Artificial Intelligence and the Future of Humans and Robots in the Economy

Friday, February 27, 2015


Event Description

MIT professor emeritus and Rethink Robotics’ founder Rodney Brooks, Carnegie Mellon’s Abhinav Gupta, and MIT’s Andrew McAfee, join Nicholas Thompson, editor at NewYorker.com, to discuss artificial intelligence (AI) and robot technology, and their economic impact on industry and society in the future. The panel remarks on the accelerated rate of AI advancement in the past couple of years, adding that AI is still limited to solving specific tasks as opposed to having common sense or dexterity. The panel comments on education, research funding, and military and security AI applications such as predictive analytics.

The Malcolm and Carolyn Wiener Annual Lecture on Science and Technology addresses issues at the intersection of science, technology, and foreign policy.

Event Highlights

Rodney Brooks on robot technology:
“…robots are good at very simple things like cleaning the floor, like doing a repetitive task. Our robots have a little tiny bit of common sense…that's a bit of common sense that, traditionally, industrial robots have not had. But we just put each of our piece of common sense in one by one, rather than automatically. I think people get confused by seeing Hollywood movies and see what the robots can do there, but we are a long, long way from having robots that can do that.”

Abhinav Gupta on research funding in the United States:
“The research funding is, in U.S.,  going down. …from 2008 to 2012, the research funding in China, which is going to fuel all this technology drive, is doubled. You can see the research outputs. They are increasing. They are doubling. We are going down. Europe is even better than U.S. I would think given the European economy, they would be worse. But if you talk to researchers, like me, Europe has much better…research funding than America has.”

Andrew McAfee on policy challenges affecting the U.S. economy:
“Why is our infrastructure so lousy? Why are immigration policies, as far as I can tell, designed by our enemies? Why are we still turning out the kinds of workers we needed fifty years ago? If we could get these things right, that would be the best thing we could do to improve the prospects for the American worker, and none of it is rocket science.”

THOMPSON: I would like to welcome you to today's Council on Foreign Relations Malcolm and Carolyn Wiener Annual Lecture on Science and Technology. The title today is "Artificial Intelligence and the Rise of the Robots" with Rodney Brooks, Abhinav Gupta and Andrew McAfee.

Rodney Brooks is a Panasonic professor of robotics, Computer Science and Artificial Intelligence Lab at MIT. But he is known in my family, I have three little boys, and he is known as the inventor of the Roomba. And in fact, this morning, I was having a hard time getting my kids dressed. Roomba is a little electronic vacuum cleaner that my toddler surfs on and the older kids love. I told my kids I have to get to work on time since I was interviewing the man who invented the Roomba and they got dressed like that. Rodney Brooks, a very important man.

Abhinav Gupta is an assistant research professor at the Robotics Institute at Carnegie Mellon University. He is doing some of the most interesting and cutting-edge research we have on the way artificial intelligence can work with visual extraction technologies.

Andrew McAfee is the principle research scientist and cofounder of the initiative on digital economy at the Sloan School at MIT. You have probably all seen his book, The Second Machine Age. If you haven't seen it, you've seen reviews of it. I work at the New Yorker and we actually used a draft of a review of his book as an edit test. So I have read 100 copies of a review of your book. That was so bad, we made it into an edit test. Sorry about that.

OK. A special thanks to Malcolm and Carolyn Wiener for making this lectureship possible. They are here in the audience. We just had a lovely lunch talking with them. We were very grateful. This is an extremely important topic.

And I would also like to welcome the CFR members around the nation and the world participating in this meeting through the live stream.

We are going to begin with a video to help us understand what exactly is going on. So if someone could please queue up Baxter. This is one of the many robots that Rodney Brooks has brought into the world. So, Baxter.

(BEGIN VIDEOTAPE)

(UNKNOWN): Think adding automation to your production line limits you to high costs, high volume, caged robots? Think automation requires specialized robots programmed by highly-trained engineers? Well, think again.

Baxter breaks out from the productive cages that contain many manufacturing robots, allowing people and robots to work side by side, doing what each does best. From machine tending to line loading, to packing and unpacking, Baxter has the flexibility to be deployed and redeployed on tasks with minimal setup or integration costs. Whether there is increased customer demand or the introduction of a new product, Baxter helps you create a collaborative workforce, one with increased productivity and efficiency.

Think about getting all this for a base price of $25,000. That translates into an ROI of months, not years.

Now think about what Baxter could add to your plant. Rethink manufacturing. Rethink what's next. Rethink robotics.

(END VIDEOTAPE)

BROOKS: In my defense, I did not choose that video.

THOMPSON: That is an excellent video. I hope everybody's excited.

BROOKS: Well, it's a little selling and I wouldn't have chosen the selling one.

THOMPSON: It's like a Super Bowl ad. After watching that, I replaced the entire staff of the New Yorker website with Baxters and it's been going quite well for the last two weeks. I hope you all visit newyorker.com.

Abhinav, would you like to—we're going to begin with the hardest stuff first. So Abhinav, I want you to give some background on what NEIL is so we can understand it, and then we will jump right into the big topics everybody wants to hear about, how the economy is changing, how the world is changing, how the military is changing and how long everyone in this room will still have their jobs until they're replaced with robots.

BROOKS: Forever, forever.

THOMPSON: Not everybody if I've read all your work.

Abhinav, do you want to start with NEIL?

GUPTA: Before I start, I think I should say that I also have a Baxter in my lab and I'm a big fan of Baxter. It's changing at least research completely with the price point which it has.

THOMPSON: Andy, do you have a Baxter?

MCAFEE: I do not yet.

THOMPSON: Do you have a Roomba?

GUPTA: You should buy one.

MCAFEE: Well, Rod and I are neighbors. I'm just going to go over and see his in Cambridge.

THOMPSON: All right. Abhinav?

GUPTA: OK. So before I give an introduction to NEIL, let me just first talk about why we need NEIL. And so if you think, there are things we humans—like facts we humans just seem to know about. For example, all of us here know sheeps are white, cars have wheels, babies have eyes, and so on. So these kind of facts, which I like to call common sense knowledge, it's not clear how we learned them. But what is clear to me is that we use common sense every day and every moment of our lives. For example, if you see a boxy object outside on the road with a wheel, you will all know that this is a car. So you are using your common sense to reason about objects. You use common sense to reason about physics. You fill your dishwasher so that by putting on one more dish, one more dish, and you know your common sense tells you that it's starting to fall. You use common sense to predict what's going to happen next. So if you throw a ball up, you know it's going to come down.

Now this common sense, which is so obvious to you and me, it's completely tricky and totally inaccessible to computers right now. Why? Just think of the skill. How many such facts can I write? Billions, trillions. I mean, I don't even—I cannot even count how many such facts will exist for us to know what is truth in this world, what—how does our world work.

So at CMU, at Carnegie Mellon, we have been taking an approach where we are building NEIL. NEIL is Never Ending Image Learner. It's a computer program that runs twenty-four hours a day, seven days a week. It's running right now as we are going to have this conversation. And what it is trying to do is something very simple. All it does is it downloads hundreds of boring images of our world every day. It tries to understand those images using the advances in AI. And as it understands more and more images, it's basically experiencing the world. And as it is experiencing, it is learning more and more common sense facts about the world. And by learning more common sense facts, it kind of becomes much more smarter and smarter in understanding the world around us.

So for example, NEIL knows our concepts and relationships. Like it knows that Columbia can have four different meanings and it learns all this automatically without having a single human in the loop. So it knows Columbia could be Columbia University, which is like nearby. It could be Columbia Sportswear Company, which is making jackets, shoes and so on. Columbia could be Columbia River, or it could be Columbia Pictures.

And then, NEIL has also been learning relationships.

MCAFEE: It's not aware there's a country?

GUPTA: It has not learned it's a country. So yeah, there is some—there are some gaps, I have to say.

BROOKS: It's only three years old.

MCAFEE: Sorry.

BROOKS: It's only three years old.

GUPTA: It's actually one year old.

BROOKS: Oh, one year old.

GUPTA: So NEIL is just one year old. Country is probably harder than you would say.

So just to point out that the kind of common sense relationships NEIL has learned, for example, trading floors are crowded. It has learned things like Corolla is a kind of a car, cars have wheels. It has learned things like bamboo forests have vertical lines, and so on.

So it has been running for one year, as Rodney pointed out. So it's still a baby. I should ask any other baby how much he or she knows to compare. It has learned more than four million objects in the world, at this point of time, 20,000 relationships. And as it grows older and older, my hope—it might not happen, as Andrew points out that there are some holes—my hope is that it will be able to understand better and better about our world, using images and videos.

THOMPSON: It is a terrifying thought of someone just surfing the web and learning everything about the web. They must know a lot about llamas today, too. I hope NEIL does.

MCAFEE: They still don't know what color that damn dress is. No insight on that.

THOMPSON: Nobody does. This is the biggest mystery of the --

(CROSSTALK)

GUPTA: Computers agree on the color of the dress.

THOMPSON: Baxter knows.

OK. Rodney, let's talk about where robots are. You've been a leader in this field for at least thirty-five years. You wrote your first seminal controversial paper trying to answer the question that I'm about to ask thirty-five years ago, I believe. But—what are they good at; what are they not good at?

BROOKS: Robots are good at very simple tasks. They're not good at complex tasks. They're certainly not good—here's something that if a venture capitalist offered me money to build a company around, I would not take it, even though I love taking money from venture capitalists but I'm too honest. I would not say, you know, give me $25 million and I'll work on a robot that can do this.

[Removes keys from pocket.]

Way, way too hard. Way to dexterous. Way too hard. We have no idea how to do that level of dexterity. In fact, in what you saw in the video was Baxter with either just two fingers going like this, without any touch sensing, or a suction cup sucking. Dexterity, we're terrible at. We haven't done at all well in the last fifty years.

THOMPSON: Why?

BROOKS: Because hands are very complex and you need to work on four simultaneous pieces of science at once. You need to work on different kinematics. You need to work on materials because our hands are very dependent on materials. You need to work on sensing and you need to work on algorithms all at once and we just have not succeeded.

THOMPSON: But aren't prosthetic hands a great medical success of the last few years? Or is that a --

BROOKS: But there's a person in the control loop.

THOMPSON: Right. So it's the control loop for the dexterity. It's not the physical dexterity.

BROOKS: You need both in order to succeed.

So robots are good at very simple things like cleaning the floor, like doing a repetitive task. Our robots have a little tiny bit of common sense. Our robots know that if they've got something in their hand and they drop it, it's gone. They shouldn't go and try and put it down. So they—that's a bit of common sense that, traditionally, industrial robots have not had. But we just put each of our piece of common sense in one by one, rather than automatically.

So I think people get confused by seeing Hollywood movies and see what the robots can do there, but we are a long, long way from having robots that can do that.

THOMPSON: Are battery technologies a main limitation right now as well?

BROOKS: You know, battery—battery technologies are certainly a—for mobile robots, they're certainly an issue. But we don't have the technology to get the energy density conversion that human muscle has, even if we had the batteries.

THOMPSON: OK.

Well, Andy, one of your great themes is how these different kind of robots are advancing and what jobs they will eliminate. So let me just read off some of the professional affiliations of the people in this room; JPMorgan Chase, Centerra Capital, Markle Foundation, Passur Aerospace, New York Presbyterian Hospital, The Dana Foundation. Who is most likely to lose their job first to a robot?

MCAFEE: Probably someone like me, who tries to survey a big mass of unstructured information, draw a hopefully correct conclusion or an inference based on that large amount of information, and then try to communicate that insight to a broad audience. Digital technologies are now pretty good at all of those tasks, and getting better along a Moore’s Law kind of trajectory, an exponential trajectory, as opposed to a linear trajectory.

And the reason that we wrote the book and what's fundamentally interesting to me is, really smart guys, previous generations of—and above Rod, have been working on a set of problems for as long as we've had computers. Call it half a century. The progress they made was deeply underwhelming from any practical point of view.

And then just in the past few years, digital technologies and all their manifestations, including robots, have been breaking out of the little box where they'd been operating and starting to demonstrate capabilities that they never, ever had before that are really impressive when you look at it. They're economically significant and we're seeing human-level and, in many cases, superhuman performance in some fundamentally important areas, like unstructured search, like natural language processing, like image recognition and image processing and image classification. A lot of these things that technologies were, traditionally lousy at, they're now really, really good and getting a lot better all the time. This was pretty unprecedented and unexpected and I think it's going to—it's already having pretty serious economic consequences. I think they're going to get bigger.

BROOKS: Can I respond to that a little bit?

MCAFEE: No.

THOMPSON: They've started fighting already.

BROOKS: People always want us to fight, but we don't really.

I think, although I agree with the general themes that Andy talks about, I think it's very easy for people who are not deep in the technology itself to make generalizations, which may be a little dangerous. And we've certainly seen that recently with Elon Musk, Bill Gates, Stephen Hawking, all saying AI is just taking off and it's going to take over the world very quickly. And the thing that they share is none of them work in this technological field.

So let me explain why—and they're all smart people, but I think they're making a fundamental error and it gets to NEIL, actually.

THOMPSON: After taking down Bill Gates, Elon Musk and Stephen Hawking, he's going to take down the Dalai Lama.

Please continue, Rodney.

BROOKS: So let's go back to an example from the '90s, when IBM's Deep Blue beat Kasparov, beat the world chess champion. And Kasparov got up and said, well, at least it didn't enjoy beating me. That was his—holding on to his humanity.

And now, today, you can get programs that run on—and that was on a supercomputer and now you can get programs that run on laptops. There's about twelve of them that have a better chess rating than any human being has ever had. So people see that --

MCAFEE: It's so bad now—let me underscore what Rod is saying. It's so bad now that they asked human grand master a couple years ago how he would prepare for a match against a computer and he said, I'd bring a hammer.

BROOKS: So they can play chess really, really well. And I think people generalize that in the way that if a person can do some task really, really well, they can do adjacent tasks quite well. But none of those chess programs can play tic-tac-toe. Imagine a chess grand master who couldn't play tic-tac-toe. It doesn't make sense. None of those chess programs can give advice to an aspiring human on how to play better. All they can be is a sparring partner.

MCAFEE: That program couldn't play tic-tac-toe without being substantially redirected, right?

BROOKS: Right. So people, I think, are seeing some of the image labeling that's going on, for instance. Google came out with image labeling, which is a great commercial problem for them. They want to be able to label images. And one of the examples was, that Jeff Hinton shows, one of the chief scientists, is, it's a picture and it says there's a baby holding a teddy bear or doll in there. You look at it, it's a baby holding a teddy—a doll.

But then if you ask the program, where is the baby? All it can say is, well, this pixel has 10 percent probability of being a baby, this pixel has 80 percent. And people have done experiments. You have a mashup of, you know, a grotesque mashup of baby parts and it says it's a baby. It's a baby. It's got all the parts. But a person says, no, that's a grotesque mashup of baby parts in the image.

THOMPSON: But Abhinav, you've solved this, right?

GUPTA: No, no, no.

BROOKS: Well, he's working towards--

GUPTA: So, can I--

BROOKS: He's working towards it because it's such a hard problem.

GUPTA: Yeah, OK. Thank you.

So since we are talking about images, I think I should chime in a little bit and tell you that—so what Andrew is talking about, that we have made big advances, again, they're very, very specific tasks. Given an image, tell me what label can you put on that image? We have gotten really good at this task. Some people claim even better than humans. I don't buy that, but let's assume that even better than humans. But that doesn't mean we can do anything else apart from that exact task. And that's what Rodney was talking about. They have no idea that—what does baby mean? What does having a baby mean in those images? No idea. You, as a human, would know, OK, if I'm saying there's a baby, it has a lot of meaning inside. You get a lot of meaning out of that thing.

So while we have made significant advances in the last two years, I want to boil it down a little bit and say we still are a long, long way to go but Elon Musk or Bill Gates, everyone is talking about, we still have a long way to go. But there's hope, and that's what I think we have to see here. Two years ago, if you asked me can computers take an image and solve this problem, I would say I have invested seven years of my life but if you give me a random image, it will not work. And now, given a random image, it will work. So all it --

MCAFEE: This was the guy who was doing this for a living and if you asked him two years ago would this happen, he would say, no. This progress is weirdly fast and is surprising insiders in the field.

GUPTA: Yeah. I agree. And I'm an insider in the field and I'm very surprised, I have to say. Now, at this point of time, I am like living in an awe of myself, in some sense, that—but it was like Rodney told—Rodney told us that in thirty-five years he never thought he'd do this all. I also thought some of it, like that, for thirty-five years, I will not solve—see this kind of classification performance. But as I'm saying this, still a long, long way to go and much harder way to go. What—where all the kind of gains have come from is the data. And I think—so technology, this like deep-learning technology has been there from '70s and '80s. Don't misunderstand that this technology came two years ago and everything's changed. This has been there for thirty years. It's just that for the first time in our technological advances we have data for this deep-learning technology to learn.

MCAFEE: Let me jump on this because I think the three of us are really agreeing, instead of disagreeing. I chose my adjective pretty carefully. I said these advances are going to be economically significant. I completely agree with Rod that they're not going to be existentially significant on any timeframe that we really need to worry about, for exactly the reasons that you're bringing up.

One way to think about this, the way I try to get my mind around it, is there are, from what I've been able to take in, there are something like between ten and twenty really fundamental challenges that these guys and their discipline have been working on. Common sense is a really great example of that.

As I've looked around, these breakthroughs that we're seeing seem to be—kind of indicate that we're making real progress on one of those challenges, the challenges of learning in a pretty unstructured environment. That's a big deal. There are lots of other fundamental challenges in the discipline where the progress has not been as fast, and these are the ones that you're working on.

GUPTA: So let me add another thing. I mean, it's economically significant but I see the positive, unlike him seeing the negative. I see more jobs created from digital analytics than jobs being lost here because of digital analytics.

MCAFEE: I wish the evidence were lining up with your optimism.

GUPTA: I feel at least, I mean, so what will happen --

MCAFEE: What you feel is not really a great counter argument to the balance of the evidence.

GUPTA: I agree. I agree. I'm not an expert in economics; that's what you are. But what I feel that over time what we have seen is the characterization of the jobs will change. I agree that the characterization—so earlier, we were like if—suppose Rodney's Baxter goes into every manufacturing company. Maybe we will not have people doing manufacturing in those companies, but we'll have some other jobs which would come up. And there are two great examples which I would like to point out. For example, this digital analytics is creating so many jobs in health care right now. My wife was basically employed because of this—sorry, I --

MCAFEE: How many advanced degrees does your wife have?

GUPTA: Well, I am—that's what I'm saying. The characterization of jobs have changed. And my wife does have advanced degree. I'm not saying she doesn't have. But what wrong in humans having advanced degree? I don't see—I mean, characterization of jobs will change, our education might need to change, but that doesn't mean that are not going to survive. I mean, we are going to have enough jobs. Robots are going to be helping us, not destroying us.

MCAFEE: I agree with exactly half of that.

THOMPSON: Hold on. We've got to pause for a second here because we are dismantling Andy's theory without having presented Andy's theory. A classic example of panel indirection.

So Andy, will you please present your theory about the economic loss, and then we'll go back to dismantling it.

MCAFEE: And I want to be clear; I'm an optimist about technological progress. I get sometimes painted as a doctor doom, which I think is a little unfair. Tech progress is the only free lunch that economists believe in. It's making the overall pie bigger, unambiguously.

THOMPSON: CFR, I just had a delicious pie --

MCAFEE: The question is about how we're dividing up that pie. And the classic bargain for the era of industrial capitalism for most of us is you offer your labor, an employer wants that and you get a decent and better-over-time middle class living as a result. There is plenty of evidence that that bargain is under some stress these days. Median American family, even after you adjust for household size and inflation and everything else we can adjust for, is not making any more money than they were in the 1990s. The recovery is gaining some jobs. Most of those jobs are not what we would call great jobs. They're not highly paid. They don't have benefits. They're kind of lower to mid-low range, fairly precarious jobs. And we're still not generating enough of them. The labor force participation rates for the American workforce, even of prime-age workers, are where they were about thirty years ago, before women entered the workforce in big numbers.

So as I look around, I see plenty of evidence that something is changing with that deep bargain and technology and tech progress is one of my culprits for it.

THOMPSON: Let me read a tiny bit from an essay that John Lanchester just wrote in the London Review of Books that I just read last night. I came out in this week's issue. "In 1960, the most profitable company in the world's biggest economy was General Motors. In today's money, General Motors made $7.6 billion that year. It also employed 600,000 people. Today's most profitable company, Apple, employs 92,600. So where 600,000 workers would once generate $7.6 billion in profit, now 92,000 workers generates $90 billion. An improvement in profitability per worker of 76.65 times, but the jobs are disappearing." Which is echoing your argument.

Let's talk about—not about the United States, but let's talk about the global economic impact of this transformation that's going on.

Rodney, why don't we start with you? You were, at lunch, telling us a great story about Roombas in Spain, which shows the complexity of what's going on here.

BROOKS: So the Roomba home cleaning robot from a previous company of mine, iRobot—biggest—fastest growing market in Europe is in Spain. Spain has terrible unemployment. Why would people be buying Roombas to clean their houses? It's because of the collapse of the construction industry. The South American workers went back to South America and their wives went with them and their wives had been the maids that people had employed. And people, even with very high unemployment rates, don't want to be maids in Spain, and so people are buying Roombas to clean their houses.

So the—it's a complex intertwining of forces. I don't think it's a very straightforward one, which the press often takes your statements and generalizes them to be very straightforward; more technology, less jobs. It's a complex --

MCAFEE: And historically, that has not been true. For 200 years we have had a ton of tech progress and the average worker has done nothing but get—but get better off and more employable over time.

THOMPSON: So why is that different? Why now? Why is that --

(UNKNOWN): Why this time?

MCAFEE: Because as I said earlier, the technologies are breaking out of their historical boxes. They're doing things they could never, ever do before. And they're eating into the bundle of skills and abilities that we all have to offer. They're eating into it in a way that would we've not seen.

GUPTA: Can I add something? I mean, I just want to say that he was saying 'I don't have data", but his data is showing what I was saying, that with technology the jobs are increasing; it's not going down. You said that I didn't have the data.

MCAFEE: Were you tuning out a little earlier when I talked about some of the statistics?

GUPTA: No, but I'm just saying that—I mean, that the kind of jobs might change. And I agree, I mean, that it's kind of—you know that these jobs of labor and stuff will definitely—(inaudible)—like in the Spanish things and so on, I mean—but there is going to be new jobs.

I think the best example I want to put is my computer learning. At times, like ten years old, my father would never touch a computer. He hated computer learning and so on. But now computer is everywhere in the life. What happened? The technology of computers came in and now everyone starts to learn about computers and now the jobs are like that. I think the jobs are just going to be much more different than they are today, and that's good. I mean, I don't see anything wrong in that at all.

BROOKS: I want to say that I've become over—I've been subjected to Andy's books for the last few years. I've—I used to say, well, why now, I don't believe now, but I think it's a little more complex. I would say, by the way, that Apple, with their 92,000 employees, also has about 1.5 million people under subcontract in China, which their boss, Terry Gou, is desperately trying to replace with robots.

THOMPSON: So my statistic is wrong but it may be right.

BROOKS: Yes. Well, because it's —are much more complex. GM just built everything.

But you know, one of the—one of the things that I've certainly cited before and other people have cited is that, you know, in the agricultural revolution, we went from a majority of the population being in agriculture to a very small percentage being in agriculture but it didn't get rid of jobs. But what—and I was reminded of this as I was walking along 59th Street past the park to get here, what did happen was the population of horses plummeted. There were horses on farms doing work that were no longer needed. The only gainfully employed horses left are here in Central Park, and it's a sad, sad thing. And the question is whether we're going to be the horses this time.

MCAFEE: From an economic point of view, that's the right question. From a societal point of view, these are our choices to make. We get to decide whether we become irrelevant or not, and I certainly hope that we don't. But people who are blithe about job and wage prospects have one pretty important phenomena to explain away. It's the fact that in the majority of countries around the world, high-wage ones, certainly, but even low-wage countries as diverse as Mexico, India and China, the share of GDP every year, getting paid out in wages, is going down. It's going down fairly quickly. We've not seen this before. And we're not outsourcing to Mars. So the globalization is not part of the phenomenon there. But the classic structure in the bargain is shifting.

BROOKS: There's a much more complex thing. You talked earlier—and we haven't talked about this—you talked earlier about the contract between working for a company, getting a middle class job. Along with that, the company took on the risk of stabilizing your employment, stabilizing your income and other benefits. And what we're seeing now is a different sort of company coming along which doesn't have employees but has contractors. And it ranges from Airbnb to Uber, different sorts of models of interaction, but across—even UPS trucks, FedEx trucks, the whole way that workers and companies relate, and it's being greased by information technology. But not about AI. It's just about being able to track things. So --

MCAFEE: About really powerful platforms, sure.

THOMPSON: So let's say I'm Barack Obama and I have to come up with a national industrial policy, and I recognize these trends that you're talking about. I see the economy transforming. I recognize that robots are coming. Baxter will only have more Baxters and better Baxters in the future, even if they can't take the keys out of their pockets. What should I do, if I'm Barack Obama and I want to adjust our economy? Well, I'm not. I'm just a guy, but you know, what should America's policy be?

MCAFEE: Are you cold-calling or do you just want to us to shoot off our mouths if we feel the urge?

THOMPSON: I'm asking you to shoot off your mouths.

MCAFEE: So in the short term, I find this a pretty easy question to answer. In the short term, despite the hard work of guys like this, the robots and the AI are not about to take all of our jobs. They can't do most of our jobs yet. So the right recipe for now is grow the economy more quickly. The jobs will come along with it. I completely agree with Larry Summers on this point.

So then the question is what's the econ 101 playbook for right now?

THOMPSON: Right.

MCAFEE: Again, I don't find that that hard a question to answer. Why is our infrastructure so lousy? Why are immigration policies, as far as I can tell, designed by our enemies? Why are we still turning out the kinds of workers we needed fifty years ago? If we could get these things right, that would be the best thing we could do to improve the prospects for the American worker, and none of it is rocket science.

BROOKS: I want to bring up something about robots and the future. I think—often in my talks, I put up a picture of a latest model S class Mercedes and I ask the audience what it is. Some of them look at it and say it's a car. Some say it's a Mercedes. And I say, no, it's an elder-care robot because what it does is it's got all these safety systems in which is letting people drive safely longer than they could have before. And that, I think, is a—and it gives people dignity to control their lives longer.

If we look at the population inversion in Europe, Japan, lagging a bit in China but coming there, too, and certainly, North America, I think we're in for a pretty hard time where the elderly will not have services available to them. And there's going to be a real pull on technology to provide those services.

So I would be worried about that because we've not taken immigrants to—I use my mother in Australia as an example. Her life—because she's not very well. She quite old. She's not able to go to the bathroom by herself, not able to get into and out of bed by herself. So her life is—immigrants who've been in the country for less than three months, want a job, they come, they're nice people, they help her with these very personal things but they didn't come on a refugee boat to help an old lady go to the bathroom. They—as soon as they can, they get a better job. So my mother's life is a succession of people she's never met coming into her life, interacting with her in the most intimate ways that she would never have let my father see her participate in during her life. I think wanting, you know, just the technology to be able to get into and out of bed when you choose, rather than when the orderly comes, is something that's going to drive technology and drive robotics. So I think there'll be more, not robots that don't have common sense, but more simple robots in our lives for the next thirty-five years.

GUPTA: Can I just add—yeah, two more points and first is, like—I'm echoing immigration, being an immigrant myself and having a hard time staying here and every given point of time is a pain. I have taught—I mean, every Indian who is here getting a PhD thinks about going back to India because they don't want to stand outside an embassy for a hundred—like hours and hours and having the pain.

So—and further, our immigration policies are probably not treating technology as good as money. For example, one example which I always like is that in the green card, like the permanent resident process, you have a category where if you invest $5 million in U.S. you can get a permanent resident. Why don't we have a kind of category where if you want to open a start-up—if you have ideas to open up start-up, you can raise venture capitals, you can still stay in this country is you want to? We are not treating technology at the same level as—and I have example. Like, I have a student who wants to open up a start-up. He has a really good idea but he cannot open because he's an F1. That kind of hurts the economy here.

The second point I'm going to sound a little selfish but I think I still have to say it. The research funding is, in U.S., it's going down. In China, it's doubling every—like in the—from 2008 to 2012, the research funding in China, which is going to fuel all this technology drive, is doubled. You can see the research outputs. They are increasing. They are doubling. We are going down. Europe is even better than U.S. That's what is the most surprising part to me. I would think given the European economy, they would be worse. But if you think --- talk to researchers, like me, Europe has much better way of, like, in the research funding than America has. And I think that's kind of—also tells you that you need to invest more in people like, not me personally but every—every researcher.

THOMPSON: You made a pretty good case of investing in you, too. I mean, and I—if anybody's in this room, this guy's doing some very important work here.

Rodney, I want to go to something that relates to this and something that you just said, and something that, Abhinav, that you've talked about in a previous panel of yours that I watched, and that is the example of the robot taking care of the older person, whether it's your mother or not, and that is that robots are very good at certain things. They're very good at lifting people. They're very good at carrying things, but they're not good at emotions. They're not good at morals. So how does our society change as robots take on a more profound role? And how does your mother's relation change if she's being cared for by a robot who's good at some things but can't empathize?

BROOKS: Well, I think we are going to see people and robots together. You know, I much prefer going to an ATM, which doesn't empathize at all with me, and in fact, every time I go to the ATM it ask me whether I want to speak Spanish. It's been doing that for ten years. Doesn't it know yet I don't want to speak Spanish? But still, I prefer to go to the ATM than to a human teller. We don't have to have the machines have empathy for them to be useful. We don't have to have the robots that are helping the elderly have empathy. I think there's still a place to have people in the equation, but the nurses and the caregivers don't have to be doing all the physical labor. They can be spending much more time just talking to people.

MCAFEE: If you define empathy a little bit narrowly as awareness of the emotional state of another, technologies are actually getting really good at that. The—there is a system now that's better than the average guy at recognizing the emotion on somebody else's face.

BROOKS: Yeah, but that's a low bar.

MCAFEE: It's a low bar, right? That means almost nothing, but it's there.

THOMPSON: They probably dress better than the average guy, too.

MCAFEE: That's right.

THOMPSON: Abhinav, do you agree?

GUPTA: I actually tend to agree a lot more—I mean, that robots don't need to have emotional—I mean, there's a lot of research, again, in this field. There are people doing research in the field called human-robot interaction, which is just designed to make humans feel comfortable with their robots around them. And I think that's pretty good. We should do all this research and it's—I don't disagree with that research. But I agree that for getting applications out there, for getting products out there, we probably don't need to have it immediately. With time, if this happens, if the robots develop empathy and so on, it would be pretty—it will be nice but I think we are—we should, like Rodney's example of ATM. I think it's a pretty good example that I also personally sometimes feel—I would feel freaky if my ATM start talking to me nicely and I was like—maybe it's like the education of me. Like once I have done enough number of times a day at my ATM I will start to feel comfortable. But if I think of it this point of time, I will start to—still feel freaky. I mean, maybe --

MCAFEE: Your freak-out's going to decrease super fast.

GUPTA: That's what I think.

MCAFEE: I remember how freaked out I was the first time somebody out there on the Internet took a—put up a picture and tagged me in it. It felt like a crazy privacy—and now it happens all the time.

I hear this argument once in a while that our humanity is going to be compromised when the robots gain any kind of emotional capability or ability to empathize. I fundamentally don't understand that argument. Our pets have that ability. They've had it for thousands of years. We humans are still plenty human. That argument just makes no sense to me.

THOMPSON: And my children are quite attached to their Roomba. They were feeding their Roomba little bits of bagel the other day because they thought the Roomba was hungry because the floor was too clean.

MCAFEE: And if you think about some of our most vulnerable populations, you brought up the elderly, think about people—kids with autism who are doing a great deal of learning and improvement with these technologies that don't mind the endless repetition that goes along with that.

I just read about a really interesting smart phone app that just by listening to voice tone, not even the content of a phone call, can help someone with bipolar disorder understand when they're in the middle of a swing and intervene more effectively in that. These are humanity-increasing technologies, not decreasing.

THOMPSON: All right. So we're at the Council on Foreign Relations. Let's talk a little bit more about the world balance of power. And you believe that—you were making an argument on the phone yesterday, or the day before yesterday, that Mexico was recently gaining on China, wages were shifting to—jobs were shifting to Mexico that are now going to shift back to China because of the trends we're seeing.

Give us a very brief synopsis of what's about to happen to the world power dynamics because of what we've been talking about.

MCAFEE: You mean on the economics front or the security front?

THOMPSON: Yeah. Let's do military next. We have about five minutes until we go to Q&A, so let's talk about global economic power and then let's talk briefly about the military.

MCAFEE: I think the shortest thing I could say on that is I would rather have our problems than just about anybody else's. The wage arbitrage, as a national growth strategy, I think is going to become dicier as the technology can do more and more at lower and lower price points.

And as we were talking about earlier, I had the chance to ask David Petraeus and Condoleezza Rice a little while back, not about the arms race but about the geek race. And they—while cybersecurity is scary as hell, they were really optimistic that the United States was winning the geek race because we do a good job of generating them ourselves and because some of the geekiest—that's a term of praise, by the way—some of the—a lot of the geekiest people in the world still want to come here to build their lives, thank heaven.

THOMPSON: Rodney?

BROOKS: Well, I think we are seeing some shifts in manufacturing. China, when you go in and you talk to the big manufacturers there, the biggest problems in mainland China are recruiting and retention. There isn't an endless supply of cheap labor anymore in China. And it's now true that the labor rates in Mexico are lower than in China.

That said, I think sort of some catastrophic event, the smart phone, laptop, that sort of manufacturing is going to stay in China for the foreseeable future because they have built a supply chain which you can't just migrate.

So what's happening in Mexico is all those boxes you see in your doctor's office, which are essentially one circuit board with plastic wrapped around them, that's cheaper to build in Mexico. It's doesn't need supply chain. Sewing goes to Bangladesh because you don't need a supply chain. But China has this asset, which is this lock on the supply chain, the vast volume high-technology electronics manufacturing. And the Chinese government recognizes that is—got rid of all import duties on robots and is even subsidizing Chinese companies buying robots because there's not enough labor.

THOMPSON: Wow, that's interesting.

Let's talk about the military for a minute. Rodney, I'm going to take a wild guess that you've been approached by DARPA roughly 150 times in your career, and yet, you made vacuum cleaners.

BROOKS: Well, actually, that company did two things. That was only 50 percent of the business. The other half of the business was the PackBot and its brethren. We had 40,000—4,500 of them in Afghanistan and Iraq.

THOMPSON: What did that robot do?

BROOKS: Doing IED—so at the start of the conflicts in Afghanistan and Iraq, U.S. doctrine was you put the youngest—they didn't say but this is what happened. The youngest guy in the unit got in the bomb suit, went out with a stick and poked the roadside bomb. They're no longer allowed to do that. You send out the robot. All sorts of good reasons for doing that. So—I could go into it in much more detail.

So robots are helping the military. We see the drones, which are tele-opped, so that's a little different. But what we saw in Iraq and Afghanistan was the asymmetry of everything. What I called at the time, which maybe I have to change the title, I called at the time Radio Shack technology. We would see in one area of Iraq some new technology for a new sort of roadside bomb. Someone came up with something and it wasn't expensive technology. It was stuff you could buy at Radio Shack and put together. We would then seek—we'd have to come up with a countermeasure. You could then see that migrate across the country in a period of about three or four weeks. So very simple new ideas are all the—the asymmetry is always harder to battle against, or has been up to now.

MCAFEE: In addition to which, the other dynamic that's going on is that we are democratizing access to incredibly powerful technologies. And overall, I'm really optimistic about that phenomenon but we can rest assured, the bad guys are going to use the same tool kit to do us harm in these asymmetric ways. And the tools they have are going to get better and better as well.

THOMPSON: Abhinav, what do you see about robots in our military future?

GUPTA: I mean, a lot of research is being funded by military. There's no doubt about that. And I think one area where, again, military is being really helped by artificial intelligence, robots, I think these guys have told a lot, is, again, I think the example of—(inaudible)—the example is—Palantir Technologies, for example. They are—they take all—exploit all this data which they get and they use all this data to help in cyberterrorism and terrorism as well. You can track and you can understand much, much better. You can predict. I think this predictive analytics is really, really getting better and better with this AI and it's going to help us tackle these challenges with this terrorism and other things. Of course, we are also going to have problems in foreign policy because you are able to eavesdrop on some people and they are not going to feel happy about it.

BROOKS: Yeah, and just to follow up a little bit on this asymmetry. What the—those big companies have is they can employ vast, vast clouds with GPUs. And the small people may be really good at doing Twitter, as we see with ISIL, or ISIS, but they don't have the access to those vast server clouds that—where they can do the big data analytics the same way that the governments can. So there's a place where maybe the asymmetry hurts the little guys.

GUPTA: The only problem is that these GPUs are getting cheaper and cheaper so quickly that who knows when this GPU will be like $100 and even --

MCAFEE: GPU for the non-geeks in the audience is the graphics processing unit that's in your Xbox.

GUPTA: Thank you for pointing that out. It's such a common term I don't even realize.

THOMPSON: All right. My predictive analytics suggest that there will be plenty of questions. So at this time, I would like to invite members to join our conversations. Reminder, this meeting is on the record. Wait for the microphone, speak directly into it. Please stand, state your name and affiliation. Please limit yourself to one question and keep it as concise as possible so as many members as possible can speak.

So in the second row, please? Both of you in the second row, first and then second.

QUESTION: Bettye Musham. Is there any crossover technology between robots and drones, because isn't a drone really a flying robot?

BROOKS: Yes. There's a lot of—there's a lot of common technology there that both—both of them—I mean, the drones being the small drones, not the $70 million drones, the small drones that we're - the FAA is currently arguing about. Both today's robots and drones have benefitted greatly from cell phone technology. It's brought down the cost of sensors, lots and lots of sensors very cheap, system on a chip, very powerful computers. They both use them so there's a lot of commonality there.

GUPTA: I think what I'm quite certain—I mean, you can think of it as the planning for rewards. I mean, the normal robot, like Baxter, or a drone will be very similar because they have to see the same world, they have to react to the same world and that's why I would say there is a very high overlap. It's just that the capabilities of these two robots are different and that's it.

THOMPSON: OK. Again, in the second row?

QUESTION: Robert Kaiser. I'm curious, Mr. Gupta, if you could give us some examples of what NEIL can do. I'm having a hard time imagining what the payoff from NEIL is in a practical way.

GUPTA: OK. So I think this is something which I have to—I was hoping someone would ask it. And there was a discussion on the phone as well that—why understanding? Why do all this common sense and something like that. Think—let's think of autonomous cars. All of think autonomous cars are going to come very soon. I'm sure we are reading it everywhere that autonomous cars are coming tomorrow, or something like that. Let's think your best pattern-matching technologies trying to solve autonomous cars. So you have to understand that our world has two kinds of distributions; things which will happen a lot. Fifty percent of these things will happen a lot. For example, cars driving on roads, cars taking turns. You will see them every day. There's lot of data that—(inaudible)—of these common things that happen in our life. And autonomous cars, they'll be really good at understanding these common things that happen every day in the life.

Now think of rare things that happen. Now you will be surprised that 50 percent of the things that are in our world are very, very rare. By rare, I mean they might happen once in a month or even much, much less. How would you adapt to those? There's not enough data. Your—all your pattern technology—matching technologies will start to fail. So what will you do in that case? In that case, you will need common sense.

And a very simple example of this I can tell you is let's, again, go back to autonomous cars and think of this as—let us suppose your car is driving on a road, it's an autonomous car. There is a wire which is crossing the street and this wire breaks on one end. The wire goes like this. You all know what's going to happen because it's still attached to one point. It's going to show a pendulum motion and it's going to come back like this. What an autonomous car sees, it goes like this. There's nothing in front of me now, I can go. And it's a live wire so you don't want to touch it or cross it.

But it has no idea how the world works. It has never seen wires crossing like this because how many times have you seen wires like this?

So we humans can reason out and say, OK, I should stop, I should not drive through this, whereas an autonomous car will continue driving through it and it might get electrified or something.

I am just, again, a very rare, rarest of  examples and you can think of many more examples where you would need common sense like this.

BROOKS: Can I just reiterate that? John Leonard at MIT, one of the founders of a technology, SLAM, which is critical for autonomous cars, put a dash cam on his car and had his students count the events which were unusual and which were not going to be in the big databases and, indeed, came up with numbers very similar to what you say.

THOMPSON: So you're hope is that through your software you'll be able to ultimately teach an autonomous car not to do those things?

GUPTA: I mean, it might be able to reason out these things. It might be able to learn about pendulum motions and it can say, OK, if I see things like this, it has to come back while it's still attached to that thing on the top.

THOMPSON: OK.

In the fourth row? Yes, right there on the left.

QUESTION: Stephen Blank, University of Ottawa. This is about jobs and strategy. A little context quickly. I graduated from high school, a rural slum high school in 1957. We didn't have very much academics. People didn't go to college, but most of my classmates had good jobs in the sense that they lived—they consumed as middle class consumers throughout most of their life.

OK. Question today. How do we provide access to the jobs, however many are being created? Do you educate people up for those jobs or do you make those jobs more accessible to people who do not have the educational apparatus—don't have an increased education apparatus? Which do you do, up or down? How do you do it?

MCAFEE: I don't think that's an either/or question. I think both are clearly going on. And like Rod points out, the bargain seems to be shifting kind of quickly these days away from guarantees of something that felt more like a guarantee of a job and benefits to a series of more spot engagements, and Uber is a really great example of that.

Opinions on that shift are really sharply divided. I personally am a huge fan of companies, like Uber, not because I like getting from point A to point B, but Uber is putting labor back into the economy at a time when we really need labor. If we're not happy about some characteristics of that job or what people—the life of an Uber driver, great. Let's intervene with policy. If we don't like where that person is, let's top them up with something like an earned income tax credit. Let's make educational options more available so they can upskill themselves if that's a good path forward. But for heaven sake, if we've got companies that are adding labor to the economy, why are we demonizing them? It doesn't make sense to me.

BROOKS: But to—sort of your question up or down, without robots that we're putting in factories, we're making it so the ordinary factory worker can learn—teach themselves how to program it. You know, if I had my smart phone in my hand, imagine you had to go take a community college course in order to learn how to use a smart phone. They wouldn't have taken off as they have. What we put in—what the companies that build them put in there was technology that teaches the consumer how to use them as they first use them and so it brings the consumer into that world.

I think from looking at industrial equipment, the way to go is to have the industrial equipment be smarter and bring the person into the world and be able to understand it so you don't need to go off and be trained in a particular technology that's going to change in ten years anyway, so it's got to be continuous on-the-job learning, and technology has to provide that.

MCAFEE: It also appears that about, again, since the turn of the century, you still want to go to college because the wage premium on a college education is a pretty good one. But even college-educated people are not seeing their real wages go up. So this idea that getting more skills will always put you on an upward trajectory of earnings, that really appears not to be true, at least since the turn of the century. The only folk who it looks like are seeing their real earnings go up are the 1 percent, the 1 percent of the 1 percent.

QUESTION: We have a—(inaudible)—core of people in this country who are not going to go to college, and that's the reality. Those are the people who need some sort of access to jobs.

MCAFEE: Which is why I'm a huge fan of Uber.

BROOKS: Yeah, those Uber drivers don't have to go to college and they're still being able to be—do something pretty good.

THOMPSON: OK. Do we have a next question here on the right? Whichever one of you --

QUESTION: Steve Rodriguez. I'm a venture partner with a VC company here in New York. So an area I focus on is technology with military applications. What in your, I guess, estimation, is the biggest barriers to companies like iRobot or even like Helen Greiner's company, CyPhy, is a former colleague of yours? What's their biggest barrier to entering or dealing with the military, from your perspective? Yeah, I'll stick with one question.

BROOKS: Having living that with iRobot, we would not have gotten anywhere near as far if there hadn't been some wars going on that we had an answer for. Very hard to break into the acquisition cycle for a smaller company. The bigger companies really do have it locked down pretty well and it's very hard to get in. It was just that we had a technology that no one else had but it never got into a program of record. As soon as the wars ran down, iRobot's share of—the percentage of their military business decreased rapidly within the company. So it's the—the companies that are there are still taking enormous parts of the military budget on things, you know, as throw weight, if you like, whereas maybe the threats of the future are going to be very different sorts of threats, and that worries me.

THOMPSON: The gentleman who is also in the fourth row?

QUESTION: Les Baquiran, Alpine Capital Advisors. My question is that I think we could all agree that technology has lowered the price for consumers in a lot of different things, but there's two core industries that it really hasn't impacted, at least to my expectations, and that's really health care and education, because health care and education have far outpaced normal inflation. So what do you see in the future in those particular sectors, technology, AI, robotics, helping to lower the cost of both?

MCAFEE: Larry Summers makes this point in a really vivid way. He says if you go look at the bundle of stuff that makes up the CPI, the Consumer Price Index, where we get our inflation figures from, it was set to 100 in—I'm going to get the year wrong—like thirty years ago, forty years ago.

(UNKNOWN): (OFF MIKE)

MCAFEE: '83?

(UNKNOWN): '82.

MCAFEE: '82? So that—everything was 100 in '82. A color TV today is 10. A hospital visit and a college education are at 600. So I couldn't agree more. We are not seeing the kind of crazy price declines in these huge industries that we're seeing elsewhere.

What's going to bring them there? Opening up these markets, getting rid of the incumbent's advantage. Education, my home industry, is a cartel. And the only thing that I think is going to change the real price of education is when employers start valuing signals that don't come with an expensive—other than an expensive degree from an accredited university. I can't wait for that day to come.

BROOKS: Let me talk about health care very briefly. I think where it impacts health care is when you change the rules and change the way it works. So putting sensors that detect when a person is going to need help is going to—so that you intercede earlier in the situation, don't have them going into hospitals, staying there. That's going to start changing things because the hospital visits continue to increase.

We're seeing companies start to do that. It's getting bundled with things like Nest, with home security systems. And it's going to be that children of the—their parents investing in that technology to keep their parents healthier longer at home.

And then there's—most of—a big part of health care costs go to the last few months of life and that's a big moral issue. How do we rethink what we want for ourselves, you know, in the last few months of life? And the problem is you never know when the last few months are so it's a terrible bargain you have to make.

GUPTA: I think the internet of things is basically what you are pointing out is that the sensors, the smart homes and everything, I think they're really going to change health care, and especially once we have enough analytics on them. We are still in the process of developing data. I think it's the—I mean, for me, when I think, it's all about data. And now, at this point of time, we don't have enough data in the health are, but because we are now starting to purchase a sensor that's become so, so cheap, it's now at a point where it's starting to go and basically create applications out of this.

BROOKS: Even simple things right now. Some companies are putting sensors in—they notice when a person isn't getting out of bed and they—the health supplier then sends someone to check on them before they sit there for three days and get into a chronic situation.

THOMPSON: My sensor just went off and said there are more questions. Over here, please?

QUESTION: (OFF-MIKE), from McKenzie. Mentioning the internet of things, the actual internet, we've been reminded vividly, is very hackable, very insecure and we're anxious about it. So as we move the internet of things and more robots, are you worrying more about making sure your stuff is hack-free? How are you thinking about that, Rodney?

BROOKS: That's a real problem.

THOMPSON: Have you ever had Baxter get hacked?

BROOKS: Well, interestingly, you know, Baxter, we started selling him to small companies that didn't have networks, so Baxter is currently not networked. So we have not had Baxter hacked.

GUPTA: I think it is a real fear. I mean, there is the more and more power goes into the hand of computers. This is going to—but I think, again, the hope is that we are—I mean, there is lots of research a lot of work on this field and hopefully we are not going to—I mean, hopefully we are not—will be in a bad position. So that has to be a downside.

THOMPSON: Wow. Pure pessimism across the panel.

OK. Back on the left, please?

QUESTION: Darryll Hendricks. The panel, earlier, made it clear you didn't think there was an imminent existential threat from, you know, robots. And I'd be interested in the panel's views on consciousness. Is that something that you see as, you know, coming in decades, coming, you know, hundreds of years, or never?

THOMPSON: That is a hard question.

BROOKS: Well, let me say I've written about this some. In principle, we can have conscious machines because I believe we are machines. You know, when you take biology 101 it doesn't say "and then the two molecules, you know, and the receptors come near to each other and the soul intervenes and they lock together". We have a mechanistic description of ourselves and that's what modern science is based on.

Whether we're smart enough to build such a machine is one question, and whether we understand it enough at the moment for it to be a problem that we can work on today, I think is a totally different question. Some questions people had back in the days of Leonardo da Vinci but it took hundreds of years to get there.

The conscious—in my group, my lab, I used to—we used to call it "the C word". We were not allowed to say it because—and we had this deal where we'd go to the international consciousness conferences once every ten years to see if there'd been any progress. And so far, I haven't detected any.

THOMPSON: Can you—let's pause for a second. I didn't ask you to define deep learning earlier in the panel, but I was tempted. I'm going to ask you to try to define consciousness as it applies to a robot.

BROOKS: Oh, I can't define it as it applies to a person. I mean, we only have this --

MCAFEE: I've heard it called a suitcase word, in that you put in it whatever you want and you kind of carry it around. Philosophers have been debating what this thing is for hundreds of years. It's an incredibly hard problem. I couldn't agree—I rarely find myself saying this, I could not agree with Rod more on this. Conceptually, we're machines. Conceptually, there's no reason that we can't—we don't have the faintest idea of how the brain actually works. We understand some things pretty well. This is a huge mystery.

THOMPSON: All right. And a new rule; when you ask a question, please identify whether you're a human or a machine.

In the front row, please?

QUESTION: Hi, Laeticia Garriott de Cayeux and I'm a humanoid. So you all took, across the panel, the position that we're not going to be the generation that pull the plugs for this AI that surpasses us, and that's your opinion. Can we hear arguments other than—and we hear you, it's extremely hard. Computational power is not today the capability where it should be. But what are your arguments that we should trust and go with your opinion and not the opinion of Elon Musk, Stephen Hawking and others?

BROOKS: So if we look at actually one of the places which makes this prediction, did a careful study of the last sixty years of such predictions, the mode of where they—the predictions were—were always fifteen to twenty years in the future. So we know the vast majority of those predictions were totally wrong because everyone always predicts fifteen to twenty years in the future. So historically, people have been wrong in predicting. Why would they be right now? We have no reason to believe—to know that they're right now. And certainly, Elon Musk has no reason to believe it. He's just making that up.

I liken it to—this whole thing about the singularity, I liken it to witches. We used to burn witches. Why did we burn witches, because we believed by existed and then we found them. So if you believe there's going to be a singularity where there's, you know, the great technology changes us and lets us live forever and have everything we want, then you're going to expect it's going to happen on a date.

QUESTION: Yeah, I'm asking from our perspective, not their perspective. I'm just asking yours.

BROOKS: Because I've been working in this for forty years and we've made that much progress on that much.

MCAFEE: I went to a conference in January of this year exactly devoted to this question, where there were people who spent their careers in the discipline who believe very differently than you do, who were giving these predictions about 2035, 2040. What was interesting is, again, they were twenty years from right now, which is the common thing that the singulatarians always say. Their point was that we have made progress on one or two of those really, really fundamental challenges and we feel like we are on a trajectory that looks like this for making progress on the rest of them because we've got so much computational power. We've got better insights into the brain. We can map things down to the individual neuron on the brain that's going to let us understand how it works. I hear their arguments. I'm still with you on this.

BROOKS: Elon Musk was at that.

MCAFEE: Yes, he was.

THOMPSON: Abhinav, where are you on singularity and --

GUPTA: Yeah. So I actually think it's very—(inaudible)—and there's a very good reason to believe so. First of all, what does singularity mean? I mean, what are we talking about here? We are talking about computers deciding themselves what is good, what is bad. Right now, we are making all these applications with a task in mind. And we are thinking of making a computer with no task deciding what is good for themselves, what is bad for themselves and just hypothesizing these kinds of questions which we need to solve. And I'm 100 percent sure there's actually no serious researcher who's thinking of these questions at this point of time.

BROOKS: But I—I want to say one more thing.

MCAFEE: Rod feels strongly about this.

BROOKS: That the basis of this belief is that AI is going to get so good that it's going to self-improve, it's going to rewrite itself. We do not have a single system that can write a ten-line program, like any person who takes a first semester of programming. It cannot restructure it, even in a ten-line program. If we were going to have something that could take these enormously complex systems with billions of lines, surely we should've gone somewhere between zero and ten lines of code. We're not there.

So what this whole argument has made me do is decide, that's my project. I want to build that system.

MCAFEE: At the same time—I'll take the other side of this argument. At the same time, there was just a paper published in Nature. A pretty serious team of geeks out of London built a machine to learn to play '80s vintage Atari videogames without being taught the rules or any of the right strategies for playing them. That computer is now the world's best Space Invaders player. It's now the world's best Breakout player. You will never score a point from it in Pong. It's a really impressive achievement. I'm still with you.

GUPTA: Again, the task is defined here. It has to do good on that task. This whole question of meta-thinking—and this is what Rodney is saying—what—I mean, we need to solve meta-thinking and I don't think - I'm pretty serious about it. There's no one actually who's taking on meta-thinking right now. We are still at a much lower level. We still want to solve simple tasks.

BROOKS: And last thing. Before we get robot—before we get AI that's going to take over the world, we're going to get really annoying AIs. And we'll be annoyed at them and stop them. It's not a Hollywood movie where suddenly it comes.

THOMPSON: So the award for the question that most flummoxed the panel goes to the woman in the front row.

Now in the fourth row, please?

BROOKS: She's a robot.

QUESTION: Rick Thoman, Corporative Perspectives, Columbia University and also former executive at Xerox and IBM. I wanted to talk a little bit about the futurologist aspect of all of this. You know, a major consulting firm was asked by a big telecom twenty-five years ago to do a study of how big the world's cell phone business could get. And they concluded it could get to be 150,000 units. Now of course, they couldn't image that the cell phones didn't weigh twelve pounds and didn't cost $7,000.

So talk to me about a world of robots in which they costs $100 or $10. And are there such a thing—is it imaginable that there could be personal robots that are two inches by two inches or one foot by one foot? So --

MCAFEE: It's not imaginable; it's guaranteed. I completely agree; futurology is a cheap ploy and very few futurologists ever get called on how bad their predictions actually are.

QUESTION: Right. Exactly.

MCAFEE: So it's a sucker's game. The things that I can say with great confidence, Moore’s Law, this exponential improvement in the elements of computing is not about to run out of gas. We've got a generation more of it to go. The amount of data available in the world is going to continue to explode, especially as the internet of things and all these sensors happen. And geeks out there are going to take that computational power and that ocean of data and do things that are going to astonish us. Those three things I know.

BROOKS: I agree.

THOMPSON: In the middle, on the—in the right? Yes, please.

QUESTION: Hi, David Preiser, I believe I'm a human although Gödel would tell us it's not possible to actually know. So I read your book. I thought it was great. The --

MCAFEE: Thank you. I think we're done.

QUESTION: No, no. So—but if I read some of the import of it, it seems like we're—we're bisecting ourselves between those geeks who will create amazing things and those who are not able to enter that promised land, don't have the drive, the intellectual capacity, the interest, whatever limitation there is. And yet, all those people will need something to do.

And I wonder, and I heard something that was interesting, which is, well, isn't part of the potential of the machines to dumb down a whole bunch of tasks so that people can feel like they're participating in this world and doing something useful, when in fact, the machine is doing most of the work?

I once read a story like that where a doctor had this fantastic mechanized doctor's bag that was programmed with all medical knowledge and he really believed he was a doctor, when, in fact, the doctor bag was the doctor. So what do you think about that?

BROOKS: I want to answer both questions at once. One of the possibilities that will happen is that we will see the maker movement in the same way we saw the Homebrew Computer Club takeover computation from the big companies, the NCRs, the Honeywells, all those companies which were great computer companies, Deck, et cetera. And that Homebrew Computer Club turned in to be the fountain of productivity.

The maker movement might turn into the fountain of productivity and there'll be much more—many more people doing making than there have been in the past and a whole locus of where / what makes a company may change, in the same way computer companies completely got obliterated in the '80s by a whole new generation.

And I'm not saying, you know, using a cell phone analogy, making it easy to use doesn't mean it's dumbing it down so that people aren't actually doing stuff with it. It just makes it more convenient to do things and you can get up to the threshold of using them. I think that's different from fooling them. I'm not arguing—I don't think it's they're fooling them. I think we can still provide useful interfaces for people so they can become much more powerful and much more powerful in their making. The same way if you're at Xerox, you know, the laser writer, the laser printer let people do typesetting where they didn't—they couldn't do it before.

MCAFEE: What's interesting is that as the technologies have proliferated and really democratized access to innovation and to making things—we've been doing this for thirty-plus years—the economy has moved in the other direction. The economy has moved toward more concentration, the big guys getting bigger, the workforce becoming more polarized instead of less.

So when I hear about this promised future land of the maker movement that's going to restore middle class opportunity, it's arguing exactly against the trends that are pretty clear in the evidence so far. So I'm really worried about this polarization of opportunity and of mobility. I think it's already a challenge. I think it's going to get a lot bigger. And what we do about that is going to be the challenge, I believe, that defines us for the next generation or so.

The good news is the problem we're going to face is not the problem of scarcity, is not the problem of not enough to go around. And I find it really important not to lose sight of that fundamentally fantastic news. Keynes was right; we're going to get away with what he called the economic problem of allocating in the face of scarcity. Fantastic. The distribution problem is a tough problem, but like you know, we put a quote from Voltaire to start off one of our chapters. He said that work saves us from three great evils; boredom, vice and need. Of those three, I find the need easiest to take care of.

THOMPSON: Back there on the right?

QUESTION: Thanks. Jamie Metzl, I'm an investor and novelist. I would love if you could say a few words about the future of robotic pets.

BROOKS: Well, I do apologize for having let loose the genre of pets riding on Roombas by having the Roomba, and therefore—people do treat their Roombas with a reverence that I could not have believed. Hardcore—hardcore Marines in war zones treating their PackBot with a reverence that I could not believe. So it is not too hard to trigger some of our empathy for machines. And there will be startups which try to do that because they're going to make a buck and so there will be machines around that some people treat as pets. Other people have disdain for. We saw that just last week when Boston Dynamics, now owned by Google, showed their new four-legged robot walking and the people were kicking it and they got eviscerated in the press for --

MCAFEE: For kicking the poor robot.

THOMPSON: All right. We have two minutes left. Last question here on the right, please?

QUESTION: Eyrique Miller from JPMorgan. I guess this ties into the last two questions. So I have two daughters that are ten and five. They have a Zoomer, both of them, which is the dog that you can teach and do tricks. And I guess my question is, and this is sort of skipping the current generation of college students, but are we doing enough for the youth that are, you know, at the younger age to educate them in terms of robotics and actually to be participants in where you think this is going?

MCAFEE: No, absolutely not. We're teaching --

QUESTION: But what can we be doing then if we're not doing enough?

GUPTA: We are not doing enough but there is, I think, again—I agree that we are not doing enough, first of all,  I should say that. But I—for example, in Carnegie Mellon, I see every day these capable students coming, like, four-year kids coming and making robots in the labs. I think that is the best hope we have.

We need to, first of all—I mean, if you don't have to wait until college education for robots to be inside their—inside your education. I think we need to take it much faster. It could be—start with a very small system—small, like easy things they make on. But I think there is—I mean, there should be something which could be done for that this education goes to—starts as early as possible, rather than waiting until the last moment, like the last four years, the bachelor technology or bachelors, to do this thing.

BROOKS: There is a little—and this is a—this is—you spoke as a parent. I'm going to speak as a parent. One of my daughters has started a company to build STEM toys for girls age seven to nine and it was actually at New York Toy Fair two weeks ago. There were a whole bunch of actually women-run companies building engineering toys for girls. So we're starting to see that. That may help the next generation be exposed as they're getting out. And they're getting pretty wide distribution. So there are entrepreneurs wanting to do this. Her major investor is Mark Cuban, but that's because he has daughters, too. So the UVCs—if you'll just invest in girls' toys companies that will be a lot better.

THOMPSON: Elon Musk will not be investing after this panel.

Thank you very much. We all have to get back to our jobs before they're taken by robots.

Thank you to Malcolm and Carolyn Wiener and thank you to our panel.

END

Top Stories on CFR

United States

Each Friday, I look at what the presidential contenders are saying about foreign policy. This Week: Joe Biden doesn’t want one of America’s closest allies to buy a once iconic American company.

Immigration and Migration

Dara Lind, a senior fellow at the American Immigration Council, sits down with James M. Lindsay to discuss the record surge in migrants and asylum seekers crossing the U.S. southern border.

Center for Preventive Action

Every January, CFR’s annual Preventive Priorities Survey analyzes the conflicts most likely to occur in the year ahead and measures their potential impact. For the first time, the survey anticipates that this year, 2024, the United States will contend not only with a slew of global threats, but also a high risk of upheaval within its own borders. Is the country prepared for the eruption of election-related instability at home while wars continue to rage abroad?