In this segment we're going to talk about the effects of AI, and particularly the societal level effects. Especially we're going to talk about jobs and the belief, the concern that the rise of artificial intelligence is going to eliminate jobs over here, over there, etc.. These are attention getting predictions. Some of them are extremely dramatic. I mean, we've heard things a few years ago that, 40 percent of jobs are going to go away in the next few years etc. The big takeaway right at the beginning is, at least so far, all those predictions, the attention getting ones have been wrong. But nevertheless they get a lot of attention. We want to talk about what we know, but also how to think about these predictions, because they're not going to go away. We're going to keep seeing them. Part of what we want to do in this section, is to get a sense about how should we think about these predictions, when we hear them. The first thing to know is to maybe spend a couple of minutes just talking about predicting the future. There's some things we're pretty good at. Actuarial science, for example. What is the likelihood that my house will burn down? Well, we can estimate that pretty well, because there have been enough houses around and enough of them burned down, and the circumstances are relatively similar, that we can assess based on attributes of my house, pretty accurately what the chances are that it might burn down in a given period of time. Economic forecasting does roughly the same thing. We're assuming that the future will look like the past. We build a model, like machine learning models in some ways, trying to predict something like the growth in the economy, GNP growth, based on a bunch of factors. We build that model if it worked pretty well for the past. Then we take it into the future by plugging in, measures for the variables in the model, today to see what that says, about growth rates. The problem with those, of course, is that if the future does not look like the past, the forecasting models won't work very well. That's an issue there. If we're looking at a prediction on a topic where there's not enough data to build a forecasting model, near future events, for example, or something like, what will the political reaction of the Democrats be to this situation something like that, then we're into the realm of asking people, market research and polling, if we're asking, trying to figure out how the election will go. Polling data is pretty good. How are you going to vote? Market research is pretty good. Would you buy this product if it was there? The problem is, sometimes we're asking about things which are bigger events. They don't happen very often, and there are things which polling is not going to work for. That is something like the introduction of new technology. These are expert judgment questions. Our colleague here at Wharton, Phil Tetlock, has studied these for decades, and he's run big experiments on them. He's also run some recently to try to understand what people with what attributes are good at expert judgments. That is predicting, for example, will the United States move toward the left? Will a particular country elect a different parliament? Will there likely be a war in the Middle East? That stuff. Here's what he finds about people who are good at this. First, people who are highly confident, who are deep subject matter experts, they know a lot about the topic, and people who are guided by theory, they do worse. People who do better, are people who are questioning assumptions, who people who consider the counterarguments, and people who look for similar situations elsewhere. They do better. When you're looking at a prediction from somebody or some organization, what you want to do is start looking down and saying, do they consider in their arguments the counterarguments? Do they question the assumptions that are built into the model that they're using, or do they at least articulate them? Are they looking for similar situations elsewhere? If they are, pretty good chance their judgment might be worth listening to. If they don't do any of those things, pretty good chance, in Phil's estimation this is like monkeys throwing darts, which is his description. This is just random. Where do the predictions come from? Well, here's something to remember about contemporary society. There's a big competition for eyeballs in the private sector. That means, if you can get attention, there's money behind that. The money tends to win too. If these are predictions by people who are selling you something, they have an interest in particular stories. Let me just point something out. I have yet to see a consultant report that says, things are pretty much going to be the same going forward. I've never seen one that says, not much to worry about here. The reason is, those don't sell. They don't get eyeballs, and they don't get much attention, even when they're true. Let me give you a couple of examples of these in the workplace that I have been around. One was in the early 2000s. There were a lot of organizations predicting that there was going to be a coming labor shortage. Of course, it never happened. That was because, initially they were misreading some of the demographic data from the census basically. Then other people started to pile on and say, yeah, coming labor shortage. Everybody had a report on the coming labor shortage. Of course, it never happened. People who understood demographics knew it was not going to happen, but that didn't prevent companies from building plans about how they were going to deal with this coming labor shortage, because everybody was talking about it. Let's talk about another one which is popular, and this is millennials and these generational differences. The National Academy of Science has published a report this past summer looking into this. They concluded basically none of this is true. There's no evidence of millennials even exist as a distinct generation compared to those born before or after them. There's nothing to any of this generational story and yet it persists. The reason it persists is it sounds cool, It sounds interesting. It sounds like things we know, even though there's no evidence that it's true and all the evidence we have suggests it's not true. Just a couple of years ago, we had this big concern that driverless trucks were going to eliminate truck drivers. I know companies where they had tasked their HR people to try to figure out what they were going to do with all their unemployed truck drivers because technology was going to eliminate trucks. These predictions were that it was going to happen very quickly. We haven't heard a thing about it since, maybe partly because other news was more important. Also we've seen no progress on that front. Let's talk about the general predictions about artificial intelligence. Let's back up a little bit, following my colleague Phil Tetlock and looking at what we know from similar situations before. We've spent a lot of time over the last generation. Looking at the effects of introducing information technology. What are the effects that that has had on jobs? Overall in the economy, has it eliminated jobs? No. The total number of jobs in the economy continue to grow. We'll talk about why that is in a bit. You might say, "Well, how about in particular occupations?" Certainly you can see some of that. The biggest one in my lifetime that I've seen is typist jobs went away. The reason typist jobs went away is not because word processing eliminated typing. It was word processing lead management to decide that even executives should now do their own typing. It was a management decision pushed along by software. That's going to be a key point. The biggest changes come when technology pushes management to make different decisions about work organizations. One of the things also to remember about IT and this will certainly be true going forward, is that it is not per se designed to eliminate jobs. In many cases, what it is designed to do is introduce new functions that didn't exist before, for example, even earlier on when you could buy in order things online, one of the things that you did not see was software making recommendations to you. That said, people who buy this also buy that. Have you thought about getting the tools that you need to put this thing together? Those recommendations that you get all the time did not exist before, didn't eliminate any jobs. The IT there added functionality. The effects of applying IT to existing jobs don't necessarily eliminate those jobs. That's something we want to think about pretty carefully. If we think about how this works, for example, and you think about how information technology and AI more broadly, what it does, a lot of what it does is not designed to eliminate jobs, even in the context of robotics. Robots, you would think are designed to eliminate jobs in assembly line, get rid of the person, put a robot in there. It turns out that is not happening. At least in the studies that we can see so far. If you look at the number of robots going into a place and you look at the number of jobs, they're not offsetting it. Part of the reason for that is what the robots are doing is helping take over one task out of the employee's. Lifting the tires on an assembly line, robots are doing that. They are doing eliminated welding early on. Here's a great story to remember, and that is the triumph of Toyota over General Motors robots. In the 1980s, General Motors decided that it was going to eliminate as many production workers as it could from its assembly plants because you couldn't trust workers and their quality was bad. They invested hugely in robotics, 43 billion dollars to put robots in their assembly plants. Toyota came into the United States and took over General Motors plant in Fremont, California, that had made Oldsmobile and Toyota started making Corollas for General Motors. They called them Novas, but they were Toyota Corollas made by Toyota with American workers and with the old equipment that was there from the 1970s using lean production at the same time. General Motors spent 43 million dollars automating it plant. Toyota's lean production is more productive, cheaper, and higher quality than the General Motors robots. It isn't the case that robots always beat employees even on things like quality and productivity. Part of the reason, as with lean production, is that people are really adaptable. Even if we think about simple tasks, low skilled tasks where you think equipment and IT and robots could do them, there's lots of things that robots could do that we're not having them do. One of the reasons why is labor is cheap. We can have room bots sweeping all our floors rather than janitors. Janitors are still cheap and it's easier to have them do it than it is to try to get robots to do it because robots are still expensive. Same thing with driverless cars. Is there actually a market for driverless cars? Here, the answer is not so clear. Sure, there are some people who would want them, but let's take the best selling vehicles in the United States. They are all trucks. We've all seen truck ads on television. The truck ads on television are some burly guy with a cowboy hat on, driving a truck, bouncing on a construction site, going through and pulling something really big at the end of it. Suppose you could make a driverless truck. Who's going to buy that truck? Are any of those people who are attracted to those truck ads going to be interested in a truck where Skippy, the robot, is driving and in the ad sitting next to him is somebody who's doing their nails while Skippy drives over the construction site? The people who buy trucks want to drive them themselves. The people who buy sports cars want to drive them themselves. I'm not saying there's zero market for driving cars that are driverless, but it's not an infinite market. The big question is simply because things are possible to do with robotics and artificial intelligence doesn't mean that there's a market for it, nor does it mean that it's going to be cheap enough to do, especially when labor is still cheap. One of the claims that we hear the most is that AI is different. This time, it's different. The problem with that, going back to our colleague, Phil Tetlock, and his views about what makes for a good prediction is whenever you get a claim that something is unique, so we're generous, this is completely unlike what we have ever seen before, you want to question that. One of the things to question, as we said before, is that what artificial intelligence is likely to do. What data science can do right now is focus on one task. Look, for example, at radiologists, so well known studies showing that if you compare radiologists screening for one particular illness like breast cancer, that an algorithm does that better than radiologists do it. Not a hugely better, but they do it better. Now, is that going to eliminate radiologists? Well, the second part of the finding, which you don't hear as much about, is that radiologists working with algorithms are even better than either of them working separately. The third thing to remember is that radiologists do a whole bunch of stuff other than reading X-rays. They make diagnoses, of course, as well, only some part of which is based on what the screens tell them. Some part of what they do as well is they do treatments with radiology as well. It might very well be the case that some of their tasks get automated, but that doesn't mean that everything they do gets automated. Driverless trucks, for example. Most of the truck drivers in the United States are delivery drivers. What they do is they drive to your house. Maybe a robotic driver can do that, but then they get out, they unload your packages, they put them on the steps, they get you to sign something. We don't have a robot that can do any of that stuff. The other thing to remember here is that management decisions matter enormously with IT broadly defined in the introduction of data science tools. Let's turn back the clock a little bit to the introduction of numerically controlled machines and the use of computers in engineering decisions. We had machinists and machines. Some of that work could be taken over by CNC machines. They have to be programmed. What we saw 20 years ago when they were first being introduced is that employers made decisions. Are we going to hire engineers to program these machines or are we going to teach the machinists who already do this kind of work how to program the machines? if you teach the machinists how to program machines, you don't lose nearly as many machinist jobs. If you hire engineers to do the programming, then you're losing all the machinist jobs and you're hiring in and creating more machinist jobs. That outcome is a management decision. So what are the things we've known about the impact of IT on jobs and on productivity? The big impact comes when managers use the IT to restructure how work is organized. At that point, you get really big productivity increases. The IT by itself doesn't do it.