Products with Personalities? My Interview with JD About Conversational AI (Tech Strategy – Podcast 31)

In this class, I talk about my interview with He Xiaodong of JD. And about how chatbots are evolving into conversational AI. Plus the basics of AI economics.

CORRECTION. In this podcast, I mentioned that JD’s Smart Customer Service has the personality of a young woman. Xiaodong was actually referreing to Microsoft’s Xiaolce. My apologies.

You can listen here or at iTunes, Google Podcasts and Himalaya.

Related podcasts and articles:
This is part of Learning Goals: Level 4-5, with a focus on:
  • #22 Basics of AI

Concepts for this class:

  • AI – Cheap Prediction
  • Mismatched and/or Crippled Scale

Companies for this class:

  • JD


I write, speak and consult about how to win (and not lose) in digital strategy and transformation.

I am the founder of TechMoat Consulting, a boutique consulting firm that helps retailers, brands, and technology companies exploit digital change to grow faster, innovate better and build digital moats. Get in touch here.

My book series Moats and Marathons is one-of-a-kind framework for building and measuring competitive advantages in digital businesses.

Note: This content (articles, podcasts, website info) is not investment advice. The information and opinions from me and any guests may be incorrect. The numbers and information may be wrong. The views expressed may no longer be relevant or accurate. Investing is risky. Do your own research.

——transcription below

Welcome, welcome everybody. My name is Jeff Towson and this is Tech Strategy. And today we’re gonna talk about artificial intelligence in China and specifically my interview with a couple of the AI heads at Jingdong JD, obviously one of China’s biggest online retailers. So not really a case for today, just sort of more moving out to the frontier of what’s happening and trying to sort of get a look around the corner a little bit. I got a call. from some of the folks over at JD. I talk with them on a regular basis. I find that they’re really sort of inventive and innovative in some kind of neat niche areas. I mean, they’re doing the big stuff like the e-commerce giants and all this. They’re doing the kind of big innovation moves you would expect, but they also have a lot of side projects, smaller stuff, farming, garbage cans with sensors in them, lot of interesting stuff at JD. So I like to talk to them and sort of see what they’re doing. Anyways, they called and said, you know, was there anyone you’d like to talk to? And I said, absolutely, I want to talk to the heads of AI because AI is kind of the big deal. It’s on the frontier and there’s a lot of experimentation and really the e-commerce companies, big surprise, you know, they’re kind of in the lead in a lot of ways. So I wanted to see what, you know, they were focusing on. And the standard answer to sort of AI and e-commerce is people say, oh, there’s a lot of focus on logistics and there’s a lot of focus on the sales and sort of demand prediction aspect. That’s kind of your standard answer. Fine, let’s be more specific. And that’s what we dug into. So this is gonna be the first of two podcasts on a couple of people I spoke with. And today we’re gonna talk about how chatbots have evolved into conversational AI. And… likely a future where products are gonna have personalities of their own, which is a very strange thing to talk about and think about. And the guy I talked to is He Xiaodong, who is overseeing that sort of customer service side where this originated, and I’ll go through some of the thinking on that. But it’s pretty fascinating, really. And we will also start to tee up the basics of how to think about AI, because it’s a big subject and we haven’t really jumped into it in the… class very much yet. So we’re gonna start to lay sort of the foundation for that. So if you don’t know anything about AI, awesome. This is sort of a little bit of an introductory lecture on that as well. But first, if you haven’t signed up, please go over to You can sign up there. There’s a 30 day free trial to join the class. And the class is moving along pretty well now. We’ve built up quite a bit of content. About more than half of it is behind the paywall and that will probably increase. And it’s really structured as a class. It’s a get on this process, step by step, week by week. We will move you up systematically in your level of understanding. And there’s gonna be more and more sort of structure to that. And then we’re moving towards a situation where we’re gonna have certificates for various completion levels and things like that. So you can go do that. Also homework for those of you who are subscribers, I’ve been pushing you to. You know, apply one concept in whatever level you’re on. First steps, level two, level three, level four. Level five is now done. Whatever level you’re on, pick out a concept this week. Anything could be simple. And look for a company that has that and sort of write up a couple of paragraphs. Try and do one of these every week so you move a little more systematically through the levels. I will be putting more structure on that, not probably homework, but tests and things like that so you can progress from level to level to level, and that’s what’s on the way. But for this week, do the homework assignment, apply the concept to one company you see in your regular life and sort of see if you can explain it in those terms. Okay, let’s get into the content. Now JD is a really interesting company. It doesn’t get as much attention as Alibaba, which is much larger and has a lot other businesses going on. It’s a bit more China specific as well. People outside of China don’t seem to know about it, although it’s here in Thailand and it’s in a couple countries around Southeast Asia. Really interesting stuff going on with JD and it’s a very interesting contrast with Alibaba because they’re both serving the same consumer group, buying and selling things, but they have somewhat different approaches, different business models. And then you have to keep in mind, they have a close partnership with WeChat Tencent. So it’s really also the WeChat world as well. Now, if you wanna know more about JD, there’s a previous podcast I did on this, which was podcast 18, which was JD versus Alibaba, how retailers and marketplace platforms compete and evolve in China. That’s podcast 18, I’ll put the link in the show notes. And there’s quite a few articles on this. The last one I wrote was for members only, and that was JD and competitive advantages versus entry barriers. That one’s for subscribers. I thought that was a pretty good take at how these things sort of, you know, compete with each other. So that’s more if you wanna read about that. And the main two ideas for today, sort of the two things that are the takeaways, is to think about mismatched and crippled scale. That’s kind of my own term, mismatched and or crippled scale. That’ll be sort of one of the two main concepts for today. And the other one is AI as cheap prediction. That will be the second one. And both of these will go under learning goals 22. Okay. Now, I mean, the point of this whole class has been how does digital change strategy? It’s Michael Porter meets Jack Ma. And digital is… It’s not a great term, but I usually, I’m talking about software and data, both. Some people call that data technology, but basically, you know, things made of bits and bytes, ones and zeros, not molecules. And how does that change business? And I put all of this in kind of three big buckets that, you know, there’s actually a couple important things happening at the same time and it’s easy to confuse them. One of them is just like platform business models. A lot of times when people talking about software and data, like it’s really just about the business model. that these platform business models, these digital platforms, can emerge in an industry when enough of it gets digitized. And that’s just a powerful business model. It’s mostly about that. Another bucket would be, okay, look, the economics of the business are starting to change and become more digital instead of making tables where the economics would be very traditional, manufacturing, scale, things like that. Suddenly the economics are a lot more about software. And I’ve given you quite a bit of talks on sort of… how economics of digital things are just strange. And it can change the economics of an industry. That’s kind of the other, the second big bucket. And we get lots and lots of crazy little use cases. Sometimes they’re a big deal, sometimes they’re not. The third bucket, which I haven’t really talked about much, is just this idea of digitized operations. When a business starts to become run on software, you’ll hear this referred to as invisible engines. as opposed to people. And when the core processes of a business just become automated, the business starts to look very different. So I’ve referred to that as digitized operations, I’ve referred to it as zero human operations or AI factories. Well, within that third bucket, AI is a big deal. Because usually to do a business, the reason we have humans in businesses is to make decisions. And other reasons, but that’s the big one. When that bit starts to get taken over by software, suddenly the bank can just take your application, look at the numbers, make a prediction, make a decision, and issue a loan in a second, and there’s no human involved, which is very different than how a loan would get processed at a traditional bank. You go in, you put in your paperwork, someone looks at it, makes the decision, loan officers. So that’s kind of the third big bucket. And within that third bucket, which is really what’s happening right now, AI is a big part of that. I’ll talk about that. And one of the interesting areas of this, and AI is just software. I mean, AI is a terrible term because it’s all sort of highfalutin and makes it sound really, it’s just software. It’s just software that makes predictions. It’s not any different. So it fits within our bucket, but sort of looking at, you know, talking with JD, the person I ended up talking to was working in customer service. which is not the sexiest area of business. I mean, people don’t get out of their base. I really wanna work in customer service. I wanna have big call centers with thousands and thousands of people answering phones about insurance claims and returned products. And I missed my flight. Can you rebook me please? Now I’ve done projects in call centers in the United States. It was not a lot of fun. It’s actually really hard because the staff quit so often because people are kind of mean to call center agents. There’s a lot of yelling going on. So you have to train your people to answer the, you know, the calls. And it could be on the phone or it could be up through a chat. And a lot of these are in Manila and, uh, Cebu and places in the Philippines cause they speak really good English, at least for Western call centers. And, you know, it’s not great, but you hire people, you train them, and then they tend to quit a lot. because it’s not a fun job. So you’re always, you have a huge amount of churn. It’s a big problem. And I sat down, or I was part of a meeting with the CTO, the Chief Technology Officer of JD, this is a couple of years ago, and he was just talking about what they’re working on, autonomous vehicles for delivery, maybe drones, which are good for media, although they don’t seem to make a big deal in practice. And one of the areas he kept bringing up was customer service. Because it turns out if you run a massive e-commerce site, one of your biggest problems is customer service. And this is where we get into this first concept for today, which I call mismatched or crippled scale. And or crippled scale. Now JD has an interesting history in this, that they started out as a business and online, firstly a physical retailer, they used to have stores, and then they moved into online retail. during the SARS epidemic, 2002, 2003. But they’ve always positioned themselves, even before they were online, as high quality. A marketplace platform like Alibaba has a lot of quality problems because people upload fakes. And you’re not curating the goods yourself, you’re not stocking the shelves. The merchants are doing that, so you never know what they’re putting up. And after the fact, you can go out and say, oh, that was a fake, remove that seller. but they can upload again. So you always have a problem with fakes and quality and there’s a lack of trust there. Well, JD never did that. JD was always a very much smaller number of items offered on their platform, but they’re always guaranteed to be the real thing. If you want a real Gucci bag, you buy it at JD. If you want a real good laptop, I just bought a laptop from JD last week, you buy it there because you know it’s a good one. And if you have a problem, their return policy is very, very good. And their customer surgence. service agents are available. You know calling customer service, no offense to Alibaba, is not great. The volume is so big, it’s automated, it’s not good. But I can always call someone at JD and get them on the phone and say look I bought a laptop it’s got a problem and they will take take it back and they take good care of you. So the three customer touch points they have would be the web page, buying goods, you know the goods are quality, delivery people. They have their own in-house delivery team. Well, they used to. And these are people in nice uniforms that show up. They don’t just zip up to your site and throw you the package, which kind of happens in a lot of cases. No, they’re very professional. They’re very well-trained. And they’re not paid on a per-delivery basis, because that’s another customer touchpoint. They don’t want bad behavior there. And then if you have a problem, you do customer service. What could be online chat, it could be a phone call, it could be a lot of things. Again, they’ve invested a lot of money there because that’s the three points where the quality aspect comes through. And that’s always been their position since day one. They’re about 17 years old as a company now, and they do very well. So they’ve always been focused on customer service for a long, long time. And the concept here of mismatched or crippled scale. Here’s the funny thing about these platform business models. Platform business models are powerful for multiple reasons, one of which they can grow very quickly. But digital can do that generally. It’s easy for Netflix to sell in 10 countries. It’s easy for Google just to open their website up to other countries, and they can grow and scale very easily. But one of the benefits of a platform business model for a marketplace is there’s a lot of time and investment. You have to build the storefront, you have to have the inventory, you have to sell it. Okay, JD and Alibaba don’t have to do that. The merchant does that for them. Facebook has the same thing. Facebook is the biggest media company on the planet, but they create no content. All that hard work, that effort, that money, that time. Well, the people who post on Facebook do that for them. The New York Times spends tons of time and money writing articles and investigating them, and then puts it on Facebook, and Facebook doesn’t have to do that work. So when you grow, you can grow incredibly rapidly because you’re not growing based on your own internal assets and work and activity and investment. You’re doing it based on the assets in the ecosystem that you connect. My standard statement is like, a traditional business like a media company or a retailer like a Walmart, that’s a house you build for yourself. Okay, a marketplace platform like a JD or a media company like Facebook, That’s the castle the entire village builds for you. But you don’t have to spend all that time and money yourself. So they can grow incredibly quickly, which we see with Alibaba, which we say with JD. That’s the scale effect. It’s not just that it scales easily because it’s digital economics and there’s no reproduction costs. It’s also because all the energy and time and investment is done by other people, not you, if you’re the platform. Here’s the problem with that. Not all aspects of your business scale that way. Certain parts do. Facebook can scale very easily in terms of content creation and distribution. But when you get to content moderation, which is a nice word for censorship and curation, that part of their business, which is absolutely necessary to do. This is like, OK, a lot of people are uploading Facebook. articles and pictures and stuff or let’s say YouTube videos in Indonesia and people are putting up horrific videos of violence or Nudity, okay, you can’t have that on your platform without the platform Becoming very toxic so you have to curate and Or censor the material Well that part of the business doesn’t scale up as nicely You know the the uploading bit scales quite quickly The curation bit, there’s no actual mechanism that does that as well. So you get one part of your business that can scale like you’re driving around in fifth gear, and you get another part of your business that scales like you can only drive in second gear. So I call that the mismatch of scaling. And it even gets worse because a lot of times, there simply isn’t a way to scale one part of your business at all. YouTube has this problem. YouTube… scales beautifully in terms of everyone can upload videos, millions of hours of video are uploaded every hour. There’s a curation and censorship problem because there’s no way to actually watch all that video. There’s no way for some editorial team to watch all those millions of hours of video every hour and actually remove horrific stuff stuck within one random video. There’s no way to do that yet. So I call that crippled scale. Now they can have a couple mechanisms they use, none of which work very well. They can have reviews. People flag things that are bad. They down vote them. And then they go maybe to an editorial team. Okay, they’re doing that. It doesn’t work very well. People game it. You know, they gang up on content they don’t like, usually for political reasons, to get people censored. It’s like a game. That’s not very good. And they can try things like doing AI. doesn’t work very well, either it’s pretty mindless, which we’ll talk about, or they can have content moderators, which is another fancy word for censors. And you know, these companies like Facebook have hired, you know, 10 plus thousand of these people. None of those work very well. So I call that crippled scale. And this is a problem. Yes, digital is amazing. You can scale beautifully. Not all aspects of your business can do that. I was at a dinner with Michael Evans, the president of Alibaba, a couple months ago. He was just talking about this. He wasn’t using the same language. These are my vocabulary. But he was talking about just the problems of fraud in e-commerce, that as you scale up, fraud happens. The mechanisms to combat fraud are more human, and they don’t work as well. They don’t scale as easily. That’s just their ongoing challenge. No, a company like Ctrip has this problem because Ctrip will book flights and hotels. Expedia. Okay, but if you miss your flight or your flight is cancelled and you’re 2am in the New Delhi airport, you need to be able to call someone or get on a chatbot and get your flight fixed. So Ctrip or Expedia, they have to have huge numbers of customer service reps to handle those calls. It’s just the nature of travel. So they have huge numbers of people in, what’s that, Hongqiao? Western Shanghai, they have big call centers and I think they also have big centers in Nanjing, that’s where they center their operations. So you have this problem in a lot of these digital businesses and it’s just the nature of the beast. So that’s kind of the first concept today. Think about, okay, yes, these are very scalable business, they scale quickly and you can lever in other people’s assets and energy and activity. But not all parts of your business scale that quickly, so you get a mismatch, and sometimes you’re just crippled. And that brings us back to sort of Jingdong’s customer service empire. I mean, they have been actively building customer service for 10 years in a very major way. Because they have this problem, they are the quality retailer, the quality e-commerce site, where you always get the goods. If you ever have a problem, you can return it. And that’s one of their major touch points with consumers is quality customer service. Okay, well they were growing quickly and in one of the quirks of history, the founder, Richard Liu, he’s from Suchian, which is a very small, used to be a poor village to the north of Shanghai. And that’s where he was from. He famously came out of there and made his way as a young man, very poor, down and just try and start business. Well, he set up you know, sort of the headquarters for JD’s customer service stuff in Suchen in his hometown. And it’s grown so big, he’s turned his sort of small, really, it was a poor village into this major customer service center for China, because mainly he was from there. It’s like people call Hangzhou a first tier city now. Why is Hangzhou a first tier city? Because Jack Ma was from Hangzhou. That’s why. If he had been from Chongqing, we’d all be talking about Chongqing a lot more. Okay. Richard Lio is from Suchien, so there’s a massive number of customer service operations in Suchien. They started this around 2010, started hiring people, putting in this stuff. They probably got about 10,000 customer service reps. They say they’ve invested about 1.5 billion renminbi over the last 10 years. And because of this, other companies have moved in. Well, actually… JD, they also have staff in Changzhou and Chengdu. I’m not sure how many. I think most are still in Sichuan. But because of this, other companies like Xiaomi and Tuneo have set up their operations there. It’s just become a little bit of the customer service city of China, or one of them. And it’s because he’s from there. Now this used to be chatbots and telephones. Okay. But their IT team has been working on this for a long time and out of this has become what… He Xiaodong and I talked about, which is conversational AI. So now we’re gonna sort of shift from this idea of customer service and chatbots, which is not that thrilling, how this is turning into something that’s much, much bigger and really pretty cool. So Mr. He sort of gave me a little presentation on what they’re doing. He’s a serious artificial intelligence gentleman about what they’re doing. And it’s really great. And the first point to sort of think about is, okay, let’s get away from this idea of chatbots. You know, the little pop-up screen on your phone and on your PC and calling in. It’s much bigger than that. It’s multimodal communication. That’s the right word, multimodal. It’s not just chatting, it’s also talking. Talking on the phone. Okay, typing something in online. and the AI can understand your question and it can reply to your question. So it’s a two-way communication. It involves text, it involves voice. In theory, I can talk to it. It will understand my words. It has to do with content creation. The chances are the AI is not just gonna give me back, like, hey, where is my package? And then the AI comes back and says, oh, well, it’s being delivered on Friday. They may well give me content in terms of, here’s an article you can read about this. Here’s a map, here’s a photo, here’s a video. It’s more than just text, it’s text, voice, JPEGs, video, music, content creation. It’s multimodal communication. And a big part of that is content creation on the AI side, whether it’s writing text, answering questions, sending me pictures, whatever. The second thing, and really the first takeaway that I got from talking with him was, it’s not just about understanding the problem. I mean, AI is about understanding. We’ll go into some of the basics of AI, but it’s about understanding a question or a problem and then making a prediction about what to do. Okay, it’s not just about solving a customer’s question or problem, it’s about being empathetic. It’s about understanding their emotions as well. Is this customer frustrated? Are they angry? How do I solve that problem? How, you know, the angry customer who’s typing something in and I can tell the AI can tell this person is upset. How do I answer in a way such that by the end of the call that person’s emotions are much better? They’re happy. So you’re solving two problems probably. You’re solving the customer’s actual technical problem, whatever it is, but you’re also assessing their emotions and you’re improving that and you have to do both. So when you talk to sort of the JD chat bot or the JD personality, you know, they have given their conversational AI, which is the right term, a personality. And the personality they have given them is basically a young woman. That is the person you’re talking to. It comes across in that way. And this person is a cheer. You could call this, I call it conversational AI. That’s the entity you’re talking to. You could also call it, they call it JD Smart Customer Service, but this young woman you are talking to, whether in voice, whether typing, is curious, very friendly, someone that people like to talk with that makes them happier to talk with. So when you come in and maybe you’re frustrated, where’s my package? By talking with this person, you feel, you put you in a better mood. They’re not managing, but they’re responding to your emotions as much as to your problem. This person is active in the social world. This person knows the new terms people use online. They know what content is popular right now, what music, what TV shows. This AI personality can create content, not just chatting and answering you, but can write songs and put little things together and show you. And have you seen this video? I mean, you’re talking with a personality. This is conversational AI. It encompasses, you know, solving the problem. in this case, customer service and emotions, personality. It’s both of those things at the same time. So you can see for basic customer service, that would be much nicer. Okay, but clearly this is bigger than customer service. Clearly this is a capability. Conversational AI is a technology. It’s a capability, a personality that you can deploy in various situations. And really what you’re going after, is you’re trying to change the human to machine interface. So that when you interact with a machine, which in this case would be started with a chat bot customer service, but it could be a sales agent. It could be your experience online. It could be talking to your television. It could be talking to your phone. It could be talking to your AI speaker in your house. It could be talking to your car. You can start to deploy this conversational AI as a way of engaging with a machine. anywhere in the world. So when we first talked about customer service, and you can see how that would work, but the other area, one of the other areas they’re working on is sales. Like this idea, like when you go on a web page, JD, and you start to browse, oh, I’m looking to buy some Nikes. And you start to look at shoes, suddenly you’re engaging with the sales process. There’s a, not a person, but there’s a personality you’re engaging with. And in that case, the goal of the personality is not really just to solve whatever customer service problem you may have called in for. In that case, the goal of the personality, the conversational AI, is to sell you. It’s a sales agent. Well, that’s a much more complicated problem. And there’s a lot of steps to this. There’s providing information. There’s education. There’s understanding that this may be an ongoing interaction. It’s not that I just logged in today and I’m looking at shoes. It probably knows that I’ve looked at shoes in the past. It’s maybe we’ve had conversations before and it’s slowly moving me through the sales process, educating me, building my interests. When I get to the point of sale, trying to convince me to do this, after I make the sale, after I do the purchase, after sales, do I have a problem? It’s a multi-touch process that this conversational AI is starting to manage. as a process and as sort of an emotional engagement. Am I having fun doing this? And so the analogy that Mr. Ho had sort of brought up was, you know, this is more like AlphaGo. This is more like an AI for chess, that I’m at one side of the chess board or the Go board and the AI is at the other. and it starts the game and then we get to the middle of the game, lots of pieces are moving around and its goal is to win the game at the end, checkmate, I made the sale. So that is a much more complicated strategic process that happens probably over a period of time as opposed to hey my package didn’t arrive. So when you go from customer service to sales, the conversational AI gets more complicated but it still has the two components of solving the problem, in this case making the sale. but also entertainment, emotion, managing that throughout the process. So that’s pretty interesting. And you can see as you deploy this into different human machine interfaces and interactions, that could be very different. And this brings us sort of to the last thing we talked about was this idea of, look, this is just gonna be a capability that exists in the world. Started with chatbots, goes to sales agents. human machine interface, you talk to your door. Hello door, I’m home. Well, it wouldn’t be the door, it’d be your AI for your home. You know, please open the door. And the purpose of the conversational AI is not just to solve your problem, but to make life interesting and engaging and emotionally satisfying. And he had mentioned something that really got my attention. He said, you know, maybe in the future, every single product will have its own personality. which gets me to the tagline for this episode. What will it be like when your shoes have a personality? Like when you go online and you go to the Nike store, because this AI capability can be given to all the merchants and brands and retailers, and they can start to put this in their own little stores. And I go on Nike and there’s a new set of Nike shoes. Actually, I like Adidas better, because I know the Adidas people in China and they’re awesome. Here’s the new type of Adidas. And the Adidas shoe itself, the new product, the new thing they’re launching this year, has a personality that I can talk to. Hello, I’m the AI for the new Adidas, 3D printed custom shoes. And I can talk to the shoe. However, this is why you should buy me. This is why I’m cool. And the AI could try to convince me along the lines of, this is why I’m better, functional, or it could just try to be funny. It could try to convince me to buy it, not because it’s an inherently better shoe. Maybe it’s a pretty basic shoe, but it’s got some sassy, funny personality. And, you know, we could have all our products in life, tables, shoes, sweaters, hats, all have different personalities that engage. And one might be the 20 year old sassy young woman who talks about what’s popular on YouTube this week. And the other could be a middle-aged man who talks about why this makes his feet feel better because he walks a lot and his feet start to hurt. So there’s this funny idea that like maybe all products and services will have their own personalities. And is the goal of the product, the product itself, you know, that’s the pitch. Or is it that just the personality is funny? What if one product is really funny and it’s just fun to talk to because they’re funny? Maybe every product will have its own personality. Maybe every store will have its own personality. Maybe every business will have its personality. Maybe every consumer, which would be me in this case, when you call me or any company tries to engage with me, they don’t engage with me anymore. They engage with my conversational AI that I’ve set up to represent me in the real world. And I will give it certain functions and capabilities, and I’ll also give it my own personality that I like. So that was most of our conversation. And I’m going to put up some slides that he had presented. It’ll go out to the members in the next couple of days. But really, you can see that this started out in something kind of standard, like customer service. And it’s evolved into something very, very interesting. And that’s a lot of what AI is going to be. It’s going to start out as something very functional. And it’s going to open up new ideas nobody ever thought of before. So. Anyways, I’ll do another, I spoke with another gentleman over at JD who was doing AI in another area and I’ll do a podcast on that shortly. But this was a good start to this. Let’s get into sort of the basics of AI and why everyone’s talking about it. Okay, so AI is probably, I probably spent 40% of my time digging into this. This is where my own sort of frontier, where I am in my own understanding. is I think I’m not a coder. I don’t know how to write in Python. I don’t do any of that. I don’t care about that. I’m looking to understand it as a tool and then to track where it is being deployed in businesses. So I’m always hunting for use cases across all types of industries. China, US, everywhere. Literally every day or two, I’m looking at various AI use cases to see what this thing can do as it continues to evolve. That’s sort of a business guy’s approach to this stuff, but I don’t know how to code. And there was a pretty easy way to think about this, which I think is accurate, which was a book that’s worth your reading. It’s called Prediction Machines by Ajay Agrawal, who is a professor. And this is kind of how business people are thinking about it. And his argument, which is very good, is look, AI is just cheap prediction. That’s what it is. Well, he says it’s prediction machines. I say cheap prediction. Whenever I hear the word AI, oh, we have a new AI program in customer service. Okay, you have a cheap prediction system in customer service. I just mentally replace the words whenever I hear AI with cheap prediction. Because that’s really what it does. And this book, they basically argue that it’s taking something that has always existed, prediction, and it’s just making it cheap. If you want to study the market and try and figure out if your product is going to do well, you might do a market study, you might interview people, you might have a lot of MBAs take the question apart. That is a process of prediction. Well, AI just does it much, much cheaper. And you could look at something like we have our taxi cabs all around town, we have our DD drivers all around town. Where should we position them? At 4 p.m. today. as it starts raining, how is the rain going to change where the drivers should be such that we maximize rides? That’s a prediction problem. That’s the kind of thing that, you know, an AI system at DD would just pop up on the screen of some manager and say, we recommend that you redeploy these 26 different cars from this location to that location. We predict if you do that, you will get 4% more rides. and then the manager will click approve or not approve. So it’s always predicting, but that doesn’t mean it’s making decisions. Usually the decisions are then given to a manager who makes the call. So think about it that way. Netflix, always showing you different shows. It’s making predictions on what you wanna watch, and then it sees, do you click on the show that it teed up next? And if it does, then okay, it feedbacks. has a feedback loop and it gets smarter. So it’s just this constant stereo of predictions. And because prediction is being done by software and not people, it’s very, very cheap. The same way spell check is very cheap. And this gets back to sort of the crazy economics of digital, that once something is done by software, the cost of reproduction, the marginal production cost is zero. So it’s very easy to deploy everywhere. And that’s kind of what’s happening. We’re seeing cheap prediction machines. being deployed everywhere. However, keep in mind, and this is really important because AI gets hyped all over the place, AI is stupid. Like it’s really stupid. It’s mindless. Like when they say artificial intelligence AI, think the appearance of intelligence. It looks like it’s intelligent because it can make predictions about what you want and what should happen and be right. It’s not actually intelligent in any way, shape or form. It’s stupid. A good story I heard about this that I thought explained it very well. It’s like, imagine wherever you’re from. If you’re not from Japan, imagine you move to Japan. You get a job as a customer service representative sitting in a cubicle with a screen computer and lots of messages come in about complaints. I have a problem, just like what we talked about. And so you see the complaint on the screen. written in and based on what you see the complaint that the little symbols you pick up a big book and you start flipping through this big book to look for okay when I see those symbols what should I reply and you flip through a massive book to find that and you don’t read Japanese at all you don’t know what the question actually is you just know the symbols that showed up on your screen you flip through a book you try and find one that matches that and then you type and then you see if it worked. So your prediction was this would solve that problem and based on what the person then does next, you either say, okay, that was successful or not successful and you do it again. Okay, AI is just a massive book. Like instead of one person sitting with a book of answers to various questions that pop on the screen, the book is insanely big. It can do thousands, millions of requests per minute and it gets better and better based on the fact that if I put in this answer, I get a success rate of whatever. If I put in this answer, I get a low success rate. So I should use the first one. It iterates and gets smarter and smarter. That’s AI as cheap prediction. It’s automating that person sitting at the terminal instead of a person which costs you money, it’s software which costs you nothing. That’s AI. But at no time do you ever actually understand what in the world you’re talking about. You have no idea what the question was. You have, we could be talking about soap, we could be talking about a product, we have no idea, we’re just looking at symbols on the screen. There is zero understanding happening. It’s just a very quick statistical sort of inference thing. So keep in mind when people talk about AI, keep in mind it’s 100% mindless. Artificial intelligence has no intelligence at all. So even though it’s very good at predicting where these cars should be, the AI has no idea we’re even talking about cars. It’s just symbols on the screen, like me or you, looking at a screen full of Japanese. That’s it. Okay, now that’s sort of the first thing to keep in mind. All right, why is cheap prediction so important? Now, Professor Aguil, he sort of lays out a very good analogy that when the transistor was invented, You know led to semiconductors what it basically did is it made arithmetic cheap the same way AI makes prediction cheap transistors made arithmetic cheap. That’s why you can do calculations on your computer and that used to take people and you know doing abacus and things like that. When that happens sort of three things happen in usually in sequence the first thing is. Arithmetic got a lot cheaper and so. companies that were already doing arithmetic, like big companies, the military, government, where you have people sitting in queues, massive warehouses full of people, all with pens and paper, doing arithmetic to track where all the boxes are moving and to track how many employees they have. All of that became dramatically cheaper to do because a computer could do it. You could have a transistor calculated or eventually a spreadsheet. That all got a lot easier. So it got cheaper. Very, very good. You used to have thousands and thousands of people sitting in office buildings calculating all our inventory as a massive oil company in 1920. Now we have much fewer people because they’re using machines like calculators. All right, so that it makes the current activity arithmetic cheaper than it was before, which usually means getting rid of people because it was people that did it. Interesting. The second thing, you start to see arithmetic deployed in places we’ve never seen arithmetic before. So photography, photography used to be chemistry based. You take a photo, it’s a lot about the chemistry. It exposes, I don’t know how photography works. It exposes the sunlight to the film. There’s a chemical reaction and then you get a photo. Well, arithmetic, which is data, digital, suddenly you could do digital photography because of that process. And so we started to see digital cameras. That’s cheap arithmetic being applied to photography, an area that had never used arithmetic before. Digital cameras emerged, the chemical cameras got wiped out. I just gave you an example of that with JD. Cheap prediction chatbots enabled them to use fewer customer service reps in Suchien. They’re doing prediction. This person has a problem. I have to make a prediction on what the solution would be to solve their problem and I have to make a prediction on hey they’re unhappy, how can I make them happier? I then do that and see if it worked. That was done by people because people are good at that. Now it’s being done by chat bots. Now it’s being done by conversational. That is phase one here. Let’s take something that’s already in the prediction business and do it cheaper, which generally means replacing humans with software. this cheap prediction to areas that haven’t had prediction before, like let’s have conversational AI, and we’ll have products that can talk to you, and we’ll have sales agents sort of manage the process for how you sell over time. Well, I guess sales agents would be more like phase one. That’s kind of replacing current prediction. But let’s talk about having products that have their own conversational AI and suddenly you can talk to your shoes online before you buy them. Not the actual shoes, but the brand. The brand will have its own conversational ability. And we’ve never seen AI. We’ve never seen prediction in that space before. Cheap prediction. Okay, so we’ll start to see cheap prediction applied to places we’ve never seen, like finding taxis on a rainy day. showing you movies, things like that. We’ll start to see it pop up in lots of different businesses where prediction hasn’t been a big deal, but it’s cheap now so we can deploy it easily the same way or cheap arithmetic could be deployed. And the third sort of impact phase would be it changes the value of other things. When cheap arithmetic got created, it basically was devastating for substitutes. So what’s your substitute for doing arithmetic with software versus doing arithmetic with a human? Those are your two substitutes. Well, it made the value of a human doing arithmetic much, much less. You don’t pay someone to sit around and doing math for you anymore, like with a piece of paper and a pencil. So it was devastating to the value of a substitute. This is radiologist’s problem, that if you wanna have someone read a chess film, You can have a piece of software do it, or you can have a very expensive radiologist do it. That’s a substitute. Well, it’s gonna dramatically decrease the value of a substitute like a radiologist and their $300,000 per year salary. Unfortunately, that’s true. So it’s devastating to substitutes, but it also increases the value of compliments. For example, digital cameras went up. in value, obviously, but so did the sharing of photos. Suddenly a company like Facebook makes your photos, if you’re going to buy a camera, what is the perceived value to you of your camera, your digital camera? Well, it’s I can see the photos, but it’s also more than that because other things increase the value of your camera. Like I can take a photo and I can share it with my friends. I can put it online. I can make advertisements, things like that. It’s the same way like electricity became more valuable when compliments like appliances in your home got invented because suddenly my electricity could enable more things to happen in my house. So as you sort of go from regular arithmetic to cheap arithmetic, a lot of compliments to arithmetic, especially data, became more valuable. Well, that’s what we’re gonna see in cheap prediction. This idea that now that prediction is cheap and the substitutes for prediction like a radiologist became less valuable, the complements to prediction are gonna become more valuable. I can now put a internet of things, an IOT sensor in my factory. I can put them all over the place. That would be a compliment. They will gather data with those sensors and that data, both complements. Prediction becomes more valuable because I can predict what’s gonna happen to various machines in my factory. It enables that cheap prediction to have more value because I can deploy it to more places. And that’s pretty much what we’re seeing. We’re seeing a lot of these devices data are making cheap prediction much more valuable. So they’re all compliments to each other, autonomous vehicles, things like that. And I kind of teed up that idea with Jingdong. Okay, you’ve built an AI, conversational AI that does cheap prediction and the first step benefit. Okay, you don’t need as many people in your call centers, fine. Second benefit, okay, we can start to deploy it in places that haven’t used Cheap Prediction like the sales process or I don’t know what they’re used for. And then third, it’s going to create complements. You can start to give it out to various merchants on your platform that are going to start to put it in places we’ve never seen and they’re going to offer things like AI speakers in your house. I don’t know, AI in your shoes, AI in your door, you can talk to your home when you’re away, who knows where this is gonna, but we’re gonna see a lot of compliments developed because of this. So we’re kinda halfway between phase one and phase two, probably right now. Anyways, I’m summarizing the thinking of Professor Agarwal. I’m probably doing it badly. So one, that’s not my thinking. I don’t get credit for that. Two, I’m sorry, I’m probably, he could do it 10 times better than me, but anyways. That’s really, I just wanted to tee up the idea at AI today and we’ll build on this and build on this and build on this, but think about it like cheap prediction and it has a couple benefits and yeah, it’s really fascinating. And e-commerce companies are sort of on the frontier. So that’s why I’m keeping an eye on JD and these companies and what they’re doing and why we did the call. Okay, I think that’s enough for today. I just, let’s sort of reiterate. I mean, the main takeaways for today, stuff to remember. AI is cheap prediction. That’s a big idea. We’re going to build on that, build on that, build on that. And it’s rippling all throughout the digital world. It’s really pretty cool. So the sort of the basic way to think about AI. I talked about mismatched and or crippled scale. This idea that so much of what we get excited about in these digital businesses is their scalability. But the problem of mismatched and or crippled scale hasn’t really been brought up that much because the early movers kind of just avoided this problem. They chose the businesses that didn’t really have that because they kind of got there first. So they chose the ones that were the easiest to do, like Netflix and things like this. They kind of dodged the hard stuff and they went for the easy stuff. But it turns out this idea of mismatched or crippled scale has always been there. We’re gonna see it more, they just kind of avoided the problem for the first 10 years. And now they’re going to go into more difficult businesses. And this is going to be more and more of an issue. Fraud, content curation, content moderation, things like that. So we’re going to see this more and more. But yeah, those are the two ideas for today. And those both go under learning goal 22, which I’ve labeled the basics of AI. We are definitely going to build a lot of this in level five of this class. So that’s it for content. I’m just relaxing in Bangkok here. I started teaching here at Sassen Business School in Bangkok, which is great. And it had a lot to do with the fact that, you know, students aren’t really delaying their classes. Like a lot of businesses can close down because of the whole COVID thing, but students can’t really do that because it, you know, suddenly impacts graduation and suddenly you’re kicked back to another six months or a year. It’s, you know, it’s pretty hard. So most universities are not. they’re not closing and saying come back later, they’re just shifting online and trying to keep things going. So anyways, I sort of had known the Sasson folks a little bit and because of the lockdown, I guess they knew I was in town. So I think that was the purpose. I don’t think it was that they thought I was terribly awesome. Well, maybe they do, I don’t know. I think it was more because I was physically here and it became impossible to fly people in and out of the country. So anyways. That’s how that started, but it turned out to be a great class. I’ve got a couple of them and I’ve got about 110 students, executives, a lot of executives, and then some MBAs as well, talking about digital stuff from the Thai perspective, because so much of digital China is, is, you know, merging in with sort of Asia and generally Southeast Asia. So Lazada, which is Alibaba’s here. And it’s a really interesting sort of strategic question, especially because, you know, most of the execs who were in my class, you know, they work at more traditional businesses like shopping malls and, you know, car companies and manufacturing companies in Southeast Asia. So they’re thinking about all this from, you know, the perspective of a traditional business, which is really where I think most of the action is. We talk about the giants of digital because it’s fun and you can learn a lot, but really the future of all this is traditional businesses becoming more digital. That’s going to be 80% of the game. But you can learn a lot from the digital giants, the natives. So anyways, that’s been kind of fun, and talking with them a lot about what’s going on. Yeah, it’s great. Other fun stuff. You know what I’ve been watching? Like, I’ve been watching Daredevil on Netflix. Like that just killed my Saturday. I turned on Netflix and Daredevil came up, and I watched the whole season. This whole binge watching thing is really counter, I mean, it really wipes out your productivity in life. But it’s such a good show. Like, season three of Daredevil on Netflix, it’s just the best. Because if you grew up reading comic books, which I did, Daredevil was pretty cool, and Bullseye was really cool, the sort of psychopath killer. And season three has him in it. And they made him a full psychopath, like, all the way. And it’s just the best. So that kind of wiped out a Saturday. But yeah, I highly recommend it. If you haven’t watched Daredevil on Netflix and you like this sort of comic book-y stuff, dude, it’s the best. And of course, because I can never get off my main topic of digital anything. It’s so devastating for competitors. Like, I get to watch all these thousands of shows, including all these original shows like Daredevil that don’t exist anywhere but Netflix. And it’s an amazing show. It’s just the best. and I get all of that for $4.32 a month. That’s my Netflix bill every month, $4.32. How in the world do you compete with that? It’s just impossible. It’s such a devastating business model. Thank God I’m not competing with it. Anyway, so I was watching Daredevil, it’s the best. If you haven’t watched that, watch that. If that’s not your thing, the other thing I’ve been watching is The Godfather, which is, depending on what part of the world you’re in, is on Netflix. That’s on there this month. But it changes depending where you are in the world. And it turns out this podcast has been doing quite well. It really sort of been bumping up. And it’s from everywhere. It’s everywhere. It’s a lot of people in China, a lot of people in Asia, a lot of people in Singapore. But then a lot of people in Brazil now, a lot of people in Canada, the United States, Germany. I mean, it’s pretty much spread everywhere. I keep trying to get some data insights into like, OK, what’s doing well and what should we do more of. And I look at the locations and it’s just kinda everywhere which is fascinating, it’s great. But yeah, so if you’re on Netflix in Brazil, I don’t know if Godfather’s on there. Whenever I’m on, my Netflix account is out of Colombia because I signed up when I was in Bogota. And whenever I sign in and I see a lot of narco stuff, like half the stuff I see is like Colombia this, Mexico that. Anyways. That’s it for me for this week. Thank you for listening. Thank you for subscribing. If you are a subscriber, do your homework. You know, stay on, keep moving, keep moving, step by step. And if you’re not a subscriber, go sign up, There’s a 30 day free trial and that’s it. I will talk to you next week.

Leave a Reply