: You know, but by that time I was so overwhelmed with information and choices and decisions that I wasn't even in a mood to ask questions.
• 49:41
  SPEAKER_05: Uh so uh in a sense, I got the worst of both worlds, right?
• 49:46
  SPEAKER_05: And then, you know, they they they put staples in my knee, and then 10 days later I went and got those removed and the uh the the wound opened up.
• 49:54
  SPEAKER_05: You know, it was a it was a colossal mess.
• 49:57
  SPEAKER_05: Um, and I wrote about the fact that gee, wouldn't it have been nice to have a machine actually look at my knee and say, you know what, like you should wait a few more days.
• 50:05
  SPEAKER_05: It it looks a little risky to open it up, you know, and take these staples out.
• 50:09
  SPEAKER_05: It might actually open up, right?
• 50:10
  SPEAKER_05: But the human didn't do that, he just sort of followed a procedure and took the things out and it opened up, right?
• 50:16
  SPEAKER_05: So at the moment, we're sort of in this phase where we're getting the worst of both.
• 50:20
  SPEAKER_05: Yeah.
• 50:21
  SPEAKER_05: You know, when it comes to healthcare, right?
• 50:23
  SPEAKER_05: We're getting humans behaving like robots, not like the old days where someone really cared about you.
• 50:28
  SPEAKER_05: You know, when I was a kid, I went to the doctor, I went to the same doctor, they cared about you, right?
• 50:33
  SPEAKER_05: And now the care has really gone out of the healthcare system.
• 50:37
  SPEAKER_05: You know, you're seeing strangers all the time who are looking at you sort of atomically as a little part of a larger process, yeah.
• 50:46
  SPEAKER_05: And there isn't this sort of intelligence of the process that's coming to the picture and saying, hey, I think this is what we need to do to offer you know Swami the best possible uh you know treatment for his injury.
• 50:59
  SPEAKER_05: Yeah.
• 51:01
  SPEAKER_05: Uh but you know, I'm I'm an optimist.
• 51:03
  SPEAKER_05: I think we'll get there, but we're just not there yet because what we've done is we've sort of made the process mechanistic, atomic, uh impersonal, and it needs to be just the opposite, where we get the best of uh both worlds.
• 51:18
  SPEAKER_05: But we're far from that yet.
• 51:19
  SPEAKER_05: But that I see as the biggest challenge going forward.
• 51:22
  SPEAKER_00: True, true.
• 51:23
  SPEAKER_00: So which means that uh if I were a youngster listening to this podcast, and I have another you know, 30-40 years of career ahead of me, uh, how do I uh you know up my thinking skills to be prepared for this AI era?
• 51:41
  SPEAKER_05: Oh, that's a great question.
• 51:43
  SPEAKER_05: You know, I I I uh I was talking to my students the other day, right?
• 51:47
  SPEAKER_05: And they're like, I don't know, between this is undergraduates, like they're between 18 and 23, and you know, my graduate students are in the late 20s to early 30s.
• 51:56
  SPEAKER_05: You know, and my um uh you know observation is that you know you guys have your lives ahead of you, and this is such an exciting time, you know, to be in this space.
• 52:06
  SPEAKER_05: Like, you know, I I've like you know created three sort of AI startups in my career, but I did it during a time when people were resistant to AI.
• 52:16
  SPEAKER_05: You know, I mean I did it anyway.
• 52:17
  SPEAKER_05: I've already had you know machine learning to Wall Street, and people say, What the hell is this?
• 52:21
  SPEAKER_05: Like what?
• 52:22
  SPEAKER_05: Like what sense does it make?
• 52:24
  SPEAKER_05: So but I did it anyway because I believed that there was a lot of merit to it and it was the future.
• 52:29
  SPEAKER_05: Whereas now the world is so receptive to it, right?
• 52:32
  SPEAKER_05: So there are so many possibilities uh you know, in terms of like you know, creating better data sets to do something, you know, because there's often value in creating data that can be used to do something useful, right?
• 52:46
  SPEAKER_05: And you know, and you talked about big tech as having data, but uh the the world is wide open in terms of the kind of data sets you can collect to do useful things.
• 52:55
  SPEAKER_05: Yeah, you know, so you know what I tell people is yeah, up your thinking in terms of like uh you know where you think AI, for example, is going to transform an industry or the way people do things, and then you know, start start you know collecting the data to make that possible.
• 53:16
  SPEAKER_00: So, which means that uh uh I've been trained as uh a student over years thinking mechanically, right?
• 53:27
  SPEAKER_00: Suddenly, when I come to the AI era or when I start to work and I'm collaborating with machines, I'm collaborating with data, you really talk about in most of your articles the fact that contextualization of the data becomes very, very important, and then suddenly my critical thinking skills, my soft skills, my ability to understand the feeling and the uh nuances of introspection becomes very important, and therefore, if I were a youngster, I would say, hey, uh, you know, working in ambiguity, but clearly ability to, you know, up my thinking skills and how I need to apply this becomes very important because you are not taught in many educational institutions today.
• 54:09
  SPEAKER_05: Indeed, indeed.
• 54:10
  SPEAKER_05: Uh I I I sort of look at this abstractly as the ability to conduct inquiry.
• 54:16
  SPEAKER_05: Right?
• 54:17
  SPEAKER_05: I mean that that's what it really boils down to is can you formulate an interesting question and then answer it?
• 54:26
  SPEAKER_05: Right?
• 54:27
  SPEAKER_05: And there are so many interesting questions out there that one could think about.
• 54:32
  SPEAKER_05: You know, uh I don't know, if I were to like just just dream, I'd say, you know, like, you know, what about assistance that uh, you know, for mental health or something, right?
• 54:42
  SPEAKER_05: Where they're they're actually able to, you know, increase your brain function and make you feel better and you know be more effective, right?
• 54:51
  SPEAKER_05: Can you actually design machines that make humans better at what they do or they make people happier, right?
• 54:59
  SPEAKER_05: So how would you even ask such a question and how would you gather the data to be able to answer it to show that, oh yeah, under these conditions, actually this kind of a machine does make people better or happier or whatever, right?
• 55:11
  SPEAKER_05: So it becomes a it becomes a question that is an interesting one because the implications are uh you know incredibly important.
• 55:20
  SPEAKER_05: But then can you then uh you know, do you have the chops to be able to answer that question?
• 55:25
  SPEAKER_05: In the old days you required a PhD to do this kind of thing, right?
• 55:28
  SPEAKER_05: This this was like you know, PhD material.
• 55:31
  SPEAKER_05: But now it's like you know, it's almost like everyone needs the ability to do inquiry, which means ask great questions and then to be able to work with data to be able to answer them, right?
• 55:43
  SPEAKER_05: And so, you know, this is also one of my themes is that what AI will force humans to do is make us better, right?
• 55:50
  SPEAKER_05: So it'll either sort of eliminate us, but those of us that doesn't eliminate, it'll force us to be better at what we do, right?
• 55:59
  SPEAKER_05: Which has been the trend of technology anyway.
• 56:01
  SPEAKER_05: Like a few hundred years ago, you were a genius if you knew how to do arithmetic, right?
• 56:05
  SPEAKER_05: Uh you know, you were special, but now you're not, right?
• 56:07
  SPEAKER_05: So it keeps technology keeps upping the game, and AI will continue to do that.
• 56:13
  SPEAKER_00: So, which brings me to a very important question, uh, Professor, uh, where uh if I were to look at AI as a career, uh, wherever uh you know, and whatever articles I'm reading, uh you need to have good knowledge of computer science, psychology, and neuroscience.
• 56:33
  SPEAKER_00: Right?
• 56:34
  SPEAKER_00: Is that the combination that probably the best AI uh solutions will be built around, and therefore I need to get prepared for those kinds of skills.
• 56:45
  SPEAKER_05: You certainly do.
• 56:47
  SPEAKER_05: I mean, I I think those are good skills to have, but I'd go even broader than that, right?
• 56:55
  SPEAKER_05: I think I think that I think that EQ is as important, right?
• 57:00
  SPEAKER_05: It isn't just about the hard skills, it's also about understanding emotions and emotional uh IQ and quotient.
• 57:08
  SPEAKER_05: So it's a it's a broad kind of problem.
• 57:11
  SPEAKER_05: But you know, it's interesting that you're asking me uh the question of like, you know, if I want to do a career in AI, right?
• 57:19
  SPEAKER_05: Because I'm asking myself, like, what does that even mean?
• 57:22
  SPEAKER_05: You know, I mean, like 30 uh 10 years, 20 years, 30 years ago, it actually meant something, right?
• 57:26
  SPEAKER_05: That you were exploring a bunch of methods and that you were at the forefront of you know uh you know innovation in terms of methods for like you know, let's say representing and using knowledge.
• 57:39
  SPEAKER_05: Um but what does it mean going forward?
• 57:41
  SPEAKER_05: Like and the answer is like I don't know.
• 57:43
  SPEAKER_05: It's like it's like it's like asking, you know, when electricity came out, you know, I want to study electricity, you know, I want to be a specialist in electricity.
• 57:51
  SPEAKER_05: Well, yeah, that's a good idea, but what does that mean, right?
• 57:55
  SPEAKER_05: Does that mean you're you know designing motors and dynamos and washing machines and televisions, right?
• 58:01
  SPEAKER_05: The things that use electricity is is what you're really getting into.
• 58:06
  SPEAKER_05: And that's sort of the way I'll use the the way I look at AI now, that you know, that some people will continue to innovate and do research in AI itself, right?
• 58:16
  SPEAKER_05: That that's sort of a core research community, but there's just this large ecosystem around it in terms of its applications that's wide open, also as a career to people who can then sort of build on AI to do whatever else they want to do.
• 58:32
  SPEAKER_00: True, so true.
• 58:34
  SPEAKER_00: And uh uh I kind of reflect on the point that you said a bit earlier.
• 58:39
  SPEAKER_00: Uh, when I look at customer service, how do I look at customer service in a machine era and customer service in an AI era?
• 58:46
  SPEAKER_00: Right?
• 58:47
  SPEAKER_00: Right.
• 58:47
  SPEAKER_00: The customer service in the uh you know internet era was all about I have CRM systems, I have data, I'll have automated emails, uh, you know, I assume that there is a journey, but humans don't follow a pattern around journeys, you know.
• 59:01
  SPEAKER_00: AI needs to then the way you think of customer service has to be very, very different with your AI models, with the context, with uh you know how my emotions are, and therefore, if I want a bank, uh I need to look at uh you know how this data uh gets first generated, or how do I interpret the data that's not there yet use small data sets to interpret what is going on for better customer service.
• 59:30
  SPEAKER_00: So the way you need to think around this problem is very different from the way you probably were thinking even five years back.
• 59:38
  SPEAKER_05: Exactly.
• 59:38
  SPEAKER_05: I mean, and and and if you think think about this abstractly, it's this trend towards like incredible personalization, right?
• 59:45
  SPEAKER_05: That's what intelligence really gives you, right?
• 59:47
  SPEAKER_05: You went you went from sort of standard processes, you know, show up at bank hours between you know 10 and 3, you know, and and you know, take a token and shut up and wait for your service, right?
• 59:59
  SPEAKER_05: To CRMs, where now you get an email, right?
• 1:00:02
  SPEAKER_05: But that's still mechanical, right?
• 1:00:04
  SPEAKER_05: To the point where you know the machine really sort of cares about you and says, hey Swami, like you know, like where have you been?
• 1:00:12
  SPEAKER_05: Yeah, you know, the last are you alright?
• 1:00:14
  SPEAKER_05: You know, you you normally come on every, you know, but uh right?
• 1:00:18
  SPEAKER_05: So it becomes personal and sort of really one-on-one because it can, it has the data and the intelligence to be able to do that.
• 1:00:24
  SPEAKER_05: So yeah, absolutely.
• 1:00:25
  SPEAKER_05: That's kind of the brave new world that we're heading towards.
• 1:00:29
  SPEAKER_00: Thanks.
• 1:00:29
  SPEAKER_00: Um, so I'm going to uh end this conversation with a couple of uh quick five questions to you.
• 1:00:35
  SPEAKER_00: Uh so the first question that I have for you is uh what does success mean to you?
• 1:00:42
  SPEAKER_05: Uh being happy, being did like spending my time doing exactly what I like to do.
• 1:00:47
  SPEAKER_00: Okay.
• 1:00:48
  SPEAKER_00: Uh that's a brilliant question, brilliant answer for this question around what is success.
• 1:00:54
  SPEAKER_00: Uh, if you were to invite uh a couple of people for a dream dinner, uh who would be the people that you would invite and why?
• 1:01:06
  SPEAKER_05: Oh wow, that is like you know, that's so hard, uh Swami, because I don't know.
• 1:01:11
  SPEAKER_05: I mean, I there are literally a few hundred people I would invite of all ilks, right?
• 1:01:17
  SPEAKER_05: Ranging from you know hardcore scientists.
• 1:01:20
  SPEAKER_05: Uh, you know, for example, I just had a podcast with Peter Ward, who's a paleontologist, you know, who's written about life on Earth, you know, just something just mind-boggling in terms of you know our the history of life, you know, or I've talked to people about free will, you know, does free will exist?
• 1:01:38
  SPEAKER_05: Um, you know, I I don't know, there's there's a there's a whole bunch of scientists, there's a whole bunch of wackos, you know.
• 1:01:44
  SPEAKER_05: I I mean I'd love to have you know Elon Musk and you know, like just like yeah, just crazy intelligent, smart, wacky people like that.
• 1:01:53
  SPEAKER_00: I'd love to have so you pretty much uh you pretty much will have a large party than a dinner.
• 1:01:59
  SPEAKER_05: Exactly.
• 1:02:00
  SPEAKER_05: It would be it would be a large party because I'm I'm I guess I'm greedy, right?
• 1:02:04
  SPEAKER_05: I want to call academics, politicians, economists, you know, policymakers, you know, innovators, you know, I'd I'd call a few of each uh each uh category, you know, whoever whoever is alive and healthy and you know kicking.
• 1:02:18
  SPEAKER_05: Uh so yeah, sorry, there wasn't a simple answer to that one.
• 1:02:21
  SPEAKER_05: Sorry about that.
• 1:02:21
  SPEAKER_00: I I I I think it was an honest and a brilliant answer, uh, to be to be uh very, very honest.
• 1:02:28
  SPEAKER_00: That's really what I got.
• 1:02:29
  SPEAKER_00: Uh if you were to recommend uh a couple of books that uh you know you would uh want anybody to read, uh what are the things that couple of books that you were inspired by?
• 1:02:42
  SPEAKER_05: You know, there's there's too much of a recency bias when you ask me about books, right?
• 1:02:47
  SPEAKER_05: Because I've like forgotten about you know the old books, you know, like Brave New World, you know, which I read in college, right?
• 1:02:54
  SPEAKER_05: Which is I would you know I would definitely recommend Brave New World, you know.
• 1:02:59
  SPEAKER_05: Um you know, going way back, you know, Death of a Salesman, you know, I loved Arthur Miller, you know, that's why I I loved uh you know that kind of literature.
• 1:03:08
  SPEAKER_05: I loved, I don't know, like a lot of American literature I grew up with.
• 1:03:13
  SPEAKER_05: You know, more recently, you know, I read Gandhi's My Experiments with Truth that I thought was fascinating.
• 1:03:18
  SPEAKER_05: I hadn't read that book, surprisingly.
• 1:03:20
  SPEAKER_05: You know, I'm an Indian, I should have read it, you know, read so much about Gandhi, but I hadn't read that.
• 1:03:24
  SPEAKER_05: So I thought that was really cool, and I you know was sort of seeing the connection to truth and AI.
• 1:03:31
  SPEAKER_05: Um yeah, I mean I could uh I could go on.
• 1:03:35
  SPEAKER_00: Perfect, perfect.
• 1:03:36
  SPEAKER_00: And uh uh if there's one great piece of advice that you got, uh and I'm sure you would have got a lot, uh, what is it one piece of advice that you got which you remember even today?
• 1:03:51
  SPEAKER_05: Um you know it's it's funny you asked that because I, you know, I said I'd spent several years on Wall Street, and I was thinking of not coming back to academia and just sort of switching careers.
• 1:04:02
  SPEAKER_05: And I was having dinner with a colleague of mine who's now passed away.
• 1:04:06
  SPEAKER_05: Um and uh I said, you know, it's it's just so hard to ignore the fact that I'd be making like 20 times what I make as a lowly professor.
• 1:04:16
  SPEAKER_05: And he and he asked me, he says, Do you have enough?
• 1:04:18
  SPEAKER_05: And I said, Yeah, I got enough.
• 1:04:20
  SPEAKER_05: And he says, Well then do what you like, do what's fun.
• 1:04:25
  SPEAKER_05: So I'll never forget that.
• 1:04:27
  SPEAKER_05: It was a simple question, do you have enough?
• 1:04:28
  SPEAKER_05: And I was like, Yep, I have enough.
• 1:04:30
  SPEAKER_05: And this has come up in sort of some of my earlier podcast conversations I had with John Sexton, you know, where I said, you know, what advice do you give people?
• 1:04:37
  SPEAKER_05: Like, you know, like and he said, You know, if you keep your needs low, you should you can do whatever you you should pursue whatever it is, you know, just keep your needs modest.
• 1:04:46
  SPEAKER_05: Um and so that was good advice.
• 1:04:49
  SPEAKER_05: Keep your needs modest.
• 1:04:51
  SPEAKER_05: I read that in your newsletter.
• 1:04:53
  SPEAKER_05: So that was a and and and and and so and and I might just put another one in by Michael Roth, who's the president of Wesleyan.
• 1:04:59
  SPEAKER_05: Like I asked him, I said, What's the purpose of higher education?
• 1:05:02
  SPEAKER_05: And he said, It's to maximize serendipitous encounters.
• 1:05:06
  SPEAKER_05: Beautiful, you know, which I thought was great, right?
• 1:05:09
  SPEAKER_05: The role of education of higher education is to maximize the probability of serendipitous encounters.
• 1:05:14
  SPEAKER_05: So put yourself out and be open to serendipity.
• 1:05:18
  SPEAKER_05: You know, there's another um quote that goes something like luck favors the prepared, right?
• 1:05:24
  SPEAKER_05: So and I think serendipity prepare uh favors the people who are sort of receptive to it.
• 1:05:31
  SPEAKER_05: So beautiful, beautiful.
• 1:05:32
  SPEAKER_05: Those are a couple of uh interesting quotes and pieces of advice that I've heard that really resonated.
• 1:05:38
  SPEAKER_00: Beautiful, beautiful.
• 1:05:39
  SPEAKER_00: And what's one thing that you believe that others don't agree with you?
• 1:05:52
  SPEAKER_05: Oh something that I believe that others don't agree with me.
• 1:05:57
  SPEAKER_05: Uh the only thing you know, I I I can think of something, but it's too embarrassing because it's like it's become socially unacceptable to talk about this.
• 1:06:08
  SPEAKER_05: Um yeah, it it has to do with why uh there are fewer swimmers in certain racial groups than others, right?
• 1:06:19
  SPEAKER_05: So, my answer to that is I find uh some people find it, you know, they don't agree with me, but I think that's because they find it socially unacceptable.
• 1:06:28
  SPEAKER_05: But I won't get into the details because that'll really get me into trouble.
• 1:06:31
  SPEAKER_05: Yeah, yeah, sure, sure, sure, sure.
• 1:06:33
  SPEAKER_00: So uh thanks, Professor Vasender.
• 1:06:35
  SPEAKER_00: It was lovely talking to you, and uh it was a fun conversation.
• 1:06:40
  SPEAKER_00: Uh really uh the time was brilliant in terms of the perspective that you gave.
• 1:06:47
  SPEAKER_00: Some of the examples were thoroughly uh brilliant, and thanks for your time.
• 1:06:53
  SPEAKER_00: Thanks for uh sharing your thoughts on the Contramines podcast.
• 1:06:58
  SPEAKER_05: I am delighted to have been here, Swami.
• 1:07:02
  SPEAKER_05: Um, you know, and I and I I'll just end by um noting something, which is that you know, some of my colleagues make uh fun of me because I believe that names are very important, you know.
• 1:07:14
  SPEAKER_05: That you know, so certain names are really important.
• 1:07:17
  SPEAKER_05: Uh you know, some names are strong, some names are weak, it's you know, whatever.
• 1:07:21
  SPEAKER_05: But I think Swami is a brilliant name, so I am so envious.
• 1:07:27
  SPEAKER_00: Thanks a lot.
• 1:07:28
  SPEAKER_00: Thanks a lot.
• 1:07:29
  SPEAKER_05: Because I can imagine people saying, so Swami, what do you think?
• 1:07:35
  SPEAKER_05: It's it's been a pleasure.
• 1:07:37
  SPEAKER_05: I really enjoyed it, and thanks for all those questions and uh and follow ups.
• 1:07:41
  SPEAKER_00: Thanks, thanks a lot.