Artificial Intelligence, Big Data & China | Martin Schmalz | Modern Wisdom Podcast 144

Martin Schmalz is a professor of Finance at Oxford University and an author. We're receiving constant warnings about the advent of Artificial Intelligence. And big data. And China. But how do all of these fit together? Expect to learn why your phone's GPS data on a night time is affecting your credit score, how the speed which you complete an online form in could change the price, where the REAL computing power behind AI is being deployed at the moment, and much more. Extra Stuff: Follow Martin on Twitter - https://twitter.com/martincschmalz Buy The Business Of Big Data - https://amzn.to/2HHg2Li Thank you to The Browser - https://thebrowser.com/ Take a break from alcohol and upgrade your life - https://6monthssober.com/podcast Check out everything I recommend from books to products - https://www.amazon.co.uk/shop/modernwisdom #bigdata #artificialintelligence #machinelearning - Listen to all episodes online. Search "Modern Wisdom" on any Podcast App or click here: iTunes: https://apple.co/2MNqIgw Spotify: https://spoti.fi/2LSimPn Stitcher: https://www.stitcher.com/podcast/modern-wisdom - Get in touch in the comments below or head to... Instagram: https://www.instagram.com/chriswillx Twitter: https://www.twitter.com/chriswillx Email: [email protected]

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Speaker 00

  People start sleeping in two different locations interchangeably at night. That tends to be a really bad credit risk. So these people tend to use up a lot of cash in the near future more than they can afford. And the story they tell me behind that is, well, those are people who have lovers. And having lovers leads to divorces and divorces are costly. And costly divorces lead to loan default. So whether that's the right story behind it or not, the boring fact is that using location data is extremely useful in predicting default.

• 0:32Chris Williamson?

  Martin, how are you doing, man? Welcome to the show.

• 0:34Speaker 00

  Very good. Thank you very much for having me.

• 0:36Chris Williamson?

  Very, very happy to have you on today. We've recently been talking about some big data stuff. Seth Stephens-Savidowitz was on recently and we were discussing about some of the interesting analysis that he'd done on Google searches and Pornhub data as well. Maybe no Pornhub today, but definitely some big data from yourself.

• 0:55Speaker 00

  That's right, yeah.

• 0:56Chris Williamson?

  Lovely. So give us your background. What do you do?

• 0:59Speaker 00

  Well, my background, I grew up in southwest Germany. And as everybody does who is from there and has any form of self-respect, I studied mechanical engineering. But at some point, I had the impression that I can much better understand what happens in the world if I study how the financial system works. And that's made my way to studying economics and going to the US and ended up being a finance professor. And in the course of that, I somehow stumbled across this topic of AI and big data and started teaching it because I somehow felt that there was a bit of a discrepancy between the demands on our graduates in industry, you know, which concerns Python and big data

• 1:42Speaker 00

  skills and what we taught them, which at the time was largely Excel. So I developed the ambition first to actually make MBA students teach, learn some Python in class and applied. And of course, the ambition is not to turn into data scientists, but to understand what the economics of data-driven businesses models is and can be and how we can understand the success of tech platforms over the last decade or so.

• 2:11Chris Williamson?

  So quite involved in the development of how we analyze the data and pushing that forward.

• 2:18Speaker 00

  Yes, so see, there are specialists on the analysis of data and you call them data scientists or so. What I'm trying to spend time thinking about is predicting the future directions of business models and the development of workplaces and yeah, just how jobs, firms and industries get transformed as a result of the of the big data revolution.

• 2:46Chris Williamson?

  Okay, so this is helping businesses to make decisions through big data.

• 2:51Speaker 00

  That's right, that helps that helps businesses make strategic decisions on what the AI revolution means for them, how they should position themselves. It also helps investors in deciding what kind of businesses they should invest in or what kind of questions they might ask businesses they consider investing in and distinguishing between the, you know, thousands of different fintech startups who are kind of figuring out which ones of those that just have AI in the name, which ones actually apply it and which ones apply it in a way that doesn't only solve a technical problem, but actually also has a decent chance of turning a profit at some point in the future.

• 3:27Chris Williamson?

  I understand. So is most of the challenge that you're coming up against here a technical one with regards to the way that you can statistically model things and the the scripts and things that you can write? Or is it more so on the side of how you interpret that data, how you apply it to the market and stuff like that?

• 3:49Speaker 00

  The latter. So the challenge is that there are very few people that have the combination of two skills or sets of knowledge. One is understanding what AI actually is and what machine learning algorithms do and specifically what they don't do. And the spoiler alert is if you read the newspaper, you don't really get a good impression of what that is. And there are, of course, people who understand that, like the engineers that or data scientists that work with machine learning models. The problem with that is that very few of them have even basic economics training or enough of economics

• 4:28Speaker 00

  training to kind of combine their data science skills with economic theory in order to predict the future of industries and markets. So there doesn't seem to be a lot of structured thought about business models and how they will change in the age of AI. And indeed, there was no book out there, which is why we wrote one.

• 4:50Chris Williamson?

  Yeah. So that, by the way, business of big data will be linked in the show notes below, of course, if you're interested in today, then head to the link there and you'll be able to find out even more. Why do you need someone whose skill set straddles both areas? Why can't you just have the data science people feed out the data that the guys who understand economic theory and AI, like why can't you have them just work in tandem as opposed to to have someone that bridges?

• 5:20Speaker 00

  Well, well, I suppose you can have them work in tandem as long as they can communicate to each other. But that is a challenge. So if a data scientist talks to an economist who doesn't know what a feature is, that is a problem. If you explain to an economist that what the data scientist called feature, you call a variable, then all is good. But somebody needs to be there who translates between these two different worlds. And that's just, that's just a challenge that that many firms have. So what happens if you don't have such a person? See, if you read the Financial Times or whatever, the business press, you might get the impression that AI is about making computers to think and to replace human beings, and that the whole world is investing it. So what do you do

• 6:05Speaker 00

  as a top executive trying to allocate capital to various projects? You might call up the IT department and tell them, hey, why don't you, you know, do AI, do big data?

• 6:16Chris Williamson?

  Yeah.

• 6:16Speaker 00

  At which point the IT department says, Well, yes. And then they build a data pool and two or three years later, they ask themselves, so why exactly did we do that? And how does it fit into overall strategy? And what's the product? Who's the customer? What produces the value in this in this case? Or, you know, the problem here being that in many cases, top management doesn't really understand what is at the core of AI and machine learning. On the other hand, engineers tend to get very excited about solving technical problems. And the technical problems ML solves is prediction. You predict some stuff. But then again, there's many things you could predict and there's a lot of data you could use

• 6:57Speaker 00

  for prediction. But ideally, that should be informed in a business that strives to make a profit by what economically important and valuable parts of prediction are. And this is where the communication between the engineers and the economists or the strategists has to come in. And that communication suffers from just, you know, basically language barriers or or lack of knowledge on both sides about what the other side is doing.

• 7:21Chris Williamson?

  I get it. I get it. So what does AI do? You've mentioned some of the things that it doesn't do. What's AI being used for mostly at the moment, other than preparing for a global takeover? And to make all of our jobs obsolete?

• 7:34Speaker 00

  Well, there you go. Exactly. So that's the kind of thing you read about, right? So Elon Musk talks about the singularity and stuff. And my point is not so much about whether that's a real thing and whether it might happen in 30 years. The point is that 95% of currently used ML and AI techniques is basically about a pretty boring technical problem, which is you use a bunch of data to predict some sort of outcome variable. So what could that be? I know everything about, you know, your listeners' Pornhub watching habits, as well as their shoe size, hair color, social network, phone records, and I don't know, a whole bunch of different variables. As well as their heart rate variability at night between two and three AM, all their sleeping habits and exercise habits.

• 8:19Speaker 00

  And I might use all that information to predict stuff. What I could I possibly predict? Well, I could want to predict, obviously, what kind of ads they're likely to click on. So what kind of products they like, what kind of movies they like to watch. But not only that, but also how much are they willing to pay for it? So if I know that you have a certain travel schedule, because I tracked your cellphone location data in a precise way, I might be able to start to infer how much you're willing to pay for airline tickets on a certain route at particular times. And obviously, that information gets valuable if you manage to exploit it in a profitable way.

• 8:56Speaker 00

  So what was the question?

• 9:00Chris Williamson?

  What does AI do and what does it not do?

• 9:02Speaker 00

  Right, so what AI does is a machine prediction that is generic in a sense of you have a bunch of data from the past and in contexts where the past predicts a future or the future works according to similar rules as the past.

• 9:21Speaker 00

  The generic types of prediction machines can do faster, better, and cheaper than human beings, and increasingly so, simply because there's more and more data around, more data than a human could possibly analyze. Computational power gets cheap. These algorithms are very computationally intensive. But yeah, that stuff got cheap. So data collection, data storage, and computation got cheap. And as a result of it, machine prediction got cheap at a given level of quality. That's what it does. And it doesn't do anything else, really. Computers, computers don't think. Like, artificial intelligence has very little to do with intelligence in a broader sense. So it is true that a pretty big part of human intelligence these days

• 10:04Speaker 00

  is being used for generic prediction. But that's exactly what will change.

• 10:08Chris Williamson?

  What like?

• 10:10Speaker 00

  What like?

• 10:12Speaker 00

  Well, there used to be a job description called loan officers. Their job is to know your dad and your your mother, your friends, and your personal character as evidenced on the Newcastle soccer team or whatnot. And from all this information, kind of predict how likely you are to repay a loan, and hence whether they should extend the mortgage to you and at what interest rate and which collateral. That is a prediction exercise. The thing is, nowadays, there are companies that know much better what your social network is, and whether your dad and sister and social network repaid loans, and what your current health and mental health status is. And those companies are called tech companies or data aggregators and so forth. So they are just becoming much better and cheaper and faster at predicting loan default.

• 10:57Speaker 00

  What else might you predict? You might predict health outcomes, you know, or the the life expectancy. That's kind of useful when you run a life insurance company.

• 11:09Speaker 00

  You might, as I said, movie watching habits, a lot of things. If you drive a car, you might predict whether the cyclist you see in the right of the road is likely to make a left turn anytime soon and get in your lane or not. All of these are prediction exercises. Yeah, once you spend some time thinking about this topic, you see that prediction exercises are all over what we do in real life these days. And some of them are generic and based on a whole bunch of data and a whole bunch of experience, and those are going to get replaced. And others are non-generative. They're not based on a whole lot of data, but are like fundamentally new. Here's an example. What about predicting how AI will disrupt your job?

• 11:53Speaker 00

  Or your firm or your industry? The thing is, that's never really happened before. So a computer cannot possibly analyze a past data set when this has happened before and therefore predict what happens now. It has to be a human being, and that's a uniquely human strength.

• 12:09Chris Williamson?

  One of the blog posts that I read last year, which was really illuminating to me, was talking about the role of a top end executive. And they classified a top end executive as a difficult to replace complex decision engine. That was what they said that the absolute top end executives are. And when you look at it from, you know, I appreciate that the listeners haven't read that blog post, but it absolutely blew my mind. You got to think the real value that is added by Tim Cook at Apple or by Elon Musk, it's not particularly his engineering ability, it's not particularly his marketing ability, it's not this, that and the other.

• 12:51Chris Williamson?

  It is their ability to compile a lot of disparate, complex, interwoven variables and make the best decision based on all of those. So that's prediction data, right? That's prediction.

• 13:07Speaker 00

  It is absolutely prediction. I would, however, say it's non generic prediction. In particular, they tend to create new products, no? So what they're trying to predict is what is demand going to be for a phone that doesn't have a keypad? I don't know if you remember that. I'm old enough to remember how the press was all over how there's absolutely not going to be demand and obviously BlackBerry was on the wrong end of this. There's absolutely not going to be demand for a phone where you'd have to key in stuff on a touchpad. Well, how do you predict that? Yeah, it is a prediction exercise, but it's not that a computer can do because it's not, it hasn't happened before in the past. So an AI algorithm, machine learning algorithm cannot analyze the past data set to make that prediction.

• 13:51Chris Williamson?

  Absolutely, yeah. That's the, the, the safety mechanism, I suppose, that is inherent in being an executive. Being a top end executive allows you to protect your own job by being the only person on the planet that knows that very particular set of variables and is able to extrapolate forward from that and say, this is what I think that we need to do. And I suppose that's where some people talk about business as an art form or being an exec as a, as a CEO, board member as being, being a creative artist in one way or another. And it is that, isn't it? It's taking all of these different variables and seeing something that potentially isn't in the data.

• 14:30Speaker 00

  So I agree with that. Now, let me.

• 14:35Speaker 00

  At one level of nuance, it's not so much about a job of a top executive, but it's about particular tasks that the executive might take that is more useful to think about. So we right now talked about predicting the demand for a no keyboard cellphone. Well, what about coming up? The creative process of even coming up with the possibility of having a no touch phone? That so far is a rather uniquely human kind of activity that a computer has absolutely no advantage doing. So the kind of, I don't know, this intuitive creativity. I guess you're referring to synthesis as well.

• 15:18Speaker 00

  Seeing a bunch of trends in the world and somehow synthesizing them in an intuitive way in order to make these creative predictions. That is something that humans are uniquely good at, so far at least.

• 15:32Chris Williamson?

  Yeah. Looking at some of the stuff that we were talking about before. Why does my bank care how fast I fill in a form?

• 15:42Speaker 00

  Why does he have to know that or why does he want to know that?

• 15:44Chris Williamson?

  Why does a bank care how fast I fill in a form?

• 15:47Speaker 00

  So see, this is an example right about in the book. I spend a lot of time talking to friends in China. China is, let's say, at least half a decade ahead in the AI and big data game.

• 16:01Chris Williamson?

  Is it really?

• 16:01Speaker 00

  Due to various reasons. And a friend of mine who is a head of AI research at a company called Huawei, that you know recently was in the news once or twice, tells me, yeah, you know, one really important variable these days is how fast people fill out online forms. I say, what? Wait, why? And he says, well, see, first of all, if you type your social security number or your national insurance number, however it's called in the country from which you're listening.

• 16:32Speaker 00

  That kind of tells you a lot about the person. If you can't fill in your social security number real fast and you're probably a fraud or I don't know, not particularly intelligent or something, I don't know. It tells you something about people's willingness or ability to repay a loan or how big of an insurance risk they are. Or similarly, if you make lots of typos while filling out a form, perhaps you're not a particularly careful person. But whatever the theory is behind why these variables predict outcomes well, they appear to work really well in predicting loan default as well as insurance risks. So companies start collecting it. I never thought until I heard this story about that anybody might track

• 17:14Speaker 00

  how fast I fill out online forms, but they do.

• 17:18Chris Williamson?

  Wow, yeah, that is that is super interesting. I was on Skyscanner, a flight comparison site only a couple of months ago. And the particular flight that I was looking at, I watched, I refreshed the page and in front of my eyes, it went up by 300, 300 pounds. And I was like, there's not, there's not this many people looking at this particular flight to Vegas right now. There's me, there's me, and there's the guy sat behind the screen that's just said, I'm gonna, I'm gonna screw him to the tune at 250 quid. And then I went back and I ended up booking the flight to Vegas just in between Christmas and New Year. And I booked it and it had gone back down to below the original price. But it changed as I was looking at it. Refresh the page changed as I was looking at the page.

• 18:03Chris Williamson?

  And I was like, there's a guy at Skyscanner that's just gone, ah, fuck it.

• 18:06Speaker 00

  Well, a guy or a machine.

• 18:07Chris Williamson?

  Or the machine's a guy as well. Yeah, yeah, you are right. There is, there's, there's a guy with a button and the button made it happen.

• 18:14Speaker 00

  Hmm, yeah. So see, I have never, I asked myself for years, why nobody does that to me?

• 18:25Speaker 00

  So you're the first time that the first data point that I hear reporting such an incident. So I don't want to extrapolate too much about it. But see, it's an interesting point, right? So the website can see how interested you are and how desperate you seem to go in a particular date and particular time, no? So what they're predicting is really your willingness to pay for a particular ticket at a particular time. Now, obviously, that can be very useful. But I imagine that you may not have felt super happy about that price increase. And obviously, they increased it so much that you didn't buy the ticket. I might even suspect that you might be less willing to use that particular site in the future if you suspect that they're using your search information against you.

• 19:09Chris Williamson?

  Yeah.

• 19:10Speaker 00

  So when you think about this from a business perspective, what a business executive might want to keep in mind as they seek to exploit the data various users generate is ethical considerations. If not because they themselves care about the ethics, but their customers might, and that might turn out to be a business risk, no? And to get the kind of intuition of what makes people upset, that's kind of like really valuable, no?

• 19:32Chris Williamson?

  Absolutely.

• 19:33Speaker 00

  People get really upset at being charged a different price from another person. But you know what people really love? Getting individualized discounts. So you got a discount, but your neighbor didn't. Now, when you think about this for more than like two or three seconds, you're like, but that's the same thing. Yeah, but it feels very different to people, no?

• 19:50Chris Williamson?

  Absolutely.

• 19:51Speaker 00

  And what we just described, all of the above, both coming up with variables that you never thought about could potentially predict your credit risk, or the intuition about what people get upset about, these are completely human exercises. And it is precisely about predicting the future of AI and how they transform from industries and jobs and websites and searching for flights and so forth, where humans have a huge edge over computers, because it hasn't happened before, you know? So educating oneself about those particular areas is actually a great skill to have as the AI revolution takes a lot.

• 20:26Chris Williamson?

  Am I right in thinking that Amazon split tests its prices? I've heard some, some, some reports about two people going on to the same product and Amazon doing a, whatever it was that you mentioned, desirability to buy or desire to buy, um, split test on some of its prices for things, just by pens?

• 20:45Speaker 00

  So I don't know that. For effect, what I have noticed myself is that when I put a certain item in my basket, in my shopping basket, but I don't immediately buy it and I come back a while later to take a look at it, I'm frequently confronted with the fact that the price of that item may have changed, increased or decreased. Now, I don't know that this is because Amazon is running an experiment on me. What I do know is that this creates data. It does create data about whether I'm buying a certain item at a certain price or whether I'm not, and that can be

• 21:24Speaker 00

  analyzed and used in order to infer individuals' willingness to pay for certain items. Now, again, I can't tell you how much deliberation is behind that, and Amazon would likely not tell me if I asked. But for sure, this is one area where listeners are becoming aware right now that they're generating data that people can go and analyze and use to estimate, you know, predict their future behavior.

• 21:52Chris Williamson?

  Hmm. Why, why does my car insurer care about my email address? Is that the same thing as the bank, the bank filling in a bank form?

• 22:01Speaker 00

  Yeah, kind of, kind of like that. I mean, I'll just ask this to the listeners and, you know, do you know anybody who has an AOL address?

• 22:11Chris Williamson?

  My, my dad did until I forced him to get rid of it and then he, he upgraded to, I think, what he considered to be modern, which was a Yahoo email address.

• 22:22Speaker 00

  Very good. So suppose that I was offering an insurance product that I really don't want to offer to older people. Based on this particular.

• 22:38Speaker 00

  Example that you just offered, I would assume that offering that to AOL and Yahoo users would not be in my best interest because I would be targeting an older audience. So it's not particularly because AOL per se or Yahoo per se are somehow bad or tell you that this person is a bad insurance risk, but being an AOL user correlates with characteristics or features of people that might very much matter for the riskiness of your insurance board.

• 23:14Chris Williamson?

  That's a data point. Inferring from that data point is a particular skew within the data that is indicative of other things and you can use that to

• 23:25Speaker 00

  That's right.

• 23:25Chris Williamson?

  make a judgment. It's starting, it's starting to fit together. What else have we got? Oh yeah. Checking your phone in the morning and where you sleep at night. What about that? Tell us about that.

• 23:36Speaker 00

  So that's another story is China, where people tell me

• 23:39Chris Williamson?

  It's everything's coming out of China, Martin. Everything is at the moment. Fucking Huawei, Huawei now. I mean, what's, I'm recording on a Logitech at the moment. What's that? What's Logitech? Is Logitech Chinese?

• 23:51Speaker 00

  I don't know.

• 23:52Chris Williamson?

  I bet, I bet that it will be. Someone, someone behind the scenes will own Logitech. They'll, they'll have a controlling share from China.

• 24:00Speaker 00

  Uh, well, okay. So you have to update me here on, on what you're after here.

• 24:04Chris Williamson?

  Just Logitech. Who are they? Where are they, where are they from?

• 24:07Speaker 00

  Uh, I, I thought they were Swiss, but I don't know who's actually behind them. So

• 24:12Chris Williamson?

  Well, that's it, China. I'm telling you, I'm telling you, man. Anyway, so China, what time you check your phone and why you sleep at night? Tell us about that.

• 24:20Speaker 00

  Well, so, uh, so, so, so what they tell me is that if people start sleeping in two different locations interchangeably at night, that that tends to be a really bad credit risk. So these people tend to use up a lot of cash in the near future, more than they can afford. And the story they tell me behind that is, well, those are people who have lovers. And having lovers leads to divorces and divorces are costly. And costly divorces lead to loan default. So whether that's the right story behind it or not, the boring fact is that using location data is extremely useful in predicting default. Let's have a less juicy story and just say, see, say you're Uber or Didi when you want to go to China.

• 25:01Speaker 00

  And you know where a person lives because, you know, you have stored a home address in your, in your app, as well as where you work. That gives you a really good idea about what the guy's annual income is. If the person moves to a richer and more prosperous neighborhood, that tells you something about whether their salary end up. What kind of restaurants they eat at, which of course, as a ride hailing company, you know, because that's where you drop them off or pick them up. It tells you a lot about a particular person as well, you know, simply how fancy the restaurant is. So obviously you can infer a lot of things about the person's current liquidity situation from simply analyzing the location data and nothing else. And guess what?

• 25:44Speaker 00

  Didi, a year ago or two, actually started a lending arm. As you would expect from a company that knows so many things about individuals that are relevant for predicting loan default.

• 25:56Chris Williamson?

  Wow.

• 25:58Speaker 00

  Well, see, okay, so, you know, it's, it's fun how I can impress people. The, the way I do this to, in MBA classes is to tell them the economic theory of how clearly a ride hailing company should be offering lending products sometime soon. And I did that in a class and I, I, I earned all the respect by my students because a week later Uber started announcing.

• 26:21Chris Williamson?

  No way.

• 26:23Speaker 00

  Now, the thing is, I didn't actually predict that. I just knew that Didi had done it a year earlier in China, so.

• 26:28Chris Williamson?

  Got you.

• 26:29Speaker 00

  As I said, as I said, it's not very hard to predict what happens in this world as long as you basically follow what's happened in China over the last five years.

• 26:37Chris Williamson?

  Why is China so far ahead with this?

• 26:41Speaker 00

  Well, I think there's at least four reasons. Let's see. We talk about some of them in the book. Let me see if I can still get them together. I mean, number one, I said at the beginning, what machine learning is really useful for is analyzing huge data sets. Otherwise, you can use conventional statistics. Now, China has a large population in a reasonably homogenous economic system. So, well, just by virtue of that, you have a lot of data. But also, for each of these individuals, you have lots of different data points because for various reasons, the Chinese people appear to tend to be less privacy concerned. Or if they're privacy concerned, they just get a lot more convenience out of the apps they have.

• 27:24Speaker 00

  So with the kind of apps that WeChat, say, offers, you can do a lot more things than with WhatsApp in the US, right? So if you think of Mark Zuckerberg trying to merge Instagram with WhatsApp and Facebook data, that's what happened in China many, many years ago. So their messaging app, you can already transfer money. So essentially, you have a useful currency there. You can make your doctor's appointment and you can buy your life insurance. So you can do all kinds of things. Yeah, basically, it's all in the same app. They call it a super app.

• 27:59Speaker 00

  So.

• 27:59Chris Williamson?

  Shit the bed.

• 28:02Speaker 00

  Now, so I said, well, why is it useful to have all these apps within one app? Wait, think about this. I just said you make the doctor's appointment in the same app that you buy life insurance. Well, you know, guess what? You can probably predict the remaining life expectancy of a particular person a lot better if you know what kind of doctor's appointments they're in the process of making, no?

• 28:28Speaker 00

  So this just illustrates that using data that you generate in one part of the business can be really useful in another part of the business. So having all these diverse sets of businesses under one roof, and you might say, why doesn't messaging service, you know, help with merging that with Facebook or Instagram data? That's kind of like the answer. You get different types of variables about the same people. And that turns out to be incredibly useful for predicting, well, all kinds of stuff.

• 28:58Chris Williamson?

  The thing that is coming to mind now, as we talk about this more and more, it's obvious that there's massive amounts of data that are being collected. It's also obvious that some of these data correlates are really useful for companies.

• 29:12Speaker 00

  Right.

• 29:14Chris Williamson?

  That makes them powerful.

• 29:16Speaker 00

  Right.

• 29:16Chris Williamson?

  One of the problems that everyone that's listening can will think of is, right, okay, so let's say my life insurance provider has direct access to my doctor bookings. From the company's side, I can see why that would be great. From the customer's side, that could be great. Or it could leave me suffering at the hands of higher insurance premiums that potentially aren't fair, or anything else. And based on the company's ethics, based on something that's completely out of my control now, I'm at the mercy of this company which knows more about me than I do.

• 29:57Speaker 00

  I couldn't agree more.

• 30:02Speaker 00

  So where does it so I think you mentioned a lot of different things. So one is fairness. Some alarm bells might ring and it might people might find things unfair. Well, the company might say, well, you're free to not use your our app. But the problem is it's kind of hard to be a functional use human being in today's society without any using some sort of technology. I know some people who try. My, my, my cycling buddy from grad school is now a computer science professor at Berkeley, teaching machine learning, and in particular fairness, by the way. And his wife works at Google in an AI team. And he does not have a smartphone.

• 30:42Chris Williamson?

  Man, this is the same as is it?

• 30:44Speaker 00

  That scares me.

• 30:45Chris Williamson?

  Is it Elon, is it Elon Musk or Tim Cook that doesn't let their kids have iPads in the house?

• 30:50Speaker 00

  Yeah, well, I, yeah, you know, it's kind of has a similar spirit to it. Now, here's the thing, I thought about the same because I don't particularly like the idea of my privacy being invaded by, I don't know whom. But on the other hand, I really like to use Google Maps because otherwise I'm having a hard time traveling and getting to places in a reasonable amount of time, you know? So basically, there's a trade off between privacy and convenience, you might call it. An anecdote I'd like to tell is in 2006 or so, the US started taking fingerprints when you entered their country. And there was a huge outcry in Germany about how this is outrageous and how can they and so forth.

• 31:35Speaker 00

  And people should just boycott traveling to the US and so forth. They said, well, that's all very nice, but I just got an offer to go to grad school in the US. I'm actually not going to go to grad school in the US because of fingerprint thing. And obviously, the answer is no. Now, fast forward 12 years later, I travel across country borders like twice a week and use these automatic, you know, scan your passport, scan your face kind of portals. And it goes a lot faster than the old passport control. And I'm perfectly happy about it. So, yes, they use my personal information. They know all kinds of things about me, all my biometrics. I have no idea where the data goes and who uses it. But it's just so much more convenient. And this is basically where things come down on.

• 32:19Speaker 00

  I think we got into this when we talked about China. When I talk to my Chinese friends, they say, well, it's not that we're not concerned about privacy, but it's just so freaking convenient to use these apps. And in this trade-off, we just come down on the side of, well, yeah, I guess, you know, people know everything about me already anyways. So what's the point of me?

• 32:38Chris Williamson?

  Going off the grid.

• 32:41Speaker 00

  Going off the grid.

• 32:42Chris Williamson?

  Yeah, like your friend. Is that the primary benefit from the user side, convenience? Because it seems like, other than convenience, you're at the mercy of the ethics of the company for better decision-making.

• 32:55Speaker 00

  Well, see, so it's not just the ethics, because again, if there's competition between companies, customers might get upset and switch companies that actually protect privacy. And, you know, that's how the whole topic of big data interacts with antitrust law and antitrust enforcement, which we can get into or not. But I'd say, see, from the economic perspective, you can say...

• 33:19Speaker 00

  The companies, the first set that machine learning, machine predictions do is just it lowers the cost of stuff. Uber is just cheaper than a taxi company for various reasons, but one of the reasons is you don't have a human being that is trying to connect a person that calls a taxi, you know, phone hotline with a driver who is currently in some part of town. You just don't have that human being in the loop and you make that process more efficient according to conventional economics. So that's, that's the first benefit. You get just basically the same product, cheaper, better and faster. That's a lot of benefit for the consumer if there's no abuse of, of the information. You then alluded to, oh, but you could do all

• 34:03Speaker 00

  kinds of nasty stuff with the data. Well, yeah, that's the concern that especially looking into the future, customers might be concerned about being price discriminated. So being charged precisely how much they're willing to pay.

• 34:17Chris Williamson?

  Because you know, you know, you've got an indication of where they work and where they live and from that you can deduce a salary and you say, well, this person that's going from a poor area or whatever, we can charge them a little bit less. Or the guy that we know has just had a raise and just bought a brand new half a million dollar house, we can charge him more.

• 34:36Speaker 00

  That's right. Or you might imagine that a ride hailing company might at some point sneakily ask you before showing you a fare for the ride, will you accept this ride no matter what the fare will be? And that actually happened to me.

• 34:50Chris Williamson?

  No way.

• 34:52Speaker 00

  And I was in a rush and just said, yes, but because whatever. And that particular model was in a rush and said, wait, snap.

• 34:58Chris Williamson?

  What the fuck have I just done?

• 34:59Speaker 00

  I'm completely price insensitive. They could have charged me whatever they wanted.

• 35:04Chris Williamson?

  Yeah.

• 35:04Speaker 00

  So that was probably not a particularly smart move on my behalf.

• 35:11Chris Williamson?

  I want to get into the trust side and I guess the litigation side of this as well, because there has to be, well, there might not be, you might tell us that there's not. I would think that there has to be some limits to what companies can do with our data, and I'd love to find out about those. But first, I just want to ask about China. Why is China so much further ahead? Have they got better engineers? Is it just the fact that they've got all of this data to play with? If the US had as much data as China, would they be able to be half a decade ahead?

• 35:42Speaker 00

  Right, so yeah, so let's get back to this topic. So more people, then they have better apps that are more convenient. So people are willing to spend more time on the apps and thus generate data. They manage to collect more features. Then there's basically no privacy or antitrust roadblocks like GDPR there of a comparable nature.

• 36:02Chris Williamson?

  Just the Wild West in China, is it?

• 36:04Speaker 00

  And they do have a ton of engineers who work on this stuff. Like, just one auto or one insurance company, Pingan, has more than 1,000 engineers that do nothing else than programming AIs. And you're going to have to look for an insurance company in the West that pushes this agenda anywhere near as much. If you go to the economic conferences and see not only the number of submissions, but also the fraction of acceptances, you start to see China, Chinese universities just getting way ahead at this point from the West. So it's all of the above. It's all the above that makes China have an edge.

• 36:43Chris Williamson?

  Okay, so antitrust, what is antitrust? What's that?

• 36:46Speaker 00

  Well, antitrust is otherwise known as competition law or the enforcement of competition laws. And this relates to this topic in various ways. One of it is...

• 37:02Speaker 00

  We talked about Facebook previously and their desire to merge WhatsApp, Instagram and Facebook data, which previously they said they would never do, but now apparently they changed their mind. And the German competition watchdog, the Bundeskartellamt.

• 37:17Chris Williamson?

  What a name. What an unbelievable name. Can you say that again? What is it?

• 37:24Speaker 00

  Bundeskartellamt.

• 37:26Chris Williamson?

  Oh my god. That is amazing. It's so badass and German, isn't it?

• 37:32Speaker 00

  They made the following argument. They say, see, it's not illegal to be a monopolist. What is illegal is to abuse a dominant position. And here's the argument they made. They said, see, people care about privacy. If there was a social media network site similar to Facebook that offers a similar benefit, but that is not Facebook, and that actually cares about people's privacy, people would potentially choose that alternative site over Facebook. But such a site doesn't exist. And Facebook does not honor the privacy preferences of the users. Hence, Facebook is abusing the dominant position. So this is obviously paraphrasing what the argument roughly is.

• 38:14Speaker 00

  And they simply prohibited Facebook from merging these diverse data sets as it pertains to users that are under their jurisdiction. And I have absolutely no idea how that is actually measured. I'm German, but I don't live in Germany, so I don't particularly know whether I fall under that or not.

• 38:30Chris Williamson?

  Is it where your signal is coming from? Is it where your phone is registered? Is it where your phone was bought? Blah, blah, blah.

• 38:35Speaker 00

  Is it where you currently are? I have no clue how that actually works. The point is just to illustrate that competition enforcers can actually come up with effective arguments in order to throw roadblocks into business models that try to pry on getting information from these different parts of the business, as we just described, is happening in China right and left. Facebook is trying to do it as well, but especially in Europe, I suppose, regulators are throwing roadblocks into these kind of attempts.

• 39:06Chris Williamson?

  Is part of that because in Europe, you've got this, you could look at Europe as one big country that's federal essentially, and you'd have to jump through the hoop of France, then the UK, then Germany, then Norway, as opposed to in America where you've just got one thing? Or is there something systemic about Europe that is anti-data, anti-big tech?

• 39:28Speaker 00

  It's a good question. See, GDPR is a Europe-wide thing, you know?

• 39:31Chris Williamson?

  And there's no equivalent in the US, right?

• 39:36Speaker 00

  There's no equivalent in the US. In the US, particular states are starting to get very sensitive about these issues as well. But it's certainly very different. So I think, yeah, it's both a federal and a European-wide issue that there seems to be more care taken with respect to data privacy. It's surprising to some, you know.

• 40:00Speaker 00

  Norway, for example, if I'm correctly informed, having Nest, you know, those home video camera kind of things, is kind of illegal here for privacy reasons. You can't have that. Which obviously is not the case in the US. And people are surprised about this because they say, wait, is Norway a country where everything is transparent? You can see your neighbor's salary and I don't know what. And the government collects all kinds of data about you? That's what my American friends, as I say, in response, I tend to say, wait, have you heard of the NSA? It's not like, it's not like your government doesn't collect the data about you. They just don't make it available for research.

• 40:35Chris Williamson?

  They don't tell you about it. Yeah, exactly. Have you heard, I don't know what the new product is from Amazon, one of their new products? Maybe like one of the Echo Dots or something like that. And as everyone was opening it up for Christmas, there's some stories of Amazon engineers being on the other side of it, pretending to be Santa Claus and doing all sorts of weird stuff. There was a lady who got death threats from the other side of there. Yeah, yeah, yeah. This is legit. This is legit. People in Amazon fucking with everyone.

• 41:03Speaker 00

  Oh, wow. So I didn't read about these particular ones, but I mean, I think what you're illustrating is just that there's just a real, let's call it business risk, if you take the perspective of the business, in how people react to the innovations that you're offering them. And again, there's a convenience trade-off and the privacy thing. And the question is which way you rob people more, you know? And whether you find Alexa more intrusive or you find it more useful. And I guess that's the, that's kind of like the cutting edge that companies have to navigate at this point.

• 41:35Chris Williamson?

  The bizarre thing is for the people like myself, I guess, who I don't fully understand all of the privacy concerns, but I've I've watched enough Netflix documentaries to be scared, right? But for me, for me, in the back of my mind, I'm always thinking about that trade off with something. So I don't have a Amazon Alexa or a Google Home Pod or any of that sort of stuff. I don't have that. I don't have it particularly because of a privacy concern. But if I was to buy one, I would think about it. I'd be like, right, okay, well, am I, am I prepared to forego my privacy for that? For me, it's not a massive, a massive element of my decision making, but it would be in there. And what's interesting is that when you get to

• 42:20Chris Williamson?

  see behind the curtain, which is my Skyscanner example, right? Had I have been anchored at the higher price, had I have logged on and refreshed the page and it had stayed at the lower price, or I'd logged on the first time and it was at £700, the higher price, and then refreshed the page and it had stayed the same. First off, I wouldn't have this story to tell you. Second, I wouldn't have mistrust in the site. So part of the game that's being played here with regards to our privacy and the way that data is being used is the transparency of our cognizance about it happening. Not whether or not it is happening, it's whether or not the user becomes aware of it. And that creates, that creates an environment for the company to

• 43:04Chris Williamson?

  purposefully try and be as obfuscating around what's being on, what's happening with your data as possible.

• 43:12Speaker 00

  Well, it's true, or to frame it differently. I previously said that people hate individualized pricing, but they love individualized discounts, you know? So it's a question of how you frame stuff as well. And yeah, I mean, you know, some companies at the cutting edge of this are way ahead of that consideration already. But, you know, many others will follow and have to make these sooner decisions. I want to, I want to throw in, like, switch, switch sides a bit and make an argument to not scare people. It's pretty easy to scare people with these privacy concerns. But here's another one. So say, why do I have a cell phone? I do have a cell phone. I have a smartphone, among others, because I find it really useful. But even if that wasn't

• 43:52Speaker 00

  a dominant consideration, do I really help myself by not generating all this data? So say,

• 44:02Speaker 00

  Say I was concerned that a health insurer is using all kinds of data about myself to price my health insurance. Am I actually concerned that I'm a worse health insurance risk than the average person in the society I live in? Probably not. I'm kind of like in shape and, you know, do sports and I'm generally healthy and all that good stuff. So I might actually want the health insurance company to know all these things about me if that makes me get a cheaper price than I would get without the phone. So true, they might still exploit my, you know, might extract some of these rents. But on the other hand, I might still get a much better deal than I would without generating all this data

• 44:45Speaker 00

  for them. So it's not entirely clear that, you know, living in autarky actually prevents you from leaving a trail because if you then apply, then you belong to the, to the set of people who don't have a smartphone.

• 45:00Chris Williamson?

  Yeah, the people that live in the woods, all they're doing is feeding on whatever they can find around the outside of their shed. Yeah, no, you, you totally correct.

• 45:11Speaker 00

  It's not entirely clear that I want to be counted as those.

• 45:12Chris Williamson?

  Yeah. Which side of the fence do you fall on? Because previously, when it's us versus company, there's always been these tricks of the trade, you know, like sort of between maybe the seventies and you hear these stories about Bill Gates getting free phone calls by like phone hacking, putting in particular tones back through the receiver to be able to make free international calls. And that was one of the first things he did, right? So there's this, there was this period, this golden era where companies were able to offer us services, but they hadn't caught up with all of the different ways in which people, the users could obfuscate their information, right? So you could be a guy who looks perfectly

• 45:52Chris Williamson?

  in shape, but know that you've actually smoked 20 a day for the last 30 years and then somehow get around that and get to whatever. But if your health insurance company knows your bank records and sees that you've bought a packet of cigarettes every day for the last 30 years, you can't get around that. But again, the, the problem is, or the, the concern is going to be, when do I fall on the side of the fence where it would have been better for them to not know this information about me?

• 46:18Speaker 00

  Well, you won't know.

• 46:21Speaker 00

  They know, but that's that's part of the problem. So see, the game you're describing is definitely being played and will be played a lot more. There are these these Fitbits or however, you know, whichever company you you you work with that counts a number of steps. Ha, you have one in SC.

• 46:43Speaker 00

  And why am I telling you this? Oh, right. If you know that your insurance premium goes down if you take more than whatever 10,000 steps a day, at some point it might occur to you that just making your dog run around with it during the day is a much

• 46:56Chris Williamson?

  You're just gaming the system.

• 46:58Speaker 00

  way to to achieve the same outcome and produce the same data.

• 47:02Chris Williamson?

  This man's this man's doing 70,000 steps a day. He's a psychopath.

• 47:06Speaker 00

  That's amazing, and he can run at 43 kilometers an hour.

• 47:13Speaker 00

  Well, in China, you can actually buy small electrical devices that do nothing else and shake around your Fitbit during the day on your desk.

• 47:20Chris Williamson?

  You're kidding.

• 47:21Speaker 00

  No, of course you can. So at some point, it has to occur to the data scientists analyzing the data that the users are actually trying to generate a certain, you know, a certain type of data that you're incentivizing them to produce. So you might actually have to change your algorithm and say, hey, we have to correlate the heart rate of the guy or

• 47:46Chris Williamson?

  That's when you need to attach it to the dog, isn't it? We're gonna have to shave, you're gonna have to shave a little bit of the back of the dog's

• 47:51Speaker 00

  Exactly.

• 47:51Chris Williamson?

  back of the dog's paw so that you can get a good connection for the optical, the heart rate monitor.

• 47:55Speaker 00

  Exactly. So all this to illustrate that this is a dynamic game and the stuff is being changed. And perhaps this is a useful, a useful place to talk about what computers can do in this game versus humans. This conservation we just said is a completely human interaction. A computer would never come up with that, that perhaps something has gone wrong with, you know, how people use the the Fitbit or the app.

• 48:20Chris Williamson?

  This poor, this poor person has been having a seizure for the last 10 hours. Yeah, yeah.

• 48:24Speaker 00

  Exactly.

• 48:26Speaker 00

  Or what, this person can have a heart rate of two hundred and forty seven?

• 48:30Chris Williamson?

  Yeah, yeah, yeah.

• 48:32Speaker 00

  You know, a computer would know what to do with that. In the end, it's a human being who stands behind that, which tells you that it's not that computers are intelligently analyzing data sets. Computers compute stuff, that's what they do. But the intelligence has to sit behind the screen and have a, you know, a human brain attached to it. And come on, in creative ways, have theories about what generates a data set, what causes the data to be generated the way it does, and therefore, which changes in the company policy or incentives that you offer to your user will actually have what kind of effect on their behavior. And that's a completely theoretical exercise, right? You're saying, hey, if I change prices, if I change the step incentives in a certain way, how will that cause people to misbehave?

• 49:16Speaker 00

  Or you offer penalties for abusing it, or I don't know what you do. This is stuff that has never happened in the past. So a computer cannot analyze a data set and predict what, how people behave in response to that. It's a completely human prediction exercise. And this is where the opportunities lie in that field, to think about what complements, economists say, what complements cheap machine prediction? And the answer is like intelligent humans that understand what the computers do and creatively think about, you know, the ethics, economic theory, think about whether you only see a correlation in the data or whether it's causal, that have some sort of moral

• 49:59Speaker 00

  compass or understand how users think about the ethics of what you're trying to collect in terms of data collection. I don't know, with humor, have you ever seen a computer come up with a joke?

• 50:13Chris Williamson?

  I follow, I follow a guy on Twitter who keeps on posting these images that he says are machine learning scripts for stand up. Now, I think that they're him writing them as if a machine did them because they're so side-splittingly funny. And it's like, it involves the, it includes the audience interaction and stuff like that. It's absolutely insane. I don't know, I haven't seen a computer be funny, not yet.

• 50:43Speaker 00

  Yeah, right. So people are actually trying to do this, right? And puns, computers are starting to be good at kind of puns. And you can see why, because you're just analyzing a whole bunch of language and puns that people find funny. And then you can predict what kind of a combination of words people find funny. Okay, fine. So far so good. And they're successful, whatever, fifty-two percent of the time. So that is, I don't know, that's probably a better rate than me in the classroom.

• 51:12Chris Williamson?

  It's funnier than me. It's funnier than me by miles.

• 51:18Speaker 00

  But you know, it's doubtlessly true that computers by now can come up with funny puns. So if you have a human being that selects all those that are funny and then gives them to an audience, great. That worked. But it's still a human judgment that's behind the exercise.

• 51:37Chris Williamson?

  Got you. So one of the things that you've mentioned so far is computers struggle to predict in situations that there's essentially no precedent for.

• 51:47Speaker 00

  Yes.

• 51:48Chris Williamson?

  Will that change?

• 51:51Speaker 00

  Well, people are trying. But let's say, this is, the main point we're trying to make in the book is that's ninety-five percent of what you read about in the newspaper, like thinking computers essentially. But it's a very small percentage of what actually happens in the real world. What happens in the real world is that people assemble humongous data sets about your behavior and predict which movie you want to watch next. Okay, so that's the stuff that actually happens, that's economically viable. So as people think about what actually affects their real life in the next five or ten years, it's definitely not thinking computers. It most definitely is a bunch of fancy statistics that is really boring to talk about.

• 52:34Speaker 00

  It's a bunch of engineers analyzing big data sets, and that's it. It has nothing to do with thinking computers and all these more exciting and hence more headline generating.

• 52:45Speaker 00

  Things that you that you read about elsewhere.

• 52:48Chris Williamson?

  I suppose, you know, thinking about Nick Bostrom or Max Tegmark, they're they're kind of either utopia or dystopia, depending on how you look at it for the future of AI. When you look at what they talk about, they're not talking about essentially the the world's best statistical modeler. What they're talking about is artificial general intelligence, right? They're talking about a thinking, a thinking computer. But it was, it would seem at the moment, you know, there was that, there was that, um, at a particular conference a few years ago, someone asked a room of experts, how long do they think it would be before we reached artificial general intelligence? I don't know that it was the singularity, but it was certainly artificial general intelligence. And like the, the consensus was like 50 years, like before, definitely before 2100, right?

• 53:32Speaker 00

  And by the way, that was a consensus in the 1950s, too.

• 53:36Chris Williamson?

  The same, oh, the same distance from the night. So it's 50 years from 1950s and now 50 and it's just moving as we, as time goes on. Okay. That's hilarious.

• 53:46Speaker 00

  So whatever the right number here is, the point here is people have talked about this for decades and decades.

• 53:51Chris Williamson?

  Yeah.

• 53:52Speaker 00

  And the horizon has shifted out. See, uh, you you can call me wrong if it happens next year and and I ridiculed the idea. But as you say, um, the closer people are actually to the forefront of research, the further out in the future, it seems to be that they expect artificial general intelligence to be there.

• 54:11Chris Williamson?

  Yeah.

• 54:11Speaker 00

  So our point is exactly, hey, this is very exciting to talk about and people have and it's a very appealing thought to human beings, apparently. But what actually matters in the real world right now and what affects people's lives is boring statistics.

• 54:23Chris Williamson?

  Hmm. And that, my point here is, we've got, this is when it's going to happen, artificial general intelligence, you know, this is what's super intelligence, which is one of my favorite books. And everyone who's listening, who really, really wants to kind of get a good grasp of how artificial general intelligence might come about should read Super Intelligence by Nick Bostrom. It's awesome. Um, this is kind of the Terminator, uh, romantic view that is being put in the press. But that's not what's being economically rewarded right now. What's being economically rewarded right now, a big data set being analyzed cleverly so that you can make good predictions. So you have to presume that as we move forward, the resources are going to be disproportionately moved in that direction. They're going to be discriminated towards what can make Amazon make more money or what can make my insurance

• 55:08Chris Williamson?

  company make more money, not. Should we throw a trillion dollars into maybe making artificial general intelligence when we don't know if it's possible?

• 55:18Speaker 00

  That's exactly right. So this is exactly the point. See, if you want to see what currently makes money for businesses, look at the market cap of companies and you'll find companies such as Amazon, Google and Facebook up there, don't you? Or of course, the Chinese tech giants. So what do they do? Well, so one thing I show to my MBA students is an annual report of Amazon in 2006. And so that's, you know, like 14 years ago or 13. And what you see is Jeff Bezos talking about that they have the ability, the data and the technical ability to analyze what economists essentially call price elasticity for demand. So if you change the price of a certain good, how are people going to react to it? Are they going to buy more or less of

• 56:01Speaker 00

  it and how much more and how much less? The techniques by which you analyze that, that's basically just dead boring PhD level or actually more like 10 year US top university professor level statistics and econometric modeling. And this is exactly the people that Amazon hired over the last few years. It's literally like hundreds of PhD economists that analyze these data sets. So if you want to use what produces a trillion dollar market cap, you just have to see what they do. And it's just, you know, econometric or economic modeling, a whole bunch of data science on huge data sets has absolutely nothing to do with artificial general intelligence.

• 56:43Chris Williamson?

  And that's what's

• 56:43Speaker 00

  It's kind of, yeah.

• 56:45Chris Williamson?

  That's what's rewarding companies right now, right?

• 56:48Speaker 00

  Right.

• 56:48Chris Williamson?

  That's what's getting them to

• 56:50Speaker 00

  Well, and the financial markets expect that to reward companies in the foreseeable future, right? Because that's what market caps of companies are. It is expectations of future profits.

• 57:02Chris Williamson?

  Talking about it on a wider scale, then, we haven't touched much on the global economics of this. But what's your predictions moving forward? Is this going to change anything? Are we going to see market caps and those predictions moving more or less? Or, you know, is it going to change the way that the actual financial markets themselves are going to be manipulated or going to be judged by the companies that are looking after them?

• 57:29Speaker 00

  Uh, yes.

• 57:31Chris Williamson?

  Okay, yes, please, please go on.

• 57:36Speaker 00

  So I'm a finance professor, so there's a risk of me diving into into finance topics a little too much for for for this audience. In financial markets,

• 57:46Speaker 00

  the most successful hedge funds and market participants are precisely those that have used computer science and models for the last few decades predicting future stock returns, and it works like a charm, and and people are rich. Okay, fine. So in financial markets, there's a huge transformation happening. I cannot get too much into that. You're asking much broader, a much broader question. So I get invited a lot to investor conferences that are like, I read about AI in the newspapers a lot, and there are thousands of startups. But first of all, is this just a fad? And second, if it's not just a fad, then which of these thousands of startups am I supposed to invest in? How do I think about this?

• 58:28Speaker 00

  And and what I tell them is, well, see, why is there so much AI? It's because there's a lot of data, hence big data. Why is there so much data? Well, because it's cheap to collect data, it's cheap to store data, and it's become cheap to analyze it. Okay? And when a product is cheap, people buy a lot of it. An economist says demand curves are downward sloping. But okay, you can forget about this again if you don't care. When things are cheap, people buy a lot of it. It's really cheap to collect data. I can get your cellphone location data if I data aggregator a few hundred bucks and they will tell me where you are. It's very simple. It used to be implausibly expensive, and I need to hire a private detective. No longer. Okay?

• 59:10Chris Williamson?

  Some guy, some guy in a leather, a leather suit and a big wide brimmed hat, yeah, smoking.

• 59:15Speaker 00

  That's right. So, hmm, okay, so that's why that's the year. Now, is this just a fad? Well, unless you tell me a story why data collection and data analysis is suddenly going to become way more expensive again, the answer is no, it's not just a fad, it's going to stay here. Now, let me, let me put a small disclaimer in there, which is, I can actually tell you a story why it becomes more expensive to collect and, and process data. And it's called legal constraints. Think GDPR in Europe. So sure, if politicians wake up to these concerns or convinced by these concerns, they might throw roadblocks in it. And it's going to differ from jurisdiction to jurisdiction. And therefore, if you're in this business, that's one really important consideration you should have in mind as you get in this field.

• 59:59Speaker 00

  Well, which leads to the second question, which kind of businesses should you invest in, given that you want to invest in the space in general? Or which one companies should you want to work for and so forth? I mean, one, one question is concerns of legal jurisdiction, but the other one is understanding what the economics of their business model is. Is it a business model to make better predictions than the competitors and therefore like offering a better product? Okay, that sounds great. Is there a business model to have some sort of useful thing people like and that people are willing to give up their personal information for? But the example I always think of is like scooters, you know, inner city scooters or bike sharing and stuff.

• 1:00:40Speaker 00

  The question of these business models has very little to do with whether the fee people pay to rent a bike actually pays for buying the bikes and, I don't know, keeping them charged and maintained. The economics of it is that you sell the people's cell phone location data to a data aggregator who sells it to Facebook or Google or whoever wants to buy it, which they use to target ads. Okay, so let me, let me go backwards in this value chain. Obviously, correctly targeting ads is a valuable thing from a business perspective, but you need a bunch of data for it. Where does the data come from? Well, in some cases, the companies like Google and Facebook collect the data themselves. But in other cases, for other variables, like how fast do you type your name in an online form?

• 1:01:25Speaker 00

  They might buy it from a data aggregator. The data aggregator buys it from whoever first came up with the idea of measuring how people, how fast people buy stuff in an online form. And whether that's an online forum for car insurance or an online forum for I want to rent a scooter in Madrid, doesn't matter the first bit, no? And this is why I'm saying, hey, the logic, the economic logic of business models really changes as data itself becomes valuable. Because the whole point of your business model might be to collect a bunch of data and it has absolutely nothing to do with the product. If you program a flashlight app for a phone, but somewhere in the terms of conditions, you say that you also want their entire address book and their, their, their, their location data, how fast they type the names and track all their activity on the internet.

• 1:02:10Speaker 00

  I mean, if you get people to sign up for that, well, good for you.

• 1:02:14Chris Williamson?

  More power to you, yeah.

• 1:02:16Speaker 00

  You don't need it for the flashlight app. So I'm sounding very cynical and I suppose I am. This is not an endorsement of these kind of practical practices or me saying this is ethical. All I'm saying is like, this is what is happening out there in the real world. Indeed, there are ethical constraints to it. Indeed, legal constraints are there and start popping up. But that's the economic logic of how business models are changing.

• 1:02:38Chris Williamson?

  Presumably as well, because this is such a fledgling market that's moving so quickly. The, there's an asymmetry in terms of how many people want data to be more available from the company side, how sophisticated they want the modeling to be able to be and the decisions that can be made off the back of that versus me and you thinking, I, I don't want my data to be sold to these people. I should speak to my local counselor who's busy like dealing with like a, you know, a proper, like a flood or like a real, a real sort of real world problem that's going to make news as opposed to me and you saying, I think if this keeps going over the next

• 1:03:20Chris Williamson?

  20 years, this could be like Orwell's 1984 and, you know, that like, so where does the rubber meet the road with regards to protection for users?

• 1:03:30Speaker 00

  Well, I think it's a societal decision, see? And in the end of the political game, so as we actually alluded to previously, in Europe,

• 1:03:40Speaker 00

  both individuals and politicians seem to be more on the cautious side concerning data collection.

• 1:03:47Speaker 00

  And in some ways, and just in a vague sense, and as I said previously, I don't know, maybe that's a bad thing. I mean, first order, having more data enables better decisions. So as long as it doesn't get abused in some sort of way, having companies that, I don't know, give me, I love Uber. Having taxis in the small town in the US where I lived for a long time was a disaster. I loved Uber coming along and offering me cheap rides, no? So first order, this is all great. The question is just when the concerns come in, and that depends a lot on how many companies start abusing the power they have thanks to the information and how politicians react to it, either to the company's concerns or lobbying efforts.

• 1:04:30Chris Williamson?

  Yeah, the lobbying, the lobbying must be so, that must be where I would imagine companies like Amazon and Facebook are really ratcheting up their spend. You know, if first off, they needed to get a lot of computer scientists, up next, they need to get the lawyers and the lobbyists who can protect the work that the computer scientists are doing.

• 1:04:49Speaker 00

  Yeah, and the economists, I suppose. So see, what you get from all of this is a sense that there's going to be a few winners in this space and there's going to be a bunch of losers, no? And obviously, that's a societal concern. Everybody talks about inequality and this is one mechanism, you know, which can drive it. And as I said, I don't expect this to stop anytime soon. It's going to go faster in some jurisdictions than in others. It goes fastest in China for the reasons we've discussed. So if you want to see whether you want to live in a society, you know, that has very little regulation around these topics, you can look there and see if you like it. There's many things I like about it. Go to Germany and try to pay with a credit card in a supermarket. It's going to be hard because they like cash.

• 1:05:33Speaker 00

  Meanwhile, in China, you walk past the supermarket counter, they scan your face and that was it. Oh my god, is that beautifully convenient. You know, it's similar like the airport, the airport face scanning technologies, whatever. It's it's not a technical problem that has been unsolved. I mean, how much would you like to not have to stand in line at a supermarket because some person in front of you is trying to, you know, count individual coins to get the precise amount, you know, like in the eighties.

• 1:06:03Chris Williamson?

  Scan my face. I've put my stuff in the bag. Scan my face.

• 1:06:06Speaker 00

  You know, it's just incredibly convenient. So I I do not want to be paternalistic about which kind of society we want to live in. It has huge benefits to live in this more technologically advanced world. My prediction is that likely we're going to move in that direction more than not. But to which extent and how fast and in which jurisdiction and who makes the money, those are the exciting questions from yeah, an individual perspective, an investor's perspective, a potential employee's perspective and so forth.

• 1:06:34Chris Williamson?

  And we can watch China. We can just see what happens in China. They're like the the canary in the coal mine or the monkey that got shot out into space. And it's like if their if their entire society breaks down and it it it becomes a dystopian waste world where Mad Max is is running around everywhere. Then we know we're like, all right, let's just pump pump the brakes, Elon. Let's have a, let's have a, let's have a chat, like just chill out a little bit for a while.

• 1:06:58Speaker 00

  Right. So I mean, see, I hope the sense that that listeners will get from this is that it's definitely not a time to just sit back and just because you can still pay with cash in your supermarket to think that you're kind of like immune from this. That's how it works. You're just five or 10 years behind. And this is not ideally where you want to be in this space because as you said, it's moving incredibly fast.

• 1:07:20Chris Williamson?

  Well, man, Martin, today's been, today's been awesome. Where should people head? They want to get the big, the business of big data. Amazon? Go on Amazon?

• 1:07:28Speaker 00

  That's, that's what they should. That's, that's where to find it. That's the only place to find it, in fact.

• 1:07:33Chris Williamson?

  Got you.

• 1:07:34Speaker 00

  And I can't wait to hear people's reactions. It's, it's our first book on this topic, the first attempt at it, and we'd love to hear back.

• 1:07:44Chris Williamson?

  Yeah, I've got to give a shout out to Yuri as well. So the the guy, you're, is your co co author on the book?

• 1:07:50Speaker 00

  Right, so we got to know each other when I was a graduate at Princeton, he was an undergrad and he wrote a popular book on statistics and I always thought that was an oxymoron, but he somehow managed to pull that off. So that tells you something about that he's a really, really good and funny writer and I very much enjoyed him there and working together with him on the book and I hope it shows.

• 1:08:13Chris Williamson?

  Man, he's awesome. I've been in touch with him ever since I got introduced to the browser by David Perel last year. I've been in touch with him, just back and forth, just nerdy stuff. I'll send him something, I've seen this thing, do you want to have a look at this thing? And he'll always, he'll always take time to reply to me. So yeah, if you haven't, to the listeners, if you want to sign up to a service which allows you to get curated articles every day and then a synopsis of them at the end of the week, this has been the only reason I actually think that I'm interesting over the last 12 months. That the only reason I have anything interesting to talk about is because of the browser's daily emails where they send me like random articles about like how high a pig can jump or like, you know, like the, whatever it might be. Yeah, it's, that's

• 1:08:58Chris Williamson?

  exactly what it is. But yeah, that'll be linked in the show notes below. Yuri, thanks so much for the link up, man. Martin, business of big data will be linked in the show notes below as well. Ladies and gentlemen, it's been awesome. Go and, go and put some sellotape over your webcam and make sure that Amazon aren't watching you do naughty stuff in your bedroom.

• 1:09:17Speaker 00

  Thank you very much.