WEBVTT - How Artificial Intelligence is Taking Over the Economy 0:00:00.240 --> 0:00:05.159 From self driving cars to robot powered factories. Artificial intelligence 0:00:05.240 --> 0:00:09.360 is taking over significant pieces of the global economy. This 0:00:09.440 --> 0:00:12.880 is great for the businesses embracing AI, but there is 0:00:12.880 --> 0:00:16.599 a downside. More robots in the workforce also means more 0:00:16.720 --> 0:00:20.560 people losing their jobs to computers. So how bad will 0:00:20.600 --> 0:00:23.919 the robot revolution be and how will it reshape the 0:00:23.960 --> 0:00:37.200 global economy? Welcome to Benchmark. I'm Scott Landman and economics 0:00:37.280 --> 0:00:40.840 editor with Bloomberg News in Washington. Returning as a guest 0:00:40.880 --> 0:00:44.199 co host is my colleague Chris Content. He's a reporter 0:00:44.320 --> 0:00:47.239 covering the Federal Reserve and u S economy also here 0:00:47.280 --> 0:00:49.960 in d C. Chris, glad to have you back, Happy 0:00:50.000 --> 0:00:52.479 to be here. So this week we're talking about the 0:00:52.600 --> 0:00:55.840 rise of the robots and how they will impact the economy. 0:00:56.280 --> 0:00:59.360 As someone who writes about the Federal Reserve, Chris, what 0:00:59.600 --> 0:01:02.680 interest you and AI? Well, I can tell you, Scott, 0:01:02.760 --> 0:01:06.640 that there is a great capacity actually of artificial intelligence 0:01:06.680 --> 0:01:11.560 to transform one of the biggest jobs of central banks. 0:01:11.600 --> 0:01:15.360 As you know, central banks are trying to regulate economies, 0:01:15.480 --> 0:01:18.160 keep them in that Goldilock zone, and they do this 0:01:18.360 --> 0:01:22.440 through short term interest rates. But that tool has a 0:01:22.560 --> 0:01:25.080 lag to it. It takes a good six st eighteen 0:01:25.200 --> 0:01:28.880 months before a move and interest rates has an impact 0:01:29.280 --> 0:01:32.920 in the real economy. So these central bankers they've got 0:01:32.959 --> 0:01:36.319 to forecast what what's the economy gonna be like a 0:01:36.440 --> 0:01:39.760 year from now, eighteen months from now. But they're not 0:01:40.080 --> 0:01:43.120 very good at forecasting the future when it comes to 0:01:43.200 --> 0:01:49.120 things like inflation, unemployment, GDP growth. So some central banks 0:01:49.160 --> 0:01:53.040 are starting to turn for a little help to artificial 0:01:53.080 --> 0:01:58.040 intelligence because artificial intelligence can be used quite effectively at 0:01:58.400 --> 0:02:03.280 spotting patterns in past events and then using that to 0:02:03.440 --> 0:02:05.480 predict the future. So it's an area it's going to 0:02:05.560 --> 0:02:07.920 take a while, but it's an area of great promise 0:02:08.280 --> 0:02:13.000 for economics. It's not just economics, but it's business and 0:02:13.560 --> 0:02:17.720 the broader economy as a whole. Forecasting prediction, it's it's 0:02:17.720 --> 0:02:20.880 a really big part of what AI is meant to do. 0:02:21.360 --> 0:02:24.080 And our guest is here to talk about that today. 0:02:24.440 --> 0:02:27.400 His name is Joshua Gains and he's a professor at 0:02:27.440 --> 0:02:31.680 the University of Toronto's Rotman School of Management. He's one 0:02:31.680 --> 0:02:35.120 of the authors of a brand new book called Prediction Machines, 0:02:35.320 --> 0:02:39.760 The Simple Economics of Artificial Intelligence, just published by Harvard 0:02:39.800 --> 0:02:43.440 Business Review Press. Joshua, thanks for joining us. Good to 0:02:43.440 --> 0:02:46.359 be here. Thanks. So we've already sort of alluded to this, 0:02:46.400 --> 0:02:50.680 but why is the book called Prediction Machines. Well, it's 0:02:50.880 --> 0:02:55.400 uh called that and not some more interesting title such as, um, 0:02:55.440 --> 0:02:59.120 you know you're wonderful, great new robot and how your 0:02:59.120 --> 0:03:02.280 life is going to get better, simply because the recent 0:03:02.320 --> 0:03:07.560 developments in artificial intelligence are not about completely replacing human 0:03:07.600 --> 0:03:11.560 intelligence per se, but actually all of being about one thing, 0:03:11.720 --> 0:03:14.960 and that is prediction. That is taking a whole lot 0:03:14.960 --> 0:03:18.160 of information that you do have and converting it into 0:03:18.240 --> 0:03:21.919 information you do not have. That's very, very different from 0:03:21.960 --> 0:03:24.840 something that does all of your choices for you and 0:03:24.880 --> 0:03:28.840 things like that. Uh, it really really just does prediction. 0:03:29.520 --> 0:03:33.400 And when we talk about this prediction function and getting 0:03:33.440 --> 0:03:36.320 all this information, can you bring us into the real 0:03:36.360 --> 0:03:39.920 world here and what kinds of industries or jobs would 0:03:39.960 --> 0:03:42.680 be most likely to or have the most room to 0:03:42.760 --> 0:03:47.880 benefit from this kind of additional knowledge. Well, prediction is 0:03:47.920 --> 0:03:51.320 something and from your business that you're in, you think 0:03:51.360 --> 0:03:55.440 of it mainly about forecasting, for instance, economic variables and 0:03:55.480 --> 0:03:58.560 things like that, and something of course I've worried about 0:03:58.600 --> 0:04:02.720 as well. But what's really interesting about these new developments 0:04:02.960 --> 0:04:07.000 is that they highlight problems that turn out to be 0:04:07.080 --> 0:04:11.600 prediction problems. For instance, the whole issue of having a 0:04:11.640 --> 0:04:14.520 digital image served up to you and knowing what's in 0:04:14.560 --> 0:04:17.960 it is a prediction problem. Invariably, when you get an 0:04:17.960 --> 0:04:20.839 image from the Internet, the label that's being attached to 0:04:20.880 --> 0:04:23.880 it is the label that human would attach to it. 0:04:24.279 --> 0:04:26.960 And so basically what Google are doing when you're searching 0:04:26.960 --> 0:04:31.080 for a picture is predicting which pictures correspond to that label. 0:04:31.600 --> 0:04:34.680 So that's a form of prediction. And it turns out 0:04:34.720 --> 0:04:40.760 that that and language translation and understanding machines, understanding human speech, 0:04:40.800 --> 0:04:46.880 and even self driving cars are all mainly a prediction problem, 0:04:46.960 --> 0:04:50.120 and so that is where this new artificial intelligence is 0:04:50.160 --> 0:04:54.800 being implemented. You wouldn't normally call that prediction problems. We 0:04:54.839 --> 0:04:58.200 normally think about forecasting the weather or forecasting an earthquake 0:04:58.279 --> 0:05:02.200 or something, but basically prediction is all around us. Joshua, 0:05:02.320 --> 0:05:04.719 I think it's clear that this can have all sorts 0:05:04.720 --> 0:05:08.000 of benefits in our lives and in the economy, but 0:05:08.040 --> 0:05:11.440 I can also have some downsides, and particularly how how 0:05:11.520 --> 0:05:16.000 these things are distributed throughout our society. Maybe a big 0:05:16.080 --> 0:05:19.160 question mark first about the benefits. Where where do you 0:05:19.200 --> 0:05:23.760 see that societally and also economically. So I think the 0:05:23.800 --> 0:05:28.240 benefits come from wherever you think about where would it 0:05:28.240 --> 0:05:32.279 be good to know things with greater certainty? And so 0:05:32.320 --> 0:05:36.120 any kind of decisions that you're doing under uncertainty are 0:05:36.120 --> 0:05:40.760 going to benefit from having better prediction so that you 0:05:40.800 --> 0:05:45.120 can imagine being applied. For instance, you're trying to manage inventory. 0:05:45.240 --> 0:05:48.200 If you've got better predictions regarding demand you're going to face, 0:05:48.240 --> 0:05:50.400 you're going to be able to manage that inventory better. 0:05:50.760 --> 0:05:53.839 You're going to be able to correctly adjust your behavior 0:05:53.880 --> 0:05:59.360 to prevent shortfalls or worse, to prevent surpluses that end 0:05:59.440 --> 0:06:02.839 up to over stocking your inventory. Those are the sorts 0:06:02.880 --> 0:06:07.119 of places where prediction machines are going to work quite well. Uh. 0:06:07.120 --> 0:06:11.640 And so basically anywhere where there's decision making being done 0:06:12.000 --> 0:06:15.839 and there's uncertainty, there is room for better prediction, and 0:06:15.880 --> 0:06:18.760 the machines might serve that up right, And that sounds, 0:06:18.800 --> 0:06:21.640 of course like it will make our companies more efficient, 0:06:21.680 --> 0:06:26.280 more productive, but also I think will disrupt how companies 0:06:26.320 --> 0:06:30.440 operate and may disrupt people's lives. Do you think this 0:06:30.520 --> 0:06:33.119 time is going to be any different. We've had lots 0:06:33.160 --> 0:06:37.719 of technology disruption in our economy over the decades, in fact, 0:06:37.720 --> 0:06:41.679 in centuries, um is the pace of change these days 0:06:41.720 --> 0:06:45.839 going to be more disruptive and more problematic to adjust to. 0:06:46.520 --> 0:06:49.200 I think there's a chance, as usual, whenever you've got 0:06:49.200 --> 0:06:54.640 a very large and radical innovation occurring and being adopted, 0:06:55.000 --> 0:06:58.679 there is the potential for disruptive change. But the hard 0:06:58.720 --> 0:07:01.400 part is trying to predict exactly where that will be. 0:07:02.520 --> 0:07:05.000 When we look back and we think about the rise 0:07:05.000 --> 0:07:08.400 of the automobile, it's no surprise that we also point 0:07:08.440 --> 0:07:13.160 to horses as being the disruptive workers in that disrupted 0:07:13.200 --> 0:07:17.040 workers in that equation. When it comes to things like 0:07:17.520 --> 0:07:21.480 better prediction, however, it's far subtler um. Some of the 0:07:21.480 --> 0:07:25.320 discussion that's going around is saying, well, this is the 0:07:25.360 --> 0:07:27.840 first Why is this time different? It's because it's the 0:07:27.880 --> 0:07:32.160 first time you really have machines taking over cognitive tasks. 0:07:32.680 --> 0:07:35.720 Well that's not true. We've had machines take over cognitive 0:07:35.760 --> 0:07:39.920 task with computers, and you know, we seem to most 0:07:39.920 --> 0:07:42.600 of us seem to be still be gainfully employed, and 0:07:42.640 --> 0:07:47.120 the adjustments that took place have been worked out uh, 0:07:47.440 --> 0:07:49.920 this time, you know, Okay, the computers are doing much 0:07:50.000 --> 0:07:53.480 much more. They're doing much more in terms of thought. Yes, 0:07:53.600 --> 0:07:56.240 But as we identify in the book, the one thing 0:07:56.240 --> 0:08:00.280 that computers can't do is set goals. You still have 0:08:00.440 --> 0:08:04.400 to have a human what we call human judgment, to 0:08:04.560 --> 0:08:07.560 set the goals the trade offs. No prediction is going 0:08:07.600 --> 0:08:09.600 to be perfect, so you have to work out what 0:08:10.320 --> 0:08:12.640 how you're going to stomach errors and things like that. 0:08:13.360 --> 0:08:17.440 Those are still roles for people to get into. It's 0:08:17.480 --> 0:08:21.720 only where better prediction was like the final thing towards 0:08:21.800 --> 0:08:25.880 getting full automation that you might see jobs actually replaced. 0:08:26.840 --> 0:08:31.400 Speaking of not getting to full automation, one interesting example 0:08:31.520 --> 0:08:34.240 you discussed a little bit in the book is that 0:08:34.400 --> 0:08:39.240 of doctors and how their jobs might change under advances 0:08:39.280 --> 0:08:43.400 in AI, in terms of they just won't have to 0:08:43.520 --> 0:08:46.600 do the diagnoses as much anymore. The computer is going 0:08:46.640 --> 0:08:48.200 to do it for them, and their jobs are going 0:08:48.240 --> 0:08:51.600 to substantially change. Can you talk about that a bit well, 0:08:52.040 --> 0:08:55.640 basically in a conference a couple of years ago. Jeff Hinton, 0:08:55.960 --> 0:09:00.000 who is one of the pioneers of the new development 0:09:00.000 --> 0:09:02.600 it's in AI. He now it works part time at Google. 0:09:03.240 --> 0:09:06.120 He basically said to the conference, well, I think we 0:09:06.120 --> 0:09:10.199 should stop training radiologists. Now, what he meant by that 0:09:10.440 --> 0:09:14.200 was his view of what a radiologist does is look 0:09:14.280 --> 0:09:18.160 at images and then decide, you know, is there a 0:09:18.240 --> 0:09:23.280 problem or not that requires further treatment. Well, obviously, prediction 0:09:23.320 --> 0:09:26.800 machines have the capability and have been demonstrated in some 0:09:26.840 --> 0:09:29.840 settings to be far superior to people looking at those 0:09:29.880 --> 0:09:33.960 pictures and identifying exactly what's going on with a far 0:09:34.040 --> 0:09:36.960 greater degree of accuracy. So if that's your view of 0:09:36.960 --> 0:09:40.680 what a radiologist does, well it sounds like curtains for them. 0:09:41.280 --> 0:09:44.240 But actually radiologists have been dealing with these sorts of 0:09:44.240 --> 0:09:47.640 issues for fifty years and they're quite aware of them. 0:09:47.640 --> 0:09:52.000 That has technology improves their jobs change. What happens in 0:09:52.080 --> 0:09:55.320 terms of radiologists is it's not some a simple choice. 0:09:56.280 --> 0:09:58.800 You get a prediction of something and you know exactly 0:09:58.840 --> 0:10:01.199 what to do. You're right. If that was the case, 0:10:01.240 --> 0:10:03.120 then anyone could just look it up in a manual 0:10:03.480 --> 0:10:06.920 and you wouldn't need a radiologist. But invariably there are 0:10:06.960 --> 0:10:11.520 other factors, other criteria, and in particular the personal dimension 0:10:11.720 --> 0:10:15.760 of the patient situations that will also impact on the 0:10:15.800 --> 0:10:19.560 what the treatment decision actually is. So prediction is an 0:10:19.600 --> 0:10:23.240 important input into that, but there are other factors going on, 0:10:23.800 --> 0:10:26.679 and we're a long way off being able to automate 0:10:26.760 --> 0:10:31.120 all of that. And thus far, the evidence is that 0:10:31.400 --> 0:10:34.319 you make the radiologists actually better at their jobs by 0:10:34.360 --> 0:10:37.000 having these prediction machines. They are able to act with 0:10:37.080 --> 0:10:41.120 more certainty and therefore able to come up with more 0:10:41.120 --> 0:10:45.200 confident recommendations. This allows them to save some time and 0:10:45.240 --> 0:10:48.400 save some other things, and develop other skills, and may 0:10:48.520 --> 0:10:53.079 change the allocation of tasks between radiologists and other medical 0:10:53.080 --> 0:10:57.320 practitioners quite a bit. So I don't necessarily see it 0:10:57.480 --> 0:11:00.800 is obvious that those jobs are going to be disrupted 0:11:00.800 --> 0:11:04.400 as quickly as some of the engineers do. What kind 0:11:04.400 --> 0:11:08.760 of jobs, Josh, would you say are the least vulnerable 0:11:08.960 --> 0:11:12.880 to being replaced or even disrupted by this kind of technology. 0:11:13.480 --> 0:11:16.679 Well that's a really interesting question. You know. We tend to, 0:11:17.120 --> 0:11:20.320 when confronted with this thing, point to all the important 0:11:20.360 --> 0:11:24.319 things that we do and how it can't be replaced 0:11:24.360 --> 0:11:27.959 by a machine. For instance, some of the jobs that 0:11:28.080 --> 0:11:31.280 people talk about as being hard to be replaced by 0:11:31.360 --> 0:11:36.640 machines are ones that require emotional input. Interestingly, Danny Khneman 0:11:36.840 --> 0:11:41.600 at a conference here just last year. He's the Nobel 0:11:41.679 --> 0:11:44.920 Prize winning economist who has been responsible for behavior or 0:11:44.960 --> 0:11:49.200 economics and thinking about decision making and judgment in the 0:11:49.240 --> 0:11:54.120 context of psychology, and his answer was equivocal. He sees 0:11:54.200 --> 0:11:57.720 humans as ultimately very flawed and doesn't see any reason 0:11:57.800 --> 0:12:01.360 why machines wouldn't take over. His view was, do you 0:12:01.440 --> 0:12:05.560 really want your care to be managed by disinterested children 0:12:06.040 --> 0:12:09.080 when you've become elderly, or would you rather have a 0:12:09.160 --> 0:12:12.319 robot who's been trained to work out exactly what your 0:12:12.360 --> 0:12:15.160 needs are. So it's very hard to work out what 0:12:15.360 --> 0:12:18.120 is safe and what isn't. All we can say right 0:12:18.160 --> 0:12:21.640 now is that the tools in their current instantiation and 0:12:21.720 --> 0:12:23.840 what we're going to see probably over the next five 0:12:23.920 --> 0:12:27.240 years and maybe ten years, is all about that prediction function, 0:12:27.520 --> 0:12:30.680 which leaves a lot of room open for people. Speaking 0:12:30.679 --> 0:12:33.880 of jobs that are likely safe from AI disruption, I 0:12:33.880 --> 0:12:40.079 think Chris will appreciate being in Washington, d C. Federal government, Congress, etcetera. 0:12:40.160 --> 0:12:43.120 Those will probably be pretty safe from disruption for a 0:12:43.160 --> 0:12:46.160 long time. So if we ever get some disruption that sector, 0:12:46.280 --> 0:12:49.560 that would be something definitely to watch. But josh I 0:12:49.600 --> 0:12:53.040 wanted to turn to another issue that you discuss when 0:12:53.120 --> 0:12:56.520 when you're talking about economic impact in the book. Uh, 0:12:56.559 --> 0:13:00.760 it's really intriguing that you talk about inequality and how 0:13:00.880 --> 0:13:05.960 that would evolve if AI were to take hold in 0:13:06.040 --> 0:13:10.280 the economy. Why could that get worse with the rise 0:13:10.400 --> 0:13:14.160 of AI. Well, it's you know, as usual, it's difficult 0:13:14.160 --> 0:13:17.800 to forecast these things. You never should trust economis on 0:13:18.160 --> 0:13:21.280 on those sorts of big trend forecasts. But the concern 0:13:21.360 --> 0:13:26.720 that we have is that certain skills become more valuable 0:13:26.840 --> 0:13:30.560 valuable and other skills become less so. And one of 0:13:30.600 --> 0:13:33.640 the problems is if you wanted to be the sort 0:13:33.679 --> 0:13:37.559 of person who could take advantage of AI, it's invariably 0:13:37.880 --> 0:13:42.079 not going to be a skill that is like routine. Necessarily, 0:13:42.720 --> 0:13:44.600 once you know, once you have the sort of job 0:13:44.600 --> 0:13:46.640 where you can look at a prediction and you know 0:13:46.679 --> 0:13:50.280 exactly what to do, your place in that is devalued. 0:13:50.880 --> 0:13:53.640 On the other hand, there are some situations where it's 0:13:53.679 --> 0:13:55.440 going to take a bit of art to understand what 0:13:55.520 --> 0:13:59.000 the prediction really is. We don't tend to think of 0:13:59.360 --> 0:14:02.800 machine learn in computer sciences art, but there are so 0:14:02.840 --> 0:14:06.480 many variables that the engineers have to control and think 0:14:06.480 --> 0:14:09.719 about that. It does tend to have that quality to it, 0:14:10.120 --> 0:14:13.840 and similarly, how we use those tools also has a 0:14:13.880 --> 0:14:16.720 bit of an artistic flare to it. And by that 0:14:16.840 --> 0:14:20.440 I mean that it's not a dents sure why someone 0:14:20.480 --> 0:14:24.520 who's more productive is more productive, but somehow they get 0:14:24.560 --> 0:14:27.240 it more and so the people who are able to 0:14:27.280 --> 0:14:30.120 do that are probably going to fare better. The one 0:14:30.160 --> 0:14:33.280 concern we have is when I'm a person who's using 0:14:33.280 --> 0:14:35.920 an AI tool, I can use it at a much 0:14:36.080 --> 0:14:41.480 larger scale and make decisions for many more across a 0:14:41.560 --> 0:14:45.040 greater domain. And in that regard, I'm sort of made 0:14:45.040 --> 0:14:48.760 a superhuman. But you can only have so many superhumans, 0:14:48.800 --> 0:14:52.520 and that's where we might worry about inequality. Josh, talk 0:14:52.600 --> 0:14:55.080 to us a little bit also about this idea explored 0:14:55.120 --> 0:14:59.680 in your book on how artificial intelligence might actually contribute 0:14:59.680 --> 0:15:04.320 to the concentration of certain industries in our economy, even 0:15:04.360 --> 0:15:09.240 aid the creation of monopolies. Well, I think this is 0:15:09.280 --> 0:15:13.200 something that has occurred with any digital platforms, especially ones 0:15:13.280 --> 0:15:17.160 that work better with larger scale. So the current AI 0:15:17.360 --> 0:15:20.320 runs on machine learning, and I have to emphasize the 0:15:20.400 --> 0:15:25.680 learning part. You get better by putting your machines out 0:15:25.680 --> 0:15:30.840 in the field and continuously adjusting them with new data 0:15:30.920 --> 0:15:34.080 and new learning. In so, what that means is that 0:15:34.120 --> 0:15:37.040 a company such as Google that has a lot more 0:15:37.560 --> 0:15:40.520 people using its search engine is able to use AI 0:15:40.640 --> 0:15:43.920 to improve that those search engines at a greater rate 0:15:43.960 --> 0:15:47.720 than competitors such as Being and Duct Duct Go, and 0:15:47.960 --> 0:15:52.240 so in that regard, you could have the preservation of 0:15:52.400 --> 0:15:55.840 a dominance or the emergence of dominance. And similarly this 0:15:55.920 --> 0:15:59.040 might hold for companies like Facebook, and it might hold 0:15:59.080 --> 0:16:02.760 to a degree for companies like Apple and Amazon as well. 0:16:03.120 --> 0:16:05.680 They just have a lot more activity and so their 0:16:05.720 --> 0:16:09.480 potential for to to use AI to learn at a 0:16:09.560 --> 0:16:13.480 faster rate might be there. But as with all of 0:16:13.520 --> 0:16:16.600 these things, is sometimes these firms do get into a rut, 0:16:16.680 --> 0:16:20.680 and sometimes people find better ways of learning and doing 0:16:20.720 --> 0:16:24.320 things that might initially perform worse but have a better trajectory. 0:16:24.880 --> 0:16:26.600 So it's not a given that we're going to have 0:16:27.120 --> 0:16:30.760 the current monopolies, be the future monopolies or anything like that, 0:16:31.040 --> 0:16:34.000 or the current large companies, but we might see new 0:16:34.040 --> 0:16:37.160 ones develop on the basis of new tools. Every other time, 0:16:37.200 --> 0:16:40.800 we've had a large technological revolution that has occurred, I'd 0:16:40.800 --> 0:16:44.240 expect it to occur this time too. Just taking a 0:16:44.400 --> 0:16:50.040 broader view of competition, you also get into which country 0:16:50.160 --> 0:16:52.800 might be the dominant force in AI. If there's a 0:16:52.840 --> 0:16:57.080 dominant force, the US has already has a clear lead, 0:16:57.480 --> 0:17:01.920 and yet there's a lot of activity going on in China. 0:17:02.000 --> 0:17:06.000 How do you see China ascending and competing against the 0:17:06.080 --> 0:17:10.200 United States in AI. Well, this is where being able 0:17:10.240 --> 0:17:12.680 to have access to the right sort of data comes 0:17:12.760 --> 0:17:15.880 to play, not just data but also talent. So let's 0:17:15.880 --> 0:17:18.800 talk about data first. The United States is sort of 0:17:18.800 --> 0:17:22.399 a middle ground in terms of privacy regulation. Europe is 0:17:22.560 --> 0:17:26.040 far more stringent, but China there's none at all, and 0:17:26.119 --> 0:17:29.920 so Chinese firms have the ability to appropriate and use 0:17:30.359 --> 0:17:33.920 consumer data to develop AI at a much greater rate 0:17:34.240 --> 0:17:37.160 than you would be able to do inside the United States. 0:17:37.400 --> 0:17:41.280 So there's a benefit. Secondly, there's the issue of capabilities. 0:17:41.800 --> 0:17:44.600 At the moment, AI resources are thin on the ground. 0:17:44.720 --> 0:17:47.959 It's hard to hire engineers they command six or seven 0:17:48.000 --> 0:17:51.080 figure sums. But the United States at the moment is 0:17:51.080 --> 0:17:54.639 cutting itself off from the global pool of talent in 0:17:55.600 --> 0:17:58.640 machine learning, and this is something other countries aren't doing. 0:17:58.680 --> 0:18:01.639 Not only they not cutting them selves off, They're also 0:18:01.760 --> 0:18:05.480 providing resources. China is spending several billion dollars on a 0:18:05.520 --> 0:18:09.280 technology cluster in this area. Russia are doing the same, 0:18:10.080 --> 0:18:14.199 and even countries like Canada, the government's actively supporting the 0:18:14.240 --> 0:18:18.320 development of news superclusters in the space. That's going to 0:18:18.400 --> 0:18:22.439 help attract talent around the world, because talent wants to 0:18:22.440 --> 0:18:25.560 be able to work, and I think that's another area 0:18:25.560 --> 0:18:29.040 where the United States faces some risks. Let's end this 0:18:29.119 --> 0:18:33.080 interview with the existential question that you address in your 0:18:33.080 --> 0:18:35.520 book or try to address. We've all seen our share 0:18:35.560 --> 0:18:39.000 of science fiction movies. The Rise of the Robots Terminator 0:18:39.080 --> 0:18:42.879 to looms very large in my mind for me, josh 0:18:43.200 --> 0:18:45.640 is this the end of the world as we know it? Well, 0:18:45.680 --> 0:18:49.000 as we say in the book, not enough time yet 0:18:49.040 --> 0:18:51.280 to tell. But you've got enough time to read our 0:18:51.320 --> 0:18:54.199 book and be on the right side of it. All right, Well, 0:18:54.240 --> 0:18:57.240 let's send it there. Joshua Against from the University of Toronto, 0:18:57.359 --> 0:19:00.280 author of Prediction Machines, thank you very much for running 0:19:00.320 --> 0:19:06.800 us on Benchmark. Thank you. Benchmark will be back next week. 0:19:06.920 --> 0:19:09.240 Until then, you can find us on the Bloomberg terminal 0:19:09.240 --> 0:19:13.360 Bloomberg dot com. Our Bloomberg app and podcast destinations such 0:19:13.400 --> 0:19:17.640 as Apple Podcast, Spotify, or wherever you listen. We'd love 0:19:17.680 --> 0:19:19.679 it if you took the time to post a review 0:19:19.720 --> 0:19:22.439 of the show so more listeners can find us. You 0:19:22.480 --> 0:19:25.200 can also check us out on Twitter, follow me at 0:19:25.280 --> 0:19:30.000 at scott Landman, Chris You're at Chris Ja Conden, and 0:19:30.160 --> 0:19:33.240 our guest Josh Gannes is at Josh gannes g. A 0:19:33.520 --> 0:19:37.080 n S Benchmark is produced by Toper Foreheaz. The head 0:19:37.119 --> 0:19:40.720 of Bloomberg Podcasts is Francesca Levy. Thanks for listening. To 0:19:40.800 --> 0:19:41.600 see you next time.