Joshua Gans: Using AI to Manage Uncertainty in Business

Joshua Gans, the co-author of “Prediction Machines: The Simple Economics of Artificial Intelligence,” discusses how AI could help business professionals make better decisions.

[powerpress]

Full Transcript:

Announcer: The following program is brought to you by the Ed Snider Center for Enterprise and Markets at University of Maryland’s Robert H. Smith School of Business. Welcome to “The Inc. Tank.” Stay with us to get the inside scoop about technologies that could disrupt and challenge the way you do business.

Christina: Hello. I’m Christina Elson. And on this edition of “The Inc. Tank,” we’ll be discussing the extent to which artificial intelligence can help businesses make better decisions. Can AI be utilized as a virtual prediction machine that reduces uncertainty in the marketplace? Our guest today, Joshua Gans, thinks so. He co-wrote the book “Prediction Machines: The Simple Economics of Artificial Intelligence.”

Joshua, good morning and thank so much for joining me today in “The Inc. Tank.” Let’s start off a little bit helping to frame some of the concepts for our viewers. And so one of the things that you’ve spent a lot of time thinking about is the concept of disruption. This is clearly a very vogue topic. It has been for a while. There’s a lot of different approaches to it. You have a slightly different take on disruption in its role in furthering renewal and business renewals. So perhaps we could start with you telling me just a little bit about your thoughts.

Joshua: Disruption is one of these words and we’ll come to others that have generated a lot of hype, have somewhat changed their meaning over time. Disruption used to be a bad thing. You didn’t want your children to be disruptive. Now, apparently, you do. So, it’s a funny term in that regard. I think it invokes two things. One is it invokes the spirit of entrepreneurship. You should be out there moving fast, breaking things. And the other is that it denotes the bad thing. If you have an established business, you don’t wanna be disrupted. That is something that may be considered an existential threat. And I think this really took off some 25 years ago when people started suggesting that, in fact, the firms most vulnerable to disruption are the ones that are managed the best, that did all their decisions the right way, that did what the best business advice and things would do and still do to that degree, yet is still vulnerable. We’re comfortable with the idea that firms might go down because they’re stupid or corrupt or something like that. But they might go down simply because they were doing the best that anyone could do. Well, that’s gonna keep a CEO up at night.

Christina: Yes, clearly. And so your view is a little more that it’s possible to manage your way through disruption. Is that…?

Joshua: Yes.

Christina: Yeah, okay.

Joshua: There was a tendency to think of that as it was a dire event. It was a all hands on deck. You must divert all resources. You must plug every hole that could potentially be plugged and things like that. Whereas actually the reality is while disruption has occurred, there are great cases. There is Nokia, there’s Blockbuster, cases like that. It is the case that the majority of well-managed firms can see disruption coming and can come up with solutions to at least preserve themselves throughout that. They may not be happy about it. I mean, it’s not fun. No one’s saying anyone’s gonna have fun managing disruption, but it is manageable to a great degree, not perfectly but far more than people were saying previously.

Christina: Yeah, and so that’s a great point, because in the strategy classes that I teach at University of Maryland, we do like to use the Blockbuster case, of course. And it’s all very dire. It’s like you don’t wanna wind up like these people. But at the same time, we also wanna look at the IBMs of the world that maybe that’s a company that you would perhaps point to.

Joshua: Yeah. IBM’s a complicated case. So Blockbuster, we like that in classrooms because it’s very straightforward. Everybody understands that we know the thing that replaced it. I guess now as the years go on, we have to explain to students what a videos store is.

Christina: Yes. I showed them the picture the one in Bend, Oregon, the last one, yeah. Yeah, this is a store.

Joshua: Well, of course, it was disruptive. This seems like a stupid idea. That’s an easy one, although Blockbuster has a nuance is that the Blockbuster saw all this. It saw flaws in its own model. The problem is sometimes the new thing that comes along is so completely new that new entrants, they just have to build something, but incumbents have to both build and destroy something. Destroy first and then build. And that’s really hard. In the case of IBM, they had this problem with respect to the personal computer. They were obviously larger computers and they were dominating the world. And then they saw what Apple were doing and they largely invented the personal computer. And they did so by setting up a completely separate division. For many years, it wasn’t Mac versus PC. It was IBM versus Mac. That’s what 1994’s Mac ad was all about. It was about IBM. They were big brother. But then what happened was the PC started to encroach on the traditional business. And as soon as you have that happen, there’s…in any organization, there’s just a plop of politics occurs. How that gets resolved is quite difficult. And IBM actually chose to resolve it in favor of the traditional business, which, of course, people can argue over whether that was right call or not. But that set its path forward.

Christina: This is very interesting point, because as these technologies come in, and you’ve been writing and thinking a lot about AI, they do impact the economy and so there’s a lot of debate and discussion around how the disruptive nature of these technologies. Is this really going to change our basic economy, the principles that we’ve all studied and thought about for a long time? So, help us understand a little bit about that. Give us like a couple minutes of basic economics, what we think now and your view on what you think might change or not change in terms of trying to analyze this.

Joshua: As an economist, there’s one thing that we’re boring about. We tend to think there are laws of economics and we don’t think they change. So in the 1990s when the internet came along and everybody said it was a new economy. Well, that wasn’t us. It looked like supply-demand. Some costs had changed. Some demands had shifted so people would be doing different stuff but the laws remain the same. So with respect to artificial intelligence, there was that same issue. Now, people like to think of it in sort of like, “Oh, artificial intelligence. That means intelligence that’s artificial. That means that a machine can do now everything that I do.” People sort of think about it as a war of machines versus humans in that regard that that’s the economics. I mean, that’s one way to picture it. But it’s not the current reality or anywhere near close to where we are.

What artificial intelligence is, is machine learning. And when you look at it very closely and you take away the hype, what it is, is an advance in statistics. A lot of people sitting there at their job saying, “Well, there’s a new statistical method, that’s it for me.” Most people use statistics. But really that’s what artificial intelligence currently is. It is an advanced statistics that allows us to have better, faster, and most importantly cheaper prediction. So the same forces that allow you to have a better forecast of the weather, it turns out that that way of synthesizing information is applicable to a lot of different problems, including, for instance, like machine translation predicting. You hear some conversation in French. You predict what that conversation is in English. That’s prediction problem. And so that is what artificial intelligence is.

So, if you say, “Okay. We’ve got this reduction in the cost of something.” Well, the laws of economics tell you two things are gonna happen. One is we’re gonna use more of it and two is we’re gonna discover new ways of using that tech. That’s the sort of potted version of the sort of simple economics that we see driving decisions and other things for businesses on AI certainly in the near future.

Christina: So that’s a great point and the idea that the predictions are going to change the way that some people are going to need to do things. There are certainly some job areas that we don’t necessarily need humans to sit around do those predictions. We can do those with machine learning. So what are some examples that you see as really ripe for this kind of takeover of prediction?

Joshua: I mean, it’s a takeover in the sense of a tool [SP]. I mean, we make decisions all the time without knowing everything we need to know. When you don’t have a prediction, you make a guess or more importantly, you form a rule. For instance, if you didn’t know what the weather is gonna be like when you’re traveling to another city, depending on your risk preferences for getting wet, you might pack an umbrella or not. That might be the thing. You might just have a bag that always has an umbrella in it or one that just never does. That’s just how people are. But you hand people a poor prediction, there’s now a third option. You can make a decision what we call as contingent. If the forecast is for rain, carry the umbrella. If the forecast is not for rain, don’t or with some probabilities in between. And different people might make that call differently. But there’s this third option rather than the rigid rule take or not take, it’s now a contingent rule. And from what that means from individuals’ point of view is they’re making a better decision.

Every time you follow a rule, you’re always going to make a mistake. But when you have a prediction, you can reduce the probability of a mistake. That’s the application to umbrellas. I always use that example because it’s very easy. It turns out most statistics do. Most statisticians are obsessed with umbrellas and getting wet. But every decision…there’s so many decisions that involve uncertainty and some of them we know about. Like if you are in a bank and you’re trying to assess the credit risk of somebody you’re gonna lend money to, well, that’s a decision on that uncertainty. And if you can get more information that informs you whether they’re likely to repay, you’ll make a better decision. In other situations, you don’t even know that you’ve set yourself up to avoid having to make a decision under uncertainty. You’ve protected yourself against the uncertainty and you’ve developed habits and other things. Predicting where we already know we need to predict, that’s something machine learning is currently doing. Predicting where we don’t currently realize we could benefit from predicting, that’s the real place where the gains are gonna be.

Christina: What’s an example of this second one? Because in the first category definitely like the more inputs we have for making a medical diagnosis or the more information we can have to make financial decisions for loans, of course, thinking through how we wanna train algorithms so that we’re not injecting bias and all that stuff. There’s a lot of thought going into that. That’s totally clear. So the second category, of course, is very exciting because that seems like that’s where some of these opportunities might be opening up for entrepreneurs. So what are you seeing there?

Joshua: It’s really early days on that front. The clearest example thus far is self-driving cars. For a decade and a half, people try to make self-driving cars. I said, “How hard could it be?” Any toddler understands that if there’s a blockage in the road, just stop. If there isn’t, you go. You turn here. You turn there. How hard could it be to do a self-driving car? Well, the problem wasn’t the sort of programming if/then statements for those sorts of obvious things. The problem was that the environment was far more complex than people were appreciating. There’s just a ton of stuff going on. It’s a miracle we as humans are able to deal with it at all.

Christina: And talk on the phone.

Joshua: So somewhere in our lives, we’re always predictions getting a feeling that the car in front of you is gonna slow down or another car is not gonna stop and it’s gonna go into an intersection and turn here, and turn there, and I might wanna go in a different direction, or there might be some road conditions, or there’s a person there, like just add it all up. And we’ve seen that stuff before. I mean, if you’re trying to teach a teenager to drive, they understood the principles but the thing you’re shouting and screaming for your life about is their poor ability to predict. That’s the thing because you could just see more than they can see. So the advance for self-driving cars was they understood that they were taking all this information and then they said instead of trying to get the car to understand really what a person is and what they might do and have some model of it. They instead said, “Well, why don’t we let these cars just watch people drive. A bunch of sensor inputs comes in and we see where the people break, turn, accelerate, etc. And we collect millions, billions of data points on that in all sorts of situations.”

Well, it turned out that that’s what artificial intelligence was good at. It could come up with a prediction with these very complicated environments and inform the cars how to react. And so the cars can react in pretty much almost all the situations we can. I think probably up to the level of an average driver. They can’t go to the best drivers and what have you. There’s a whole lot of other issues, but they can do it. There was essentially taking this problem and reformulating as a prediction problem. Now, the bad news is they’re not gonna do any better predicting than the best human driver necessarily or they might do as well as the collective wisdom of the best drivers. So they’re not gonna necessarily be able to shoot at lightning speed, or we’d get rid of stoplights because everybody kind of knows how to sort of weave through that. I don’t think we’re quite there yet but that’s still quite an advance.

Christina: Yeah, exactly. I mean, that’s one area where we will see a lot of change and progress and more market development in that there. In the book, in “Prediction Machines,” you and your co-authors talked a lot about how prediction is going to provide the ability to do things sort of better, faster, cheaper. And so that’s a big…of course, for an entrepreneur that’s looking to think about how they can compete against established firm or someone who’s already been in the market. What are some of the things that you think are opportunities that you’re seeing for entrepreneurs?

Joshua: I think the opportunities are actually quite huge and for the most part sort of the lowest hanging fruit are people who have domain experience in some sort of decision class that is across a wide variety of firms, and being able to graft on a prediction machine to that, and therefore be able to sell that as a tool to existing firms. Well, I think that’s a good solid opportunity. There is this separate issue of it turns out that prediction will be the solution to a long-standing problem. I think that will occur. I’d be an entrepreneur if I could tell you what that would be. But let me give you an example.

So for instance, Google is great. Google does a wonderful job at certainly better, if you remember when it came in, better than anything was previous and certainly as at least it’s gonna be probably better than everything else that currently is. Is it the best that you can imagine in terms of being able to find what you want? I don’t think so. As soon as you ask that question, you say to yourself, “Well, what is it that’s preventing Google from becoming that.” And maybe somebody applying artificial intelligence and conceiving of the prediction problem the correct way will work out how to…as simply as Google did, come up with a different way that leads to answers being given better far more accurate than what Google does now. So when you think about it in those terms you say, “Well, big heavyweights, they’re not beyond,” I guess for lack of a better term, “disruption.” They’re not necessarily immune to it.

Christina: Yeah. And that’s a great point because there’s a lot of concern and discussion about whether or not we’re going into an economy in the environment that is just going to be totally dominated by these large firms and it’s gonna be hard for competition. And we’re seeing a different kind of economy. And that engenders a lot of fear, not just in business but just in sort of the normal, “What am I going to do? Will I have a job if I can’t work at Google?” And I see that with students. They really are concerned about what are the opportunities.

Joshua: History tells us that, yeah, for sure, we do not like this monopolization, but we’ve had it all the time and things come and things go. Microsoft was monopolist over an important part of the IT system throughout the 1990s. However, the thing that they were the monopoly over was just a fraction of the spent of IT. IT budgets were 100 times what they were handing over to Microsoft. And I would say that would be the same to everything else. Facebook, Google, Amazon, Apple, etc., it’s just not that high a fraction of our expenditure. However, it’s a big fraction of our attention. That’s where Facebook comes in. That’s important, although even there if we added up all our attention, it can only be so much.

Christina: Right. Yeah, hopefully.

Joshua: And it’s only for advertisers how they get through to people with critical channel [SP]. So those things are ever present. There’s just so much more of the economy going on around these large firms. And even Microsoft in the 1990s, it was a shadow of what it is today. Microsoft have…as much as Amazon, they’re probably in a few years gonna be a trillion-dollar company as well. And people don’t even count them. And it’s just extraordinary to me, yet they’re everywhere. They’ve done this excellent job of downplaying their ubiquitous.

Christina: Their presence in the market.

Joshua: I think people do get concerned about that and I think there is a reason to be vigilant and I certainly would not detract from that. But I also think history has told us that change comes.

Christina: Yeah, I like that diminish the fear and think about the progress and how we can encourage that. We had an entrepreneur on the show and he was mentioning that his firm, this fella Mark Walsh is using what he referred to as off-the-shelf AI. And I just love that term so much because the idea that this technology is becoming more and more accessible to people who can, as you’ve pointed out, use it in this sort of better, faster, cheaper and really get in there and see what they can do with it. It was just fascinating.

Joshua: It is a glorified statistical package. And off-the-shelf AI sounds good but that’s great news. It’s just like we use off-the-shelf cloud services or something like that. It’s just telling you that that’s not where the innovation is. The innovation is in finding the ways to use that off-the-shelf stuff to actually generate the new products. It is an input. It is not just a thing that walks around doing AI. It needs to be molded. It needs to be targeted. It needs to be trained. It needs to be all of those things. And there’s a lot of art involved in it. It’s not just like science, like anyone can do it. There is actually still requires skill and insight. And I think this is underappreciated.

Christina: Yeah. So let’s talk about that because you all do discuss it in a really meaningful and insightful way, the role of AI in terms of helping business people think about strategy. But when I teach strategy, everyone’s like, “Well, what is strategy?” And I’m like, “Well, hopefully by the end of this class you’ll understand what strategy is but I can’t really tell you at the beginning.” So tell us a little bit about this relationship for strategic insight.

Joshua: For most businesses, most of what AI is doing at the moment is impacting on just how you currently do things and doing them a little bit better so you can improve efficiency. You might be able to personalize a product more or something like that. These are real incremental changes. They have to be aligned with strategy. I mean, you don’t want people just doing any old thing but that’s just your normal strategic management. Where they come to have an impact on strategy is when areas where prediction becomes so good that you might change the way you do things in a fundamental way. It might mean that your organization has developed different capabilities and it might have different markets that it’s competing in and so on. You might have importantly a different way in which it’s organized. It might become more vertically integrated or less vertically integrated depending on the case may be. And the example we give in the book is it’s a thought experiment because we’re not there yet. We say think about online shopping. Currently an online shopping, it’s great. You go to a webpage. You spend time searching for what you wanna do. You look at reviews, etc. You click and buy something. A few days later, it turns up. Not too dissimilar from the way things have always been done.

But imagine for a moment that an Amazon became so good at predicting what you want to buy that instead of waiting for you to go to a web page and click on that stuff, it can just send it to your door. And you get home and you open a box and you say, “Oh, hell, I needed dental floss. Yeah, I’ll take that.” And it can’t currently do that. I can’t bring the whole door to you. It’s just too costly. But if it becomes really good at predicting, then it can do that. Okay. That’s great. But think about what then Amazon will need to do. It will need to completely rethink its logistics. It completely rethink about how it has delivery drivers and UPS and all that sort of stuff because not only are they gonna drop stuff off but they’re probably gonna be picking stuff up a lot as well. It can think about where things are and its whole distribution network. It can think about what Amazon Prime means for consumers, how it gets a payment, how it stops people from defrauding it or something like. There’s all sorts of other things to think about as part of that. And Amazon aren’t gonna introduce that idea without sign-off and it being understood in the strategy right at the top. Now, it may experiment it with it like it’s currently doing with Amazon Go, the little shops that stuff when it gets new thing and we may see Amazon starting to experiment like this with some areas. But flipping the switch and saying, “Now, this is our company,” that’s a big strategy decision.

Christina: That’s something that, as you’re pointing out, can affect so many different facets of the company. So as a manager and an executive in this position, we get back to these sort of core competencies or core skills that you also talk about, the complementarity of skills. And I’d love to hear your approach to that.

Joshua: Think about it this way. We have jobs that are comprised of tasks, a lot of little tasks and in tasks that we have decisions that we need to make and actions that we have to take. And so we drill. Forget the jobs. We drill all the way down to those tasks and decisions. And as part of a task and decision, well, one element of that is prediction. You want to be able to make that decision with the full information you can. So that’s where your AI comes in. So if somebody was doing prediction, AI can come in and start doing some of that. And that might happen with credit checking and HR functions or whatever. But at the same time in order for the whole decision to be made, all sorts of other things that we have. You have to collect data to be able to inform that prediction. You have to train the prediction thing in the first place. Okay. So these are things that…the value of which increases result of being able to predict. You have to take actions afterwards and get the outcomes. And then finally, a decision isn’t just prediction. A decision is understanding what the objectives are, what the tradeoffs are, and things like that.

I mean, we’ve all seen this. When you use navigation software like Waze to find the quickest route, it will give you the quickest route. If route A is 30 seconds quicker than route B, it’s gonna tell you route A. But if you are a driver of a car, you know that speed isn’t everything. I want route A for sure if it’s 30 minutes quicker than route B. But 30 seconds quicker, well, I can ask myself some things about route A. For instance, is route going through all these back streets, back and forward, running around and I have to pay tons of attention and constantly look about turning versus route B is straight down to main roads, and sure I’ll hit some traffic but that’s easy. That happens everywhere all the time. Even if you come up with a prediction of something, that does not automatically lead to a decision. There are all sorts of other factors that way. And while the self-driving car makers have to do all of that stuff. That’s what’s really hard. For most of our decisions, we’ve still gonna have a human in the loop to just weigh those tradeoffs. You might not even appreciate those tradeoffs and still use that getting predications, things like that.

Christina: That’s a great analogy because…I love Waze but I do call it Crazy Waze. And in strategy particularly in business strategy, you want it to be clean and elegant and not full of crazy ways decision like making processes. So that’s a great analogy.

Joshua: No computer, no prediction thing knows any of that stuff. The best that we can hope for is that computers might look at a generic person that has in common everything of all of us and could do it that way, and driving is about as close to that because most people share driving preferences. If you wanna go quickly, you don’t wanna get killed, etc. That’s easy. But choosing what to watch on Netflix or something like that can’t be done. In the end, you gotta sort of do it. People still complain that what Netflix’s prediction is not good enough. It can’t capture their feelings at that particular time of what they wanna watch or who happens to be sitting in the room and all this sort of stuff.

Christina: Sure. So at the end of the day, what we’re looking at is that we want to encourage people to be comfortable with unpredictability. I mean, the prediction advances you’re talking about are gonna help us deal with that, but there is some uncertainty in predictability that we have to deal with and in terms of these other things like the emotional intelligence aspect and the creativity aspect, those are things that were still good for a while.

Joshua: If I want to, I can reduce emotional aspect to just another tradeoff. But either way, that’s what we’re talking about.

Christina: Last thoughts. I would love to hear where you think AI is gonna go in the next like 10 to 20 years. Where do you think we’ll be?

Joshua: There’s different bits and pieces of this. Will we ever get general AI sometime in the future? I think so because we have ourselves. Nature has produced this thing. We don’t understand it yet but at least we know it’s possible. It’s the proof of concept is walking all around us. You just look at animals. They’re pretty darn intelligent as well in this regard. So, it should be possible. So it’s all an issue of timing. And what is true is the current wave of stuff in AI. I do not believe is the thing that gets us to general AI. It’s an important piece of the puzzle. It’s gonna have lots of applications. It could be transformative for large parts of the economy and all that sort of stuff. But it’s not a fully reasoning, independently minded thing that we imagine.

And I know that there are technologists out there that say, “Oh, we’re doing really just predicting. Everything else is just an illusion.” And maybe they’re right. But I don’t currently think so. Currently AI is amazing, but I don’t think it’s so good. I think it’s really, really hard to replace people with stuff. It takes an awful lot of work and no matter how single-minded an engineer is to be able to do that. In fact, all the lowest hanging fruit in AI is working with people rather than working against them. And I think that’s gonna be a while before we get anything more than that. I think even when someone comes up with the right concept that can get this to general AI, the computers that we have are still way below what is needed to actually do it.

Christina: Yeah, so good takeaways. There’s still some room for the next generation of students to find meaningful work. We have some great entrepreneurs.

Joshua: I can think of five things that are gonna kill us all before that. I’d like to speed up general AI because it might stop us from doing all the other five things.

Christina: Oh, that’s a good point too, yeah.

Joshua: That’s my [inaudible 00:29:00].

Christina: Yeah. Okay. Well, thanks so much for being with me today and I wish you a lot of luck with your book “Prediction Machines” that came out in March.

Joshua: Yes.

Christina: It’s available, of course, on Amazon and other places.

Joshua: All the book stores, yeah.

Christina: I know that you’ve been doing some speaking about it. Hopefully, people can check out your website and social media and see where you might be next.

Joshua: predictionmachines.ai if anyone’s interested.

Christina: Yeah. Thank you. So I would encourage anyone who’s thinking business goal or…certainly I’ll be recommending it to the students in my classes at the University of Maryland. So, thanks a lot for your time, Joshua, and we’ll look forward to touching base with you again in the future.

Joshua: Thanks. This has been fun.

Christina: Okay. Take care. Bye. The far-reaching potential of artificial intelligence is still largely unknown. Exploring AI as a positive force in the business world makes sense. I look forward to sharing more about the tools we’ll likely be using in our daily lives in the near future. Thanks to Joshua for talking with me today. Until next time, this is Christina Elson in “The Inc. Tank.”

Announcer: Subscribe to “The Inc. Tank” on Spotify, Google Play, and Apple Podcasts. A special thank you to the Kauffman Foundation for their support. From the Robert H. Smith School of Business at the University of Maryland, thank you for joining us in “The Inc. Tank.”

This episode of The Inc. Tank would not be possible without:

Christina Elson, Host and Executive Producer
Stevi Calandra, Executive Producer
Podcast Village Studios, Production and Editing

The Inc. Tank Theme Song “Key to the Foot” provided by Clean Cuts Music Library
The Inc. Tank logo was designed by Kasia Burns

This podcast is brought to you by The Ed Snider Center for Enterprise & Markets and the Kauffman Foundation.