The Megyn Kelly Show - The Benefits and Dangers of Artificial Intelligence, with Nick Bostrom and Andrew Ng ｜ Ep. 151

00:00:00.580 Welcome to The Megyn Kelly Show, your home for open, honest, and provocative conversations.

00:00:12.380 Hey everyone, I'm Megyn Kelly. Welcome to The Megyn Kelly Show.

00:00:15.600 Oh, we have a fascinating show for you today. Fascinating.

00:00:18.920 It's about artificial intelligence.

00:00:21.140 I've been asking my team to line up a show on this and we have the two greatest guys,

00:00:25.020 the most brilliant, just greatest guys to talk about it with.

00:00:28.800 Like, don't you want to know where this is going, right?

00:00:31.880 Like, okay, there's Amazon Alexa and then there's something called super intelligent computers

00:00:37.560 that are going to take over the world and possibly eliminate humanity.

00:00:42.160 Opposite extremes.

00:00:43.360 It can be wonderful and it can be life-changing in a great way and it could also potentially

00:00:47.320 be life-extinguishing if it gets into the wrong hands and so on.

00:00:50.840 So we've got all these angles covered.

00:00:52.540 You are going to love, love, love this show.

00:00:55.540 We're going to kick it off with a guy named Nick Bostrom.

00:00:58.800 He's a professor at Oxford.

00:01:00.920 He's the director of something called the Future of Humanity Institute.

00:01:04.980 He's done so many things.

00:01:07.160 He's been a teacher at Yale.

00:01:09.000 He did his postdoctoral fellowship at Oxford.

00:01:11.640 He's the founding director, as I say, of this Future of Humanity Institute.

00:01:14.500 That's at Oxford as well.

00:01:16.120 Researches the far future of human civilization.

00:01:19.820 A professor of philosophy at Oxford.

00:01:21.440 He has been included in Foreign Policy's Top 100 Global Thinkers list repeatedly.

00:01:28.600 He was listed by Prospect Magazine and their list of world's top thinkers.

00:01:32.000 You get it?

00:01:32.600 Sensing a theme.

00:01:34.100 And he's probably best known for his incredibly bestselling book, Super Intelligence, Paths,

00:01:40.560 Dangers, Strategies.

00:01:42.600 It's been recommended by everyone from Elon Musk, who's a huge fan of our guest, Nick

00:01:47.000 Bostrom, to Bill Gates.

00:01:49.140 And he is one of the leading thinkers on where super intelligence, what it is, where it's

00:01:54.120 going and how it needs to be handled.

00:01:56.540 That's sort of where the machines become smarter than the humans.

00:02:00.420 Now we're going to talk to him.

00:02:01.260 Then we're going to be joined by a guy named Andrew Ang.

00:02:04.420 He's also incredibly brilliant.

00:02:06.020 So excited to talk to these guys.

00:02:07.480 He's the founder of deeplearning.ai, co-founder of Coursera.

00:02:13.220 Coursera is huge.

00:02:14.480 This is the world's leading massive open online courses platform.

00:02:19.420 He's also an adjunct professor at Stanford.

00:02:22.060 He's the founding.

00:02:23.460 He was the founding lead of the Google Brain team.

00:02:27.120 He coined the term Google Brain.

00:02:28.960 He was the chief scientist at Baidu, which is China's Google.

00:02:33.180 There's no Google in China.

00:02:34.260 This is China.

00:02:34.700 I mean, this guy's been he's led a 1300 person AI group for China's Google.

00:02:40.440 All right.

00:02:40.620 So he's done.

00:02:41.180 He's basically been in charge of everything.

00:02:43.340 He's cool.

00:02:43.720 He's a globally recognized leader in AI.

00:02:46.080 And I would describe him as more of a happy warrior when it comes to AI.

00:02:49.160 Very optimistic about it and what it can do.

00:02:51.620 And talk about how it could change your life for the better.

00:02:54.080 And I think you're going to be delighted with the show.

00:02:56.740 And I predict you'll be sharing it with everyone, you know.

00:02:58.960 OK, so we're going to start with our guests in one minute real quickly.

00:03:02.200 Here's this.

00:03:04.700 There is so much that I want to go over with you.

00:03:10.780 Just treat me like I am AI 101 because I know almost nothing about this field, but am dying to know more.

00:03:19.060 And just having read what I've read now of your work and having listened to your TED Talks and so on.

00:03:24.020 I'm terrified.

00:03:26.140 I'm terrified.

00:03:28.220 So let's start here.

00:03:31.520 What is super intelligence?

00:03:34.940 I just use it as a term for any form of a general artificial intelligence that greatly surpass humans in all cognitive abilities.

00:03:45.140 And so, in other words, when the machines get smarter than we are.

00:03:47.640 Yeah.

00:03:49.480 OK.

00:03:50.240 And how likely is it to come into existence?

00:03:56.300 I think it is highly likely that it will eventually come into existence.

00:04:00.860 I think almost a certainty if we avoid destroying ourselves through some other means before.

00:04:07.980 But if science and technology continue to advance on the wide front, then I think eventually we'll figure out how to produce high-level machine intelligence and super intelligence.

00:04:18.440 Is it in the works right now?

00:04:21.520 Well, I mean, in some sense, it has been in the works for a long time in that people have been trying to understand better how the brain works, how to use statistical methods to better extrapolate from past data, how to build faster computers.

00:04:36.840 All these are potential ingredients.

00:04:38.880 And, of course, the field of artificial intelligence has really burgeoned in the last eight years or so with the deep learning revolution.

00:04:51.620 And so there's quite a lot of excitement now about what is becoming possible with machine learning.

00:04:58.320 But predicting how far we are from being able to match and then maybe surpass human-level intelligence is really hard.

00:05:05.960 And I think we just have to acknowledge that there's enormous uncertainty on the timeline of these kind of things.

00:05:13.580 Now, we're going to be joined after you by another guest who his belief is, the way he phrases it, is there's two types of AI.

00:05:24.420 There's ANI, artificial narrow intelligence, and AGI, artificial general intelligence.

00:05:29.660 And he says artificial narrow intelligence is basically like the stuff we've seen already where you're typing on your computer and it recognizes the word you're typing and completes it.

00:05:41.200 You know, or you're, I don't know, maybe Amazon Alexa or the self-driving car, like those things that are improving our day-to-day living.

00:05:49.700 But general intelligence is what you're talking about, super intelligence, which is, that's a whole different realm.

00:05:56.060 And that's the thing, as I understand it, that you're sounding the alarm on.

00:06:00.240 Yeah, or at least trying to draw attention to ask something that would be very important.

00:06:06.480 I think it has an equally large upside if we get this transition to the machine intelligence here on right.

00:06:13.540 But, but, but I do think also there are significant risks associated with this, but yeah, I think it is useful to make this distinction between kind of specialized AI systems that can only do one thing, maybe sometimes at the superhuman level.

00:06:27.380 So for a long time, we've had like chess computers that can beat any human, but contrasting that to something that matches humans, say in our general learning abilities and reasoning abilities that, that make it possible for a human to learn any of thousands of different occupations or to solve novel problems that you've never seen before and to use common sense.

00:06:51.280 So why would we be seeking superintelligence, you know, because we're going to get into the risks of it and, you know, the possibility that machines not only get smarter than humans, but actually take over the world and possibly eliminate humans.

00:07:04.740 Why would we be even going down that route?

00:07:07.600 Why wouldn't we have just seen that future and said, why would we create another being on earth that's smarter than we are that could take over this planet?

00:07:14.860 For the most part, we are not seeking superintelligence, but greater intelligence, like to have, you have some AIs today and it'd be nice if they made fewer errors and were a little bit more capable.

00:07:27.520 But then of course, if we succeed in that, we would want them to be better still.

00:07:30.840 So it's not so much that there are a lot of people who are specifically trying to create superintelligence, but there are huge strivers for making progress in having better forms of machine intelligence.

00:07:43.000 I mean, in general, it's not as if human civilization has some kind of great master plan either, right?

00:07:51.120 I mean, we are not sort of having a hundred year plans for which technologies we're going to promote and which less.

00:07:57.100 So for the most part, things just happen and there are these local reasons why people do things.

00:08:04.580 And that I think is also true for the field of AI.

00:08:08.100 I know that you've said you, a possible scenario is we create a machine, a computer that has general intelligence below human level, but is superior mathematically.

00:08:21.860 And in this scenario, human beings understanding the risks of creating a superintelligent machine would take safety measures.

00:08:28.660 They would pre-program it, for example, so that it would always work from principles that are under human control.

00:08:34.340 We would try to box it in with limitations.

00:08:37.200 We would try to be careful.

00:08:39.420 How do you see the possibility of that machine that we've tried to take these precautions with, nonetheless, on its own, becoming a superintelligent being, for lack of a better word?

00:08:51.600 Well, so I think we will keep trying to make machines smarter.

00:08:55.820 And if we succeed in this, at some point, they will become smarter than us.

00:09:01.960 I think at that point, once you have maybe even weak superintelligence, development is likely to be very fast for various reasons.

00:09:12.400 For a start, at this point, the technology would be extremely economically valuable.

00:09:17.820 So massive investments would flow in to running these AIs on even larger data centers or applying even more human ingenuity to improve them still further.

00:09:28.260 At some point, also, you might get this feedback loop when the AI itself is able to contribute to its own further improvement.

00:09:34.780 So you might get a kind of intelligence explosion where you go from something maybe just slightly human level to something radically superintelligent within a relatively brief span of time.

00:09:49.340 And then the question becomes, would we be able to steer what such a superintelligent system would decide to do?

00:09:58.560 Like, it would be very powerful for basically the same reasons that we humans are very powerful on this planet today compared to other animals.

00:10:07.760 The gorillas are much stronger than us, and cheetahs are much faster. 1.00

00:10:12.940 And yet, the fate of the gorillas depend a lot more on what we humans decide to do than what the gorillas do. 0.74

00:10:18.900 So if you have something that radically outstrips us in terms of its general intelligence, its ability to strategize, to develop new technologies,

00:10:28.560 then it might well be that the future will be shaped by its preferences and its decisions.

00:10:34.640 And it might be non-trivial for us to make sure that those are aligned with our human values,

00:10:40.400 especially if we need to get it right on the first try.

00:10:44.260 Right. You've been saying that for a while, saying if we're going to do this,

00:10:48.340 we have to make absolutely sure that they are aligned with our human values,

00:10:52.680 and there are all sorts of dangers in doing it anyway.

00:10:56.040 I mean, who's going to determine what the values are, and what if not everybody's on the same page,

00:11:00.960 and what if we do it, but it gets into the wrong hands, and people misuse it, and so on.

00:11:07.540 But let me just stick with the gorilla thing. That's interesting.

00:11:09.920 So, I mean, because I've heard you use the example of the tiger, too.

00:11:13.140 The reason the tiger gets in the cage and can be controlled by us is because we have superior intelligence to it.

00:11:19.420 So it may be more powerful, more lethal, but we're smarter, and so we can trick the tiger into the cage and keep it there.

00:11:26.660 And the same is true of the gorilla.

00:11:28.520 And so in the scenario where we have a super intelligent machine, we're the gorilla.

00:11:33.020 Well, that would be one type of scenario or one type of risk that could arise from future advances in AI,

00:11:40.760 that the AI itself somehow takes over or runs a mark or is poorly aligned.

00:11:47.880 I think there are also scenarios in which we maybe manage to tie it to our purposes,

00:11:53.540 but then we do with it as we have done with practically every other general-purpose technology in human history,

00:12:01.840 that we've also used it for a lot of bad ends to oppress each other, to wage war against each other.

00:12:09.400 And so that's another way in which advances in AI could turn out to be harmful

00:12:15.000 if they become a means of kind of amplifying human conflict,

00:12:18.520 or if they empower more people to develop other dangerous technologies,

00:12:24.820 like maybe you could use AI to more rapidly invent new biological warfare agents

00:12:29.760 or something like that, that might proliferate.

00:12:32.700 So I think there are several distinct classes of dangers that one would have to be aware of

00:12:39.160 as we move into this future.

00:12:42.780 Well, I know, I mean, you think that if you're the creator of it, you can control it, right?

00:12:47.800 You can program it such that it won't get smarter than you, and it won't...

00:12:51.580 How could you... I look at the computer on my desk, how could it ever control me?

00:12:54.600 How is that... It doesn't seem possible that...

00:12:57.460 Because you're not talking about robots running around, you know, threatening us with knives and guns.

00:13:01.340 You're talking about this thing, this thing sitting on the desk,

00:13:04.780 getting smart enough that somehow it's controlling humans.

00:13:07.500 And you think about that in the abstract, and you think, how could that ever...

00:13:10.400 That doesn't make any sense.

00:13:11.420 How could this thing sitting on my desktop ever control me?

00:13:14.080 Yeah, I mean, presumably not the thing that actually sits on the desktop now.

00:13:18.120 But, I mean, you're right.

00:13:19.520 It's easy enough to not develop superintelligence for any one individual or group.

00:13:24.800 But I think it's likely that we as a civilization will nevertheless do it.

00:13:29.940 And I think actually, probably we should be doing it.

00:13:33.780 I see it kind of as this portal in a sense that all plausible paths to a really great future

00:13:40.720 will eventually go through.

00:13:42.660 Now, it might be that it would be wise for us to go a little bit more slowly as we approach this gate,

00:13:49.760 so we don't kind of slam into the wall on the side.

00:13:52.140 We certainly should be very careful with this transition.

00:13:54.880 But I think it's kind of unrealistic to think that everybody, all the different countries,

00:14:00.760 all the different labs would decide to refrain from pushing forward with this,

00:14:05.660 when it has such enormous potential for positive applications in the economy,

00:14:10.720 for medicine, for security, for arts and entertainment,

00:14:14.680 for practically any area at all where human intelligence is useful,

00:14:19.040 which is pretty much any area.

00:14:20.360 So I think our focus should be not so much should we do it or not,

00:14:24.700 but like how can we position ourselves in the best possible ways?

00:14:31.520 Do the research in advance, say on how to align, to find scalable methods for AI alignment,

00:14:38.500 try as much as we can to build cooperative institutions and norms and practices

00:14:43.160 around the deployment of AI, and then proceed cautiously.

00:14:48.160 But can you walk us through that scenario for people who don't,

00:14:52.160 I mean, this is a big concept for folks who don't work in your field.

00:14:55.740 How could it ever be that the machines would take over?

00:14:58.540 I mean, I know you've spoken about, look, it could happen.

00:15:01.580 They could be controlling us.

00:15:02.940 They could control all the other computers and things.

00:15:05.480 Humanity could cease to exist, and we need to be cognizant of this possibility.

00:15:10.240 But how?

00:15:12.340 Well, I mean, so if we look at, say,

00:15:14.800 humans have caused a lot of mischief over the course of history,

00:15:18.620 that it's for the most part not because they use their own personal bodily strengths

00:15:24.960 to wield a sword and go around chopping people's head up.

00:15:28.020 It's they've used maybe their pen or their voice to issue commands,

00:15:31.560 to persuade others, and then thereby to exert great influence.

00:15:35.160 So those modes of action would be available even just to a laptop, right, sitting on a desk.

00:15:43.720 If it could print text on a screen, I think that's already enough

00:15:46.560 for a sufficiently great intelligence to be very powerful.

00:15:51.320 But of course, there is no reason to think it would have to stop with these indirect methods.

00:15:57.400 You could maybe persuade humans to be your arms and legs to do your work in some lab 0.98

00:16:05.280 to develop different robotic systems that you could use or hack into

00:16:12.480 or maybe develop some kind of nanotechnology

00:16:15.400 that would then give you more direct access to the world.

00:16:19.960 I think there are many ways with a sufficient level of intelligence

00:16:24.200 to kind of think above and around and through humans and achieve your ends.

00:16:31.960 It's also likely that if we develop this, we would want to give them access to a lot of stuff

00:16:36.620 because that would make it more useful, right?

00:16:38.860 If you could have an AI that drives your car, that's more useful than an AI

00:16:42.500 that just sits and tells you how to drive the car.

00:16:46.440 If it could run your factories, if it could pilot your airplanes.

00:16:50.020 Maybe we will have a lot of robots by the time this transition happens

00:16:54.860 so that there would be an even more ready-made infrastructure for it to tap into.

00:17:00.080 Well, so let's talk about the factories

00:17:01.680 because I've heard you say that this super intelligent computer

00:17:05.040 or these computers could, quote,

00:17:07.480 create nanofactories covertly distributed at undetectable concentrations

00:17:12.140 in every square meter of the globe

00:17:14.840 that would produce a worldwide flood of human-killing devices on command

00:17:19.760 and that AI would then achieve world domination.

00:17:23.580 What?

00:17:25.280 That doesn't sound good.

00:17:28.200 No.

00:17:29.000 And I think there is a kind of...

00:17:34.500 I mean, it's kind of almost by definition impossible for us

00:17:37.060 to know exactly what the best strategy would be

00:17:39.580 that would come into view

00:17:40.840 if you were a malicious super-intelligence

00:17:43.320 because kind of by definition it can think

00:17:45.400 much more deeply in the strategic space than we can.

00:17:49.840 But I think what that particular scenario is meant to illustrate

00:17:53.140 is the idea that one of the things

00:17:55.500 that a super-intelligence certainly could do

00:17:57.240 would be to invent new technologies

00:17:59.780 that we can already see are physically possible

00:18:03.700 that we haven't yet, however, been able to actually manufacture and build

00:18:08.380 because they involve a lot of detail work

00:18:10.800 to kind of figure out the specifics.

00:18:13.240 But if research were done on a digital timescale

00:18:16.100 rather than on a kind of slow biological human timescale,

00:18:18.740 then these futuristic technologies

00:18:21.340 might become available quite quickly

00:18:23.900 after you have super-intelligence.

00:18:27.100 And then using those futuristic technologies

00:18:30.160 would possibly be one way to leverage its power

00:18:33.020 and get the kind of advantage.

00:18:35.820 It's not the only one, but I think that's one possible path.

00:18:40.080 Whether it would be specifically by developing nanorobots

00:18:43.840 or whether there's some other technologies

00:18:45.520 that we haven't yet thought of,

00:18:47.080 I think we'll get stuff to be agnostic about.

00:18:49.840 It makes me think of the movie War Games

00:18:51.780 to go back to the 1980s when I grew up.

00:18:54.940 And in War Games, they created this computer

00:18:59.160 that could help with nuclear war

00:19:02.960 and planning out the war games

00:19:04.620 that the United States was going to be engaged in

00:19:06.540 with presumably Russia.

00:19:08.760 And they couldn't stop it.

00:19:11.280 The computer sort of got a mind of its own.

00:19:13.720 It started, it was going to launch the missiles anyway,

00:19:16.840 even when they had figured how to like

00:19:18.440 turn off the computer.

00:19:20.560 But when they were trying to deal with it,

00:19:22.200 first one guy says, 0.99

00:19:23.080 why don't you just unplug the damn thing, right? 0.99

00:19:25.380 And that wasn't going to work. 0.99

00:19:26.900 And then even when they found a way

00:19:29.280 of telling it to stand down,

00:19:31.160 it wouldn't stand down.

00:19:32.620 It had a mind of its own and it kept going.

00:19:34.680 Just to bring it to sort of an example

00:19:36.480 that a lot of people may have seen.

00:19:38.220 Is that basically what we're talking about?

00:19:39.900 That once, and you've said before,

00:19:41.980 it may not be possible

00:19:42.920 to put the genie back in the bottle.

00:19:44.460 Once we create the thing,

00:19:45.880 it's not going to be so simple to just unplug it

00:19:48.740 or tell it not to do the thing

00:19:50.500 that we find awful.

00:19:51.640 Yeah, I think like,

00:19:53.940 I mean, so you might not know if like

00:19:55.000 the apes that we evolved from were still around

00:19:59.280 and they thought,

00:19:59.980 well, maybe these humans were a bad idea.

00:20:02.000 Like it would kind of be hard for them

00:20:03.380 to unwind what had happened.

00:20:07.160 And similarly,

00:20:07.820 if you have some super intelligence,

00:20:09.660 once in existence,

00:20:10.480 it might have strategic incentives

00:20:12.120 to avoid us shutting it down.

00:20:16.880 And so if it's very skillful

00:20:18.440 at achieving its goals,

00:20:19.680 then in particular,

00:20:20.860 it would be skillful

00:20:22.200 at achieving this goal

00:20:23.640 of preventing its own shutdown.

00:20:25.920 Or maybe it will make surreptitious copies

00:20:27.760 in other computer systems

00:20:28.960 so that it doesn't matter

00:20:29.840 if sort of the original is terminated

00:20:32.600 or spawn other sub-agents

00:20:35.380 that can do,

00:20:37.140 you know,

00:20:37.680 execute on its preferences.

00:20:38.880 So I think we shouldn't rely on that

00:20:42.140 as these salt methods

00:20:43.340 of ensuring a safe future

00:20:45.440 that we build systems.

00:20:46.720 We don't bother to align them

00:20:48.000 just to see what they do

00:20:49.360 and then planning,

00:20:50.500 well,

00:20:51.020 if things go wrong,

00:20:51.780 we just unplug them.

00:20:52.760 I think we need a better strategy than that.

00:20:55.540 You know,

00:20:55.760 when I was reading up

00:20:57.000 on some of the things

00:20:58.120 that you've written and said,

00:20:59.680 it made me feel like

00:21:01.240 here in America,

00:21:02.100 we've had,

00:21:02.860 you know,

00:21:03.120 a couple hundred years

00:21:03.880 of feeling pretty well protected

00:21:05.220 from the world

00:21:06.000 thanks to these oceans

00:21:07.040 that surround our country,

00:21:09.080 you know,

00:21:09.260 on the East and the West Coast.

00:21:10.760 And obviously with nuclear weapons,

00:21:12.720 that's less the case,

00:21:14.180 but we've reached a detente

00:21:16.160 with other nuclear powers

00:21:17.900 that we understand

00:21:18.800 it would be mutually assured destruction

00:21:20.200 and we don't launch those

00:21:21.280 for the most part.

00:21:24.240 This threat doesn't recognize

00:21:26.920 oceans,

00:21:27.960 boundaries.

00:21:28.720 This makes everyone accessible.

00:21:30.520 If as you've posited,

00:21:33.080 the supercomputer,

00:21:34.740 the super intelligence,

00:21:35.520 can create drones

00:21:38.620 that come right to your doorstep

00:21:40.140 and drop a bomb

00:21:40.940 or create an office robot 0.90

00:21:42.800 that may be cleaning

00:21:44.260 the carpets at night,

00:21:45.580 but then assassinate the CEO

00:21:47.320 when they turn around. 0.82

00:21:48.200 Like,

00:21:48.560 there's all sorts of nefarious ways

00:21:50.900 in which it could be unleashed

00:21:52.580 on people worlds away.

00:21:56.020 Yeah,

00:21:56.240 I think with respect

00:21:57.860 to super intelligence,

00:21:58.800 like,

00:21:59.080 yeah,

00:21:59.340 I think we're all kind of

00:22:00.740 eggs in the same basket.

00:22:02.020 at least with respect

00:22:04.980 to this class of dangerous

00:22:07.400 that arise from the AI itself

00:22:09.100 doing something

00:22:10.660 that is on the line

00:22:12.480 with its creator's intention.

00:22:15.160 So,

00:22:16.120 yeah,

00:22:16.360 I think we have a common cost there

00:22:17.860 to try to figure out

00:22:19.420 how to align these systems.

00:22:21.700 So,

00:22:22.920 and I mean,

00:22:23.800 I'm reasonably hopeful

00:22:25.300 about that.

00:22:25.880 When I was writing

00:22:26.740 this book of mine,

00:22:28.380 I think it came out

00:22:29.220 in 2014,

00:22:30.500 this was an almost

00:22:32.200 entirely neglected field.

00:22:33.520 It looked like

00:22:34.020 we were moving

00:22:35.200 towards developing

00:22:37.000 the most important

00:22:37.860 technology ever

00:22:38.740 and hardly anybody

00:22:40.020 was thinking about

00:22:40.760 what would happen

00:22:41.420 if we were succeeding

00:22:42.600 in this goal of AI.

00:22:44.700 It's all along been

00:22:45.320 to not just do specific tasks,

00:22:47.340 but to make machines

00:22:48.800 generally smart like humans.

00:22:49.920 But it was like

00:22:51.180 that was such a radical goal

00:22:52.460 that the imagination

00:22:54.340 exhausted itself

00:22:55.260 in just conceiving

00:22:56.020 of this possibility

00:22:56.920 of matching humans

00:22:57.920 that it couldn't take

00:22:59.040 the obvious next step

00:23:00.040 that if we reach that

00:23:00.860 we will have super intelligence

00:23:01.980 and then thinking

00:23:02.680 about the consequences.

00:23:04.260 So,

00:23:04.760 yeah,

00:23:05.060 drawing attention to that

00:23:06.020 was a big part

00:23:07.420 of the reason

00:23:07.860 for writing this book.

00:23:10.360 But since then,

00:23:11.580 there has now sprung up

00:23:12.580 a kind of technical subfield

00:23:13.880 of people doing

00:23:14.920 serious research

00:23:16.060 and actually trying

00:23:16.720 to figure out

00:23:17.240 how to align

00:23:18.640 arbitrarily capable

00:23:22.060 AI systems

00:23:22.860 by harnessing

00:23:24.640 their ability

00:23:25.280 to learn

00:23:25.980 to make them

00:23:28.500 better able

00:23:28.940 to learn

00:23:29.660 and understand

00:23:30.360 what our intentions

00:23:32.700 are when we ask

00:23:33.560 them to do something

00:23:34.240 or what

00:23:34.720 to train them

00:23:35.640 on specific tasks.

00:23:37.540 That's crazy

00:23:37.880 that that was only

00:23:38.460 seven years ago

00:23:39.300 that this wasn't

00:23:40.360 even being discussed

00:23:42.180 that seriously

00:23:43.200 by those academics

00:23:44.960 and so on

00:23:45.440 who are now

00:23:45.960 taking such a hard

00:23:46.540 look at it.

00:23:46.880 Meanwhile,

00:23:47.680 this is probably

00:23:48.940 going to be an industry

00:23:49.600 that employs

00:23:50.240 many of our children,

00:23:51.900 grandchildren,

00:23:52.580 and so on.

00:23:53.620 Yeah,

00:23:53.860 I mean,

00:23:54.140 I think,

00:23:54.840 I mean,

00:23:56.560 certainly if there are

00:23:57.660 advances in AI,

00:23:58.440 it's going to have

00:23:58.860 a big economic impact.

00:23:59.920 I mean,

00:24:00.060 it might be that

00:24:00.700 if you get super intelligence,

00:24:01.920 then the effects

00:24:03.000 on employment

00:24:04.400 will be,

00:24:06.060 I mean,

00:24:06.340 at some point,

00:24:07.120 like if you have

00:24:07.740 sufficiently generally

00:24:08.980 capable AI,

00:24:10.800 basically all jobs

00:24:11.780 become automatable.

00:24:12.860 so I think

00:24:16.740 in a good scenario,

00:24:19.200 I mean,

00:24:19.480 in some sense,

00:24:20.060 the goal

00:24:20.440 is full-on employment,

00:24:21.820 right?

00:24:22.120 So the idea

00:24:23.080 is to try

00:24:23.480 to develop

00:24:24.320 technologies

00:24:24.980 so powerful

00:24:27.740 that we don't

00:24:29.640 have to do

00:24:30.380 stuff we don't

00:24:31.480 like to do.

00:24:33.440 And so if you

00:24:34.460 define work

00:24:35.100 as the kind

00:24:35.680 of things

00:24:35.980 people have

00:24:36.940 to pay you

00:24:37.400 to do,

00:24:37.680 then,

00:24:38.700 yeah,

00:24:40.400 almost all of that

00:24:41.320 could theoretically

00:24:42.400 be done

00:24:42.820 by a sufficiently

00:24:43.560 capable AI system.

00:24:45.460 It sounds totally

00:24:46.300 unfulfilling.

00:24:47.020 It sounds awful.

00:24:48.980 Well,

00:24:49.660 I think it would

00:24:50.780 be a situation

00:24:51.380 where we would

00:24:51.980 have to rethink

00:24:53.060 a lot of

00:24:53.660 our assumptions

00:24:55.100 about what it

00:24:55.960 means to be human

00:24:56.640 kind of from

00:24:57.160 the ground up.

00:25:00.100 I actually believe

00:25:01.300 that there would

00:25:01.800 be some extremely

00:25:02.440 wonderful possibilities

00:25:03.700 that would be

00:25:04.980 unlocked

00:25:05.380 by this.

00:25:07.160 but it

00:25:09.620 would require

00:25:10.360 a pretty,

00:25:11.140 yeah,

00:25:12.760 grounds-up

00:25:13.180 rethink.

00:25:15.000 We would,

00:25:15.840 for example,

00:25:16.300 have to find

00:25:18.380 our dignity

00:25:18.960 and meaning

00:25:19.920 in life,

00:25:20.640 not in what

00:25:22.560 we do for a

00:25:23.520 living or

00:25:23.980 like being a

00:25:24.660 breadwinner,

00:25:26.000 but in

00:25:28.440 other areas,

00:25:29.660 in relationships,

00:25:30.960 in hobbies,

00:25:31.740 in things we

00:25:33.340 do for their

00:25:33.780 own sake,

00:25:34.300 rather than as

00:25:34.880 a means to

00:25:35.440 some other

00:25:37.140 end.

00:25:39.360 But,

00:25:40.220 yeah,

00:25:40.560 I mean,

00:25:40.860 that I think

00:25:41.340 would be a

00:25:41.820 kind of high

00:25:42.540 quality problem

00:25:43.260 to have for us.

00:25:44.680 I think first

00:25:45.200 we need to

00:25:45.560 make sure

00:25:45.880 we don't

00:25:48.180 kind of

00:25:48.940 crash into

00:25:49.640 something on

00:25:50.180 the way there.

00:25:51.520 Well,

00:25:52.060 and before we

00:25:52.540 get to sort

00:25:52.940 of the

00:25:53.640 benefits,

00:25:54.840 let's talk

00:25:56.200 about the

00:25:56.620 possibility of

00:25:57.700 a terrorist

00:25:58.380 getting a

00:25:59.740 hold of

00:26:00.500 this technology

00:26:01.220 if we create

00:26:02.440 it or it

00:26:03.020 creates itself

00:26:03.880 from something

00:26:04.800 we've created

00:26:05.520 or even

00:26:06.900 an actor

00:26:08.080 like China, 0.69

00:26:08.880 which is very

00:26:09.460 advanced in

00:26:10.060 the AI field

00:26:10.860 and our

00:26:11.580 defense secretary

00:26:12.320 has made clear

00:26:13.000 that this is

00:26:14.100 an area in

00:26:14.620 which we're

00:26:15.700 equal.

00:26:16.320 We're not,

00:26:16.920 at best,

00:26:17.660 we're equal

00:26:18.060 with China.

00:26:19.300 It's not like

00:26:19.920 our military

00:26:20.500 is so much

00:26:21.340 more powerful

00:26:21.800 than theirs.

00:26:22.400 It is,

00:26:22.940 but I'm just

00:26:23.300 saying in this

00:26:23.720 department,

00:26:24.420 which is a

00:26:24.800 potential security

00:26:25.800 threat,

00:26:26.940 they're on par

00:26:27.860 with us and

00:26:29.020 they're working

00:26:29.580 it and they aim

00:26:30.160 to be the world

00:26:30.680 leader in AI.

00:26:31.740 And we don't

00:26:32.760 trust China 0.83

00:26:33.320 for good

00:26:34.000 reason.

00:26:34.380 So we do

00:26:35.480 need to be

00:26:35.880 worried about

00:26:36.300 what they're

00:26:36.700 going to

00:26:36.900 create,

00:26:37.980 not to

00:26:38.780 mention,

00:26:39.100 as I say,

00:26:39.520 somebody more

00:26:40.520 nefarious like

00:26:41.200 a terrorist

00:26:41.600 actor.

00:26:42.140 So what is

00:26:43.020 the likelihood

00:26:43.440 of that?

00:26:44.660 I think at

00:26:45.620 present,

00:26:47.460 the West

00:26:47.840 is ahead

00:26:48.940 in AI,

00:26:50.380 certainly in

00:26:51.040 this kind

00:26:51.540 of basic

00:26:52.120 research of

00:26:53.720 trying to

00:26:54.160 develop general

00:26:56.000 artificial

00:26:57.000 intelligence.

00:26:58.680 But it's

00:26:59.200 not a huge

00:27:00.080 lead.

00:27:00.300 It's not

00:27:00.580 20 years

00:27:01.760 ahead or

00:27:02.120 something like

00:27:02.560 that.

00:27:02.760 the field

00:27:03.680 is very

00:27:04.240 open.

00:27:05.780 Researchers

00:27:06.120 publish their

00:27:07.120 findings and

00:27:08.240 so other

00:27:08.780 teams can

00:27:09.360 catch up

00:27:09.880 within six

00:27:10.860 months or

00:27:11.320 a year or

00:27:12.320 so.

00:27:14.640 I'm not so

00:27:15.560 worried really

00:27:16.260 about terrorists

00:27:17.880 using AI

00:27:19.740 for particular

00:27:21.060 things.

00:27:21.560 I would be

00:27:22.060 more worried

00:27:22.520 about terrorists

00:27:23.080 using, say,

00:27:24.360 biological

00:27:24.900 weapons,

00:27:26.180 which at

00:27:28.280 the moment

00:27:28.660 would be a

00:27:29.900 lot more

00:27:30.260 destructive

00:27:30.660 and are

00:27:32.400 also becoming

00:27:33.100 much easier

00:27:34.180 to use

00:27:35.020 or obtain

00:27:35.500 through

00:27:36.200 advances in

00:27:36.880 synthetic

00:27:37.220 validity.

00:27:40.520 But it is

00:27:41.680 plausible that

00:27:42.620 AI will

00:27:43.140 become one

00:27:44.140 dimension of

00:27:45.260 a great

00:27:46.520 power of

00:27:46.940 competition

00:27:47.400 as it

00:27:50.220 becomes an

00:27:50.820 increasingly

00:27:51.260 important both

00:27:52.100 economic

00:27:52.700 factor and

00:27:54.460 also

00:27:54.880 factor in

00:27:57.680 national

00:27:58.120 security.

00:28:00.500 Because I

00:28:00.700 know you've

00:28:00.980 said that

00:28:01.960 the first

00:28:03.340 super

00:28:04.300 intelligence to

00:28:04.940 be creative

00:28:05.460 will have a

00:28:06.260 decisive

00:28:06.760 first mover

00:28:08.220 advantage,

00:28:08.680 that there

00:28:09.200 will be a

00:28:09.600 lot of

00:28:09.940 power in

00:28:10.880 being the

00:28:11.180 first one

00:28:11.600 to come

00:28:11.920 up with

00:28:12.220 it.

00:28:12.680 And so,

00:28:13.540 I mean,

00:28:13.780 how worried

00:28:14.280 should we be

00:28:14.820 that somebody

00:28:15.360 not all that

00:28:16.140 friendly to

00:28:16.560 the United

00:28:16.840 States will be

00:28:17.500 the person

00:28:17.880 who has it?

00:28:18.780 Yeah,

00:28:19.780 well,

00:28:20.020 I mean,

00:28:20.280 it's possible

00:28:20.980 that it

00:28:21.340 would have

00:28:21.680 this

00:28:21.980 decisive

00:28:22.600 first mover

00:28:23.200 advantage.

00:28:23.860 I'm not at

00:28:25.300 all sure

00:28:25.700 about that.

00:28:26.240 You could

00:28:26.540 also imagine

00:28:27.180 scenarios in

00:28:27.880 which the

00:28:28.900 transition

00:28:29.460 happens a

00:28:30.100 bit more

00:28:30.460 gradually.

00:28:31.800 If it's not

00:28:32.820 like an

00:28:33.180 overnight or

00:28:33.840 overweek

00:28:34.280 thing where

00:28:34.720 you get

00:28:35.000 from human

00:28:35.460 to radical

00:28:35.980 super

00:28:36.380 intelligence,

00:28:36.940 but suppose

00:28:37.460 it takes

00:28:37.900 several years,

00:28:38.780 then you

00:28:39.260 could easily

00:28:39.680 have multiple

00:28:40.940 labs or

00:28:43.180 countries

00:28:44.040 being more

00:28:46.260 or less

00:28:46.620 going through

00:28:47.700 this

00:28:48.080 transition

00:28:49.840 in tandem

00:28:50.340 and you

00:28:51.480 might then

00:28:51.780 have a

00:28:52.040 multipolar

00:28:52.500 outcome.

00:28:54.620 But,

00:28:55.200 yeah,

00:28:55.400 I do think

00:28:55.860 it's potential

00:28:56.620 for

00:28:57.780 exacerbating

00:28:59.860 conflicts

00:29:00.460 of different

00:29:01.080 kinds

00:29:01.480 or empowering

00:29:03.020 say

00:29:03.560 despots

00:29:05.200 to,

00:29:05.740 you know,

00:29:06.800 make themselves

00:29:07.400 more immune

00:29:08.060 from overthrow

00:29:08.760 by intelligence

00:29:10.320 applications,

00:29:11.360 surveillance

00:29:11.600 applications,

00:29:12.440 and so forth.

00:29:12.980 That is certainly

00:29:13.660 one concern.

00:29:16.540 I think

00:29:17.580 it will

00:29:18.580 be important

00:29:20.340 to try

00:29:20.700 to manage

00:29:21.180 that both

00:29:21.820 to kind

00:29:22.680 of avoid

00:29:23.740 conflicts

00:29:24.640 on that,

00:29:25.020 but also

00:29:25.340 because I

00:29:25.840 think it

00:29:26.120 might make

00:29:26.600 the first

00:29:28.300 danger

00:29:29.000 harder to

00:29:30.600 avoid,

00:29:31.200 the danger

00:29:32.000 coming from

00:29:32.480 the AI

00:29:32.820 itself.

00:29:33.840 Like,

00:29:34.040 if you're

00:29:34.280 thinking about

00:29:34.840 this,

00:29:35.220 suppose you

00:29:36.820 are like

00:29:37.320 some researchers,

00:29:38.540 you've like

00:29:38.920 got to the

00:29:39.860 point where

00:29:40.440 you have

00:29:40.820 something almost

00:29:41.460 human level,

00:29:42.060 you think

00:29:42.360 with a bit

00:29:42.760 more work,

00:29:43.340 we can make

00:29:43.820 it super

00:29:44.200 intelligent.

00:29:44.560 And ideally

00:29:46.020 at this

00:29:46.480 point,

00:29:46.740 you would

00:29:47.060 really want

00:29:47.820 to go

00:29:48.080 slow,

00:29:48.540 right?

00:29:48.780 And really

00:29:49.440 check everything,

00:29:50.740 double check it,

00:29:51.720 make sure it's

00:29:52.340 all right,

00:29:53.560 increment it

00:29:54.280 step by step,

00:29:55.680 not just

00:29:56.140 turning on the

00:29:56.820 gas full throttle.

00:29:58.560 And maybe

00:29:59.160 over several

00:30:00.040 years,

00:30:01.220 like trying to

00:30:01.820 do this while

00:30:02.760 having a lot

00:30:03.300 of people

00:30:03.680 helping you

00:30:04.260 and looking

00:30:04.640 over what

00:30:05.620 you're doing

00:30:05.980 to make

00:30:06.320 sure it's

00:30:06.640 right.

00:30:07.000 But if you

00:30:07.460 are in

00:30:07.700 some kind

00:30:08.100 of arms

00:30:08.940 race,

00:30:09.560 then that

00:30:11.180 might be

00:30:11.500 very hard

00:30:11.900 to do.

00:30:12.160 It might

00:30:12.780 basically mean

00:30:13.820 that if

00:30:14.140 you go

00:30:14.400 slow,

00:30:14.780 it just

00:30:15.080 means you

00:30:15.440 lose the

00:30:15.800 race and 0.82

00:30:16.140 become

00:30:16.380 irrelevant.

00:30:16.960 So you

00:30:17.360 feel forced

00:30:18.400 to rush

00:30:18.840 ahead as

00:30:20.140 quickly as

00:30:20.540 you can,

00:30:20.940 and then

00:30:21.300 you throw

00:30:21.780 caution to

00:30:22.320 the wind,

00:30:22.780 and then

00:30:23.120 this risk

00:30:24.180 from the

00:30:24.600 AI itself

00:30:25.200 creating

00:30:26.760 a destruction

00:30:27.920 will increase.

00:30:29.520 So the

00:30:29.880 two problems

00:30:30.400 are connected.

00:30:31.840 Yeah,

00:30:31.940 it's kind

00:30:32.300 of like a

00:30:32.580 Dr.

00:30:32.860 Frankenstein

00:30:33.420 situation.

00:30:34.580 It's a

00:30:34.720 Frankenstein

00:30:35.120 situation,

00:30:35.680 right,

00:30:35.820 where the

00:30:36.180 entity you

00:30:36.720 create becomes

00:30:37.900 super dangerous

00:30:38.800 and turns on

00:30:39.460 you,

00:30:40.220 even though

00:30:40.820 we're so

00:30:41.580 full of

00:30:41.920 hubris,

00:30:42.440 I think

00:30:42.700 most humans

00:30:44.080 would believe

00:30:44.740 that they

00:30:45.060 could continue

00:30:45.600 controlling a

00:30:46.880 machine.

00:30:48.140 Again,

00:30:48.720 it's hard,

00:30:49.260 I think,

00:30:50.200 to conceptualize

00:30:51.300 that the

00:30:51.640 machine I

00:30:52.280 control now

00:30:53.040 will someday

00:30:53.700 have the

00:30:54.200 capability of

00:30:55.080 controlling me,

00:30:56.220 but I know

00:30:57.020 you've said

00:30:57.400 right now

00:30:58.000 the potential

00:30:58.720 for this

00:30:59.140 super intelligence,

00:31:00.300 right now

00:31:00.840 it's lying

00:31:01.320 dormant,

00:31:02.260 but it's

00:31:02.900 akin to the

00:31:03.620 power of

00:31:04.160 the atom

00:31:04.640 and how

00:31:05.440 it laid

00:31:06.380 dormant

00:31:06.880 through much

00:31:07.500 of our

00:31:07.820 human history

00:31:08.620 until 1945,

00:31:10.120 in which

00:31:11.200 case it

00:31:11.580 was very

00:31:12.320 much not

00:31:12.880 dormant and

00:31:14.280 we saw

00:31:14.780 its power

00:31:15.580 in really

00:31:16.740 raw and

00:31:17.420 disturbing

00:31:17.820 ways.

00:31:19.740 Yeah,

00:31:20.040 I think

00:31:20.420 in general

00:31:21.100 we are

00:31:24.260 kind of

00:31:24.660 reaching into

00:31:25.960 this giant

00:31:26.480 urn of

00:31:26.960 invention.

00:31:27.780 This is

00:31:28.200 almost like

00:31:28.960 the picture

00:31:29.360 of human

00:31:29.740 history.

00:31:30.300 We reached

00:31:31.340 in,

00:31:31.640 we pulled

00:31:31.960 out one

00:31:32.360 ball after

00:31:32.880 another,

00:31:33.240 one idea,

00:31:33.980 one technology,

00:31:34.620 and I

00:31:35.820 think we've

00:31:36.160 kind of

00:31:36.380 been lucky

00:31:36.860 so far

00:31:37.360 in that

00:31:37.700 for the

00:31:38.080 most part

00:31:38.640 the net

00:31:39.640 effect of

00:31:40.220 all this

00:31:40.640 technological

00:31:41.580 progress has

00:31:42.300 been hugely

00:31:42.760 positive.

00:31:44.760 But if

00:31:46.120 there is a

00:31:46.520 black ball 0.99

00:31:47.080 in this

00:31:47.440 urn,

00:31:47.860 some

00:31:48.360 technology,

00:31:49.300 there could

00:31:49.600 be some

00:31:49.920 technology

00:31:50.360 that is

00:31:50.860 just such

00:31:51.700 that it

00:31:52.420 invariably

00:31:53.440 discovers

00:31:54.040 the

00:31:54.400 civilization

00:31:55.000 that

00:31:55.480 invents

00:31:56.920 it.

00:31:57.840 It looks

00:31:58.540 like we're

00:31:58.920 just going

00:31:59.220 to keep

00:31:59.660 reaching

00:32:00.080 into this

00:32:00.540 urn

00:32:00.860 until we

00:32:02.380 get the

00:32:02.640 black ball

00:32:03.040 if it's 0.86

00:32:03.360 in there.

00:32:03.860 And while

00:32:04.620 we have

00:32:04.940 developed a

00:32:05.800 great ability

00:32:06.360 to pull 0.59

00:32:07.440 balls out 0.98

00:32:08.240 of the urn,

00:32:08.680 we don't

00:32:09.100 have an

00:32:09.460 ability to

00:32:09.940 put them

00:32:10.240 back in.

00:32:10.720 We can't

00:32:11.220 uninvent

00:32:11.680 our

00:32:11.920 inventions.

00:32:13.300 So it

00:32:14.780 looks like

00:32:15.260 our strategy

00:32:15.880 such as

00:32:16.360 this is

00:32:16.900 basically

00:32:17.480 just to

00:32:17.880 hope that

00:32:18.420 there is 0.98

00:32:18.940 no black 1.00

00:32:19.360 ball in

00:32:19.720 the urn.

00:32:21.720 And I

00:32:22.160 think not

00:32:22.520 just AI

00:32:23.000 but some

00:32:24.120 other

00:32:24.560 technologies

00:32:25.340 as well

00:32:25.760 could be

00:32:26.440 potential 0.88

00:32:27.480 black 1.00

00:32:28.800 balls. 1.00

00:32:29.360 I alluded

00:32:30.140 to

00:32:30.520 synthetic

00:32:31.140 biology

00:32:31.580 before.

00:32:33.360 which is

00:32:33.980 one area

00:32:34.460 where we

00:32:35.500 might discover

00:32:36.300 means that

00:32:36.940 would make

00:32:37.280 it a lot

00:32:37.600 easier to

00:32:38.220 create

00:32:38.560 highly

00:32:39.940 enhanced

00:32:40.440 pathogens.

00:32:42.020 In some

00:32:42.440 sense we

00:32:42.740 were lucky

00:32:43.180 with nuclear

00:32:43.660 weapons.

00:32:44.120 They are

00:32:44.260 enormously

00:32:44.640 destructive

00:32:45.200 but at

00:32:45.760 least they

00:32:46.220 are hard

00:32:46.500 to make.

00:32:47.680 You need

00:32:48.220 highly enriched

00:32:49.060 uranium or

00:32:49.600 plutonium to

00:32:50.340 be able to

00:32:50.800 make a

00:32:51.120 nuclear bomb. 0.96

00:32:52.480 And that

00:32:52.860 requires large

00:32:53.900 facilities, huge

00:32:54.740 amounts of

00:32:55.180 energies, really

00:32:55.940 only states

00:32:56.620 can do

00:32:57.040 this.

00:32:57.960 But suppose

00:32:59.360 it had

00:32:59.720 turned out

00:33:00.200 that there

00:33:00.560 were an

00:33:00.800 easier way

00:33:01.280 to do

00:33:01.580 this, to

00:33:02.060 unleash

00:33:02.400 the

00:33:03.140 energy

00:33:03.500 of the

00:33:03.760 atom.

00:33:04.680 Before we

00:33:05.300 actually did

00:33:05.780 the relevant

00:33:06.200 nuclear

00:33:06.580 physics, how

00:33:07.260 could we

00:33:07.580 have known

00:33:07.880 how it

00:33:08.200 would turn

00:33:08.460 out?

00:33:08.800 But if

00:33:09.520 there had

00:33:09.720 been some

00:33:10.040 easy way,

00:33:10.660 like something

00:33:11.600 baking

00:33:12.280 sand in the

00:33:12.920 microwave

00:33:13.260 oven between

00:33:13.860 two nickel

00:33:15.460 plates or

00:33:15.960 something like

00:33:16.440 that, then

00:33:17.100 that might

00:33:17.920 have been the

00:33:18.340 end of

00:33:18.620 human

00:33:18.820 civilization

00:33:19.260 once we

00:33:20.420 discovered how

00:33:20.980 to do

00:33:21.220 that.

00:33:21.500 Because then

00:33:21.880 anybody

00:33:22.360 would be able

00:33:24.300 to destroy

00:33:24.860 a city.

00:33:26.480 And in a

00:33:27.300 sufficiently

00:33:27.660 large

00:33:28.080 population,

00:33:28.800 there's

00:33:28.940 always going

00:33:29.320 to be a

00:33:29.680 few

00:33:29.960 individuals who

00:33:31.300 would choose

00:33:31.920 to do

00:33:32.160 that.

00:33:32.400 Whether

00:33:33.040 because they

00:33:33.540 are mad or

00:33:33.980 they have

00:33:34.240 some grudge

00:33:34.780 or they

00:33:35.100 have some

00:33:35.440 extortion

00:33:35.840 scheme or

00:33:36.520 some

00:33:37.080 ideology.

00:33:39.040 So we

00:33:39.820 can't really

00:33:40.240 afford a

00:33:41.200 kind of

00:33:41.620 democratization

00:33:42.940 of the

00:33:44.140 ability to

00:33:45.020 cause mass

00:33:46.380 destruction.

00:33:47.640 But if we

00:33:48.420 discover some

00:33:48.940 easily, you

00:33:49.820 know,

00:33:50.680 implementable

00:33:51.280 recipe for

00:33:51.900 this, then

00:33:53.260 it looks like

00:33:54.080 we are in a

00:33:54.580 pretty dire

00:33:55.480 situation.

00:33:56.060 up next,

00:33:58.400 how important

00:33:59.020 is it going

00:33:59.400 to be to

00:33:59.820 have

00:34:00.120 ethicists

00:34:01.060 involved in

00:34:02.460 creating these

00:34:03.180 super intelligent

00:34:04.060 beings, right?

00:34:05.040 It's not just

00:34:05.540 the people who

00:34:06.000 can make the

00:34:06.420 machines function,

00:34:07.480 it's those who

00:34:08.020 can lay in

00:34:08.500 some sort of

00:34:08.920 an ethical

00:34:09.340 code, and

00:34:09.860 is that even

00:34:10.280 possible?

00:34:11.120 And then we're

00:34:11.500 going to get

00:34:11.740 into the

00:34:13.040 future of

00:34:13.880 humanity and

00:34:15.060 how it can

00:34:15.520 be helped

00:34:15.960 by technology.

00:34:17.320 This is

00:34:17.700 something that

00:34:18.140 Nick has

00:34:18.400 studied for a

00:34:18.960 long time.

00:34:19.760 Could there

00:34:20.080 be something

00:34:20.540 like an

00:34:20.940 anti-aging

00:34:21.520 pill coming

00:34:22.660 our way?

00:34:23.060 And how

00:34:24.000 far away

00:34:24.960 is that?

00:34:26.000 And also

00:34:26.740 cryogenics.

00:34:27.760 Is he going

00:34:28.260 to freeze

00:34:28.520 himself?

00:34:29.020 And why?

00:34:29.720 And should

00:34:30.040 you?

00:34:30.640 Stay tuned.

00:34:34.640 I know

00:34:35.300 you've said

00:34:36.040 you could

00:34:36.760 use this

00:34:37.460 technology or

00:34:38.440 some bad

00:34:40.040 actor could or

00:34:40.660 the super

00:34:41.100 intelligent

00:34:41.420 computer could

00:34:42.080 for ethnic

00:34:42.620 cleansing.

00:34:43.160 I mean,

00:34:43.280 imagine if

00:34:43.740 Hitler had 0.99

00:34:44.160 control over

00:34:45.260 this type of

00:34:45.820 technology where

00:34:46.780 he could target

00:34:47.860 some particular

00:34:49.400 groups this

00:34:50.980 way.

00:34:51.300 It would be

00:34:51.820 efficient.

00:34:52.460 It could be

00:34:52.780 a killing

00:34:53.080 machine and

00:34:53.940 this is why

00:34:55.040 we need

00:34:55.380 ethical people

00:34:56.040 creating the

00:34:56.780 technology if

00:34:57.540 it gets

00:34:57.780 created at

00:34:58.300 all.

00:34:59.200 And that

00:34:59.520 leads me to

00:35:00.380 something I

00:35:00.740 heard recently

00:35:01.220 and I know

00:35:01.520 you've been

00:35:01.880 saying all

00:35:02.620 along,

00:35:03.020 which is not

00:35:04.000 only are we

00:35:04.500 going to need

00:35:04.900 if we're

00:35:05.400 going down

00:35:05.860 this route

00:35:06.280 and we're

00:35:06.540 going to

00:35:06.700 create

00:35:06.980 super

00:35:07.240 intelligence

00:35:07.600 or something

00:35:08.280 less than

00:35:09.140 super

00:35:09.400 intelligent

00:35:09.760 computers

00:35:10.220 for the

00:35:11.500 good that

00:35:11.780 they can

00:35:12.020 do,

00:35:12.780 one of the

00:35:13.280 most important

00:35:13.780 roles we're

00:35:14.280 going to have

00:35:14.740 in this

00:35:14.960 process will

00:35:15.560 be ethicists,

00:35:17.460 philosophers.

00:35:18.600 It's not all

00:35:18.980 about kids

00:35:20.000 now who are

00:35:20.520 in robotics or

00:35:21.540 kids who are

00:35:21.980 somehow trying to

00:35:22.780 study AI.

00:35:23.800 We're going to

00:35:24.580 need people who

00:35:25.240 consider and

00:35:26.700 can even

00:35:27.200 program for

00:35:28.600 the ethics

00:35:29.300 of a well

00:35:32.040 meaning life,

00:35:32.980 of a well

00:35:33.520 meaning existence.

00:35:35.280 Yes, I mean

00:35:36.020 not necessarily

00:35:36.840 people whose

00:35:38.180 jump title are

00:35:39.060 ethicists at

00:35:39.800 some university,

00:35:40.660 but yeah,

00:35:41.200 certainly ethics

00:35:42.580 and other

00:35:45.140 sources of

00:35:46.320 wisdom about

00:35:47.620 what we want

00:35:48.680 and what we

00:35:49.840 should be

00:35:50.740 wanting I

00:35:51.300 think will be

00:35:51.820 important.

00:35:52.100 it's not

00:35:52.540 just a

00:35:52.920 purely

00:35:53.100 technical

00:35:53.540 problem,

00:35:54.020 it's a

00:35:54.240 kind of

00:35:54.600 all of

00:35:55.340 society

00:35:55.820 problem,

00:35:56.360 how to

00:35:56.740 figure out

00:35:57.480 how to

00:35:57.900 create a

00:35:58.540 happy world

00:35:59.100 with this

00:36:00.160 new technology.

00:36:00.900 I mean it

00:36:01.140 will have

00:36:01.540 economic

00:36:02.040 implications,

00:36:02.880 right?

00:36:03.060 We've

00:36:03.320 alluded to

00:36:03.880 the security

00:36:04.340 implications

00:36:04.900 before.

00:36:06.680 And then

00:36:07.440 more cultural

00:36:09.020 dimensions are

00:36:10.220 like what do

00:36:12.280 we want the

00:36:12.740 role of humans

00:36:13.320 to be versus

00:36:13.980 our technology

00:36:15.020 and automation

00:36:16.280 in the future?

00:36:17.520 I think it's

00:36:19.640 like, yeah,

00:36:20.240 it needs to

00:36:20.740 draw on all

00:36:21.220 the different

00:36:21.780 aspects of

00:36:23.480 human wisdom

00:36:24.400 such as it

00:36:24.980 is, it's

00:36:26.360 not much to

00:36:26.980 boast about,

00:36:27.640 but we'll

00:36:29.280 have to do

00:36:29.640 our best at

00:36:30.300 least.

00:36:30.560 Yeah, I think

00:36:31.400 it will require

00:36:32.020 a much wider

00:36:32.740 purview than

00:36:33.440 just a narrow

00:36:34.260 technical focus,

00:36:35.260 although the

00:36:35.660 technical focus

00:36:36.280 also is really

00:36:37.000 important for

00:36:37.640 AI alignment.

00:36:38.180 I've read

00:36:39.180 that many

00:36:40.540 leading

00:36:41.180 researchers in

00:36:42.060 this field

00:36:42.520 say it's

00:36:43.620 extremely likely

00:36:44.460 this will

00:36:44.820 happen by

00:36:46.040 as early

00:36:46.840 as 2075

00:36:48.140 to 2090.

00:36:49.680 That's in

00:36:50.360 the lifetime

00:36:50.720 of our

00:36:51.260 children right

00:36:52.020 now.

00:36:52.880 Do you agree

00:36:53.540 with that?

00:36:53.940 Could it

00:36:54.140 happen that

00:36:54.520 soon?

00:36:56.340 Yeah, I

00:36:57.080 think that's

00:36:57.740 a real

00:36:58.160 possibility.

00:36:59.340 That's hard

00:37:00.140 to get your

00:37:00.480 arms around,

00:37:00.960 right?

00:37:01.120 Like my own

00:37:02.060 children alive

00:37:02.960 right now could

00:37:03.540 be dealing

00:37:04.000 with computers

00:37:06.440 in the corners

00:37:07.140 of the world

00:37:07.760 that are

00:37:08.080 trying to

00:37:08.620 erase

00:37:09.480 humanity?

00:37:10.620 Or save

00:37:12.400 humanity,

00:37:13.060 right?

00:37:13.380 I mean,

00:37:13.840 that's the

00:37:15.500 other possibility

00:37:16.280 or help us.

00:37:16.980 Well, I'm

00:37:17.120 not worried

00:37:17.460 about that

00:37:17.860 one.

00:37:18.540 That one

00:37:19.140 sounds good.

00:37:20.520 I mean,

00:37:20.980 that would

00:37:24.340 be what

00:37:24.660 people are

00:37:25.100 aiming for 0.92

00:37:25.740 almost

00:37:27.140 universally.

00:37:28.020 Like AI

00:37:29.340 researchers,

00:37:30.140 I suppose,

00:37:31.120 they're

00:37:31.940 usually quite

00:37:34.180 well-meaning

00:37:34.640 and idealistic

00:37:35.300 people.

00:37:35.940 Some are

00:37:36.220 just curious

00:37:36.800 about what

00:37:37.360 they're doing

00:37:37.800 and think

00:37:38.320 it's fun.

00:37:38.920 Some have

00:37:39.320 a kind

00:37:39.640 of general

00:37:40.280 sense of

00:37:41.000 wanting to

00:37:41.900 do something

00:37:42.300 good for

00:37:42.660 the world.

00:37:43.080 So I

00:37:43.300 think the

00:37:44.480 intention

00:37:44.860 is positive

00:37:47.300 and it's

00:37:49.320 just the

00:37:49.660 outcome that

00:37:50.340 there is

00:37:50.740 more uncertainty

00:37:51.260 around.

00:37:52.420 Right,

00:37:53.040 because the

00:37:53.280 dark side

00:37:53.740 is this

00:37:54.920 could be

00:37:55.220 more dangerous

00:37:55.820 than any

00:37:57.240 pandemic,

00:37:58.040 than nukes,

00:37:59.560 than

00:37:59.780 catastrophic

00:38:00.920 climate change.

00:38:02.180 This could be

00:38:02.940 more than an

00:38:04.140 asteroid.

00:38:04.600 this could be

00:38:05.940 the thing

00:38:06.200 we're not

00:38:06.440 paying that

00:38:06.900 much attention

00:38:07.300 to,

00:38:07.580 to your

00:38:07.860 example

00:38:08.300 about what

00:38:08.620 was going

00:38:08.900 on in

00:38:09.140 2014,

00:38:10.400 that really

00:38:11.220 is an

00:38:11.640 existential

00:38:11.960 threat to

00:38:12.700 humanity

00:38:13.480 on the

00:38:14.640 earth.

00:38:16.080 And on

00:38:16.400 that front,

00:38:16.780 one of the

00:38:17.040 things I

00:38:17.340 wondered in

00:38:18.000 reading about

00:38:18.800 you,

00:38:19.040 Nick,

00:38:19.180 is whether

00:38:19.600 do you

00:38:21.140 worry about

00:38:21.600 anything

00:38:22.140 other than

00:38:22.620 this?

00:38:23.040 I mean,

00:38:23.200 if this

00:38:23.680 were my

00:38:24.040 world,

00:38:24.740 that I

00:38:24.960 were immersed

00:38:25.800 in full-time

00:38:26.440 thinking about

00:38:27.000 it,

00:38:27.340 I don't know

00:38:27.760 that I'd

00:38:28.460 worry about

00:38:28.860 anything else.

00:38:30.040 Would I

00:38:30.460 worry about the

00:38:30.980 crime rate?

00:38:31.560 Would I

00:38:31.780 worry about

00:38:32.360 the erosions

00:38:33.540 of free

00:38:33.900 speech?

00:38:34.280 We're seeing

00:38:34.640 government power

00:38:35.580 growing too

00:38:36.100 large.

00:38:36.760 Some of it

00:38:37.520 relates,

00:38:38.020 but I just,

00:38:39.940 do you walk

00:38:40.900 through the

00:38:41.140 day worried

00:38:41.580 about nothing

00:38:42.000 other than

00:38:42.360 this?

00:38:44.900 No,

00:38:45.520 I mean,

00:38:45.880 I think

00:38:46.360 it's useful

00:38:48.580 maybe to have

00:38:49.080 some division

00:38:49.600 of labor

00:38:50.100 here.

00:38:51.160 Also,

00:38:51.780 I think

00:38:52.140 it might,

00:38:53.480 for somebody

00:38:53.820 like me,

00:38:54.680 be worth

00:38:55.240 trying to

00:38:56.040 focus more

00:38:57.140 efforts on

00:38:57.880 relatively

00:38:58.500 neglected

00:38:59.020 areas where

00:39:00.020 one extra

00:39:00.980 person like

00:39:01.620 me might

00:39:01.940 make a

00:39:02.240 bigger

00:39:02.540 difference

00:39:03.800 than,

00:39:04.260 I mean,

00:39:04.440 global warming

00:39:05.040 people,

00:39:06.260 so many

00:39:07.740 thousands of

00:39:08.260 people have

00:39:08.660 been worrying

00:39:09.200 about this,

00:39:09.700 and I think

00:39:10.060 it's a

00:39:10.660 smaller concern.

00:39:11.980 But it's not

00:39:12.420 the only one.

00:39:12.980 I mean,

00:39:13.160 I think

00:39:13.480 certain advances

00:39:14.760 in biotechnology

00:39:15.640 are quite

00:39:16.500 concerning as

00:39:17.160 well.

00:39:19.820 That might

00:39:20.660 just make it

00:39:21.200 too easy

00:39:21.820 to create

00:39:22.760 really dangerous

00:39:23.560 stuff.

00:39:25.280 I mean,

00:39:25.840 to make it

00:39:26.320 concrete,

00:39:26.760 so we have

00:39:27.280 DNA synthesis

00:39:28.400 machines that

00:39:29.200 can print

00:39:29.680 out DNA

00:39:30.980 strings,

00:39:31.760 if you

00:39:34.820 have a

00:39:35.140 digital

00:39:35.400 blueprint.

00:39:36.020 We also

00:39:36.500 have in

00:39:36.840 the public

00:39:37.220 domain,

00:39:38.020 the DNA

00:39:40.200 sequence of

00:39:41.520 a lot of

00:39:42.260 really dangerous

00:39:43.520 viruses,

00:39:44.440 and ideas

00:39:45.040 how to make

00:39:45.840 them even

00:39:46.160 more dangerous.

00:39:47.220 So as

00:39:48.580 this sequencing

00:39:49.320 technology becomes

00:39:50.260 good enough to

00:39:51.160 print out

00:39:51.920 whole viral

00:39:53.100 genomes,

00:39:53.780 then,

00:39:54.600 I mean,

00:39:55.420 you just

00:39:57.020 connect the

00:39:57.460 dots,

00:39:57.700 that would

00:39:58.400 give anybody

00:39:58.860 with access

00:39:59.360 to these

00:39:59.820 DNA synthesis

00:40:01.020 technology the

00:40:02.340 ability to

00:40:02.980 create things

00:40:03.580 far worse

00:40:04.400 than COVID.

00:40:06.300 And at the

00:40:06.840 moment,

00:40:07.160 anybody can

00:40:07.820 buy one of

00:40:08.300 these DNA

00:40:08.740 synthesis

00:40:09.120 machines,

00:40:09.680 and if that

00:40:10.060 continues to

00:40:10.640 be the

00:40:10.900 model as

00:40:11.500 they improve

00:40:12.080 in capacity,

00:40:12.920 then we

00:40:14.380 soon get to

00:40:16.240 an unsustainable

00:40:17.280 situation.

00:40:17.920 So there needs

00:40:18.320 to be some

00:40:18.700 kind of

00:40:19.080 global

00:40:19.460 regulatory

00:40:20.520 framework,

00:40:22.420 I think,

00:40:23.140 imposed on the

00:40:23.960 DNA synthesis

00:40:24.660 market where

00:40:26.240 DNA synthesis

00:40:27.180 is provided

00:40:27.840 as a service.

00:40:28.640 Like if you're

00:40:29.020 a researcher,

00:40:29.600 if you need

00:40:30.000 a particular

00:40:30.800 string,

00:40:31.280 there's maybe

00:40:31.740 five or six

00:40:33.380 companies in

00:40:33.980 the world,

00:40:34.500 you'd send

00:40:34.860 a request,

00:40:35.440 you'd get

00:40:35.760 your product

00:40:36.360 back in a

00:40:36.860 vial,

00:40:37.120 you don't

00:40:37.360 need to

00:40:37.660 have the

00:40:37.940 machine

00:40:38.200 yourself

00:40:38.600 in a

00:40:39.260 lab.

00:40:39.500 Or if

00:40:39.780 you do,

00:40:40.140 that would

00:40:40.380 have to

00:40:40.660 be some

00:40:40.960 sort of

00:40:41.240 controls.

00:40:42.560 That's just

00:40:43.040 one example,

00:40:43.780 but it's a

00:40:45.120 real cauldron.

00:40:45.760 People are

00:40:46.100 inventing so

00:40:46.680 many cool

00:40:47.160 new stuff

00:40:47.680 in biotech

00:40:49.320 all the

00:40:49.580 time that

00:40:50.600 new ways of

00:40:52.400 creating mostly

00:40:53.340 good stuff will

00:40:54.040 come into

00:40:54.380 view,

00:40:54.660 but there

00:40:55.080 could also

00:40:55.480 be some

00:40:55.900 bad stuff

00:40:56.800 in there.

00:40:57.560 It's a

00:40:58.060 kind of

00:40:58.320 wild west

00:40:58.800 at the

00:40:59.120 moment.

00:40:59.580 The ethos

00:41:00.340 is very

00:41:00.720 much open

00:41:01.420 science.

00:41:02.280 Let's encourage

00:41:03.140 biohackers,

00:41:04.140 let's just

00:41:04.600 make it

00:41:05.300 available to

00:41:05.920 all,

00:41:06.320 because that's

00:41:06.900 nice.

00:41:07.540 Everybody has

00:41:08.260 equal access

00:41:08.760 to it.

00:41:09.360 With some

00:41:10.040 technologies,

00:41:10.780 like with

00:41:11.260 nuclear weapons,

00:41:12.160 we don't

00:41:12.640 think that's

00:41:13.080 the right

00:41:13.400 way.

00:41:13.780 I think

00:41:14.080 biotechnology

00:41:14.780 will have

00:41:15.220 to similarly

00:41:15.760 change to

00:41:18.080 something.

00:41:18.460 I think

00:41:23.700 I can

00:41:23.900 speak on

00:41:24.220 behalf of

00:41:24.480 my audience

00:41:24.800 and say

00:41:25.040 that we're

00:41:25.520 glad that

00:41:25.860 people like

00:41:26.200 you, I

00:41:26.440 know Stephen

00:41:26.780 Hawking has

00:41:27.240 joined you,

00:41:27.680 many others

00:41:28.280 have sketched

00:41:29.900 out some

00:41:30.700 priorities for

00:41:31.520 making this

00:41:32.700 field safer

00:41:33.660 and imposing

00:41:34.740 some guidelines

00:41:35.660 so that we

00:41:36.280 don't just

00:41:36.680 jump into

00:41:37.180 this willy-nilly.

00:41:38.800 Glad you're

00:41:39.440 out there.

00:41:39.880 It reminds me

00:41:40.180 of my husband

00:41:40.660 used to run

00:41:41.380 an internet

00:41:41.800 security firm

00:41:42.800 and he used

00:41:43.680 to speak of

00:41:44.060 the white

00:41:44.440 hat hackers

00:41:45.500 and the

00:41:46.360 black hat

00:41:47.000 hackers. 0.88

00:41:48.200 And, you

00:41:48.700 know, the

00:41:48.940 Russians have 1.00

00:41:49.460 black hat 0.58

00:41:50.120 hackers who

00:41:50.720 will try to

00:41:51.140 get into

00:41:51.460 your bank

00:41:51.780 accounts and

00:41:52.380 your private

00:41:53.000 information and

00:41:53.640 so on.

00:41:54.600 And he

00:41:55.680 was in a

00:41:56.220 firm that

00:41:57.320 he would

00:41:58.720 say he

00:41:59.080 employed the

00:41:59.580 white hat

00:42:00.140 hackers who

00:42:01.000 understood how

00:42:01.520 to do all

00:42:01.860 that stuff but

00:42:02.460 would try to

00:42:02.840 stay one step

00:42:03.380 ahead of the

00:42:03.740 bad guys to

00:42:04.860 protect you.

00:42:05.420 And it seems

00:42:05.680 like this is a

00:42:06.180 field where

00:42:06.720 everyone needs

00:42:07.640 to have the

00:42:08.060 same devious

00:42:09.520 skills but we

00:42:10.720 have to make

00:42:11.240 sure we have

00:42:11.720 more people

00:42:12.200 employing them

00:42:12.740 for good than

00:42:13.260 for bad.

00:42:13.820 Let me switch

00:42:15.460 it to you

00:42:15.880 with you for

00:42:16.700 one second

00:42:17.200 because I

00:42:17.820 do know

00:42:18.240 that you

00:42:19.420 is it

00:42:20.200 transhumanism

00:42:21.100 that you've

00:42:21.460 described yourself

00:42:22.000 as into

00:42:22.400 being transhumanism

00:42:23.820 I think

00:42:24.640 like your

00:42:25.700 proselytization

00:42:27.260 about the

00:42:28.520 medical field

00:42:29.160 and what could

00:42:29.660 happen for

00:42:30.240 humans during

00:42:31.020 our lifetime

00:42:31.560 to make life

00:42:32.300 better to

00:42:32.840 make to

00:42:33.260 improve life

00:42:33.900 when it comes

00:42:34.260 to these

00:42:34.540 scientific

00:42:34.900 advances made

00:42:36.160 me feel

00:42:36.500 hopeful you

00:42:37.320 know potential

00:42:37.880 anti-aging

00:42:38.680 pill a

00:42:39.080 potential the

00:42:40.120 potential to

00:42:40.740 bring one

00:42:41.100 back thanks to

00:42:42.120 cryogenics after

00:42:43.460 one dies can

00:42:44.320 you just talk

00:42:45.300 about that a

00:42:45.800 bit

00:42:46.000 yeah so I

00:42:48.480 I mean I

00:42:49.060 sometimes I

00:42:50.200 think I'm

00:42:51.560 mistaken for

00:42:52.320 some kind of

00:42:53.120 technophobe a

00:42:53.980 lot because I

00:42:54.840 spend a lot of

00:42:56.020 time describing

00:42:56.700 some specific

00:42:57.480 concerns or

00:42:58.300 dangers but I

00:42:59.340 mean broadly

00:42:59.720 speaking I'm

00:43:00.600 very excited

00:43:01.480 about the

00:43:02.440 potential to

00:43:03.300 improve the

00:43:05.280 human condition

00:43:05.960 through through

00:43:07.460 advances in

00:43:08.440 technology and

00:43:09.520 including by

00:43:11.520 enhancing the

00:43:12.100 human organism

00:43:12.760 itself in

00:43:13.500 different ways

00:43:13.900 like I

00:43:14.140 mean I'd

00:43:14.420 love to see

00:43:14.880 some kind

00:43:15.360 of anti-aging

00:43:17.020 pill that

00:43:17.480 really worked

00:43:18.100 or you know

00:43:20.300 something that

00:43:20.900 could make us

00:43:21.340 smarter or

00:43:22.100 improve our

00:43:23.200 quality of life

00:43:24.020 in other ways

00:43:24.560 and so

00:43:26.480 sometimes that's

00:43:27.920 referred to as

00:43:28.420 transhumanism I

00:43:29.440 don't tend to

00:43:30.040 use the word

00:43:30.540 very much because

00:43:31.480 a lot of people

00:43:33.940 use it for a lot

00:43:34.680 of different

00:43:35.160 ideas

00:43:35.940 yeah now it's

00:43:36.320 confusing

00:43:36.660 which are kind

00:43:37.240 of kooky

00:43:37.640 yeah and it

00:43:38.860 attracts some

00:43:39.600 some kind of

00:43:41.040 miscellaneous

00:43:41.640 folk that I

00:43:42.520 wouldn't necessarily

00:43:43.160 agree with on

00:43:44.200 all issues so

00:43:44.920 but yeah I

00:43:47.120 think I think

00:43:48.360 there's I think

00:43:50.060 there's a lot of

00:43:50.920 room for

00:43:51.300 improvement

00:43:51.700 do you think

00:43:54.020 academics are with

00:43:55.100 you on on

00:43:56.900 that desire

00:43:57.440 because I I've

00:43:58.340 heard you say

00:43:58.740 something like

00:43:59.260 there are a lot

00:44:00.020 of people in the

00:44:00.420 field who are

00:44:00.800 like why what

00:44:01.460 about overpopulation

00:44:02.360 and why would

00:44:03.140 we want to

00:44:03.520 extend our time

00:44:05.420 here on earth

00:44:06.180 significantly

00:44:07.440 right like

00:44:08.020 what's the

00:44:08.840 boredom of

00:44:09.720 living longer

00:44:10.360 is reason

00:44:10.800 enough not

00:44:11.200 to do that

00:44:11.640 so do you

00:44:12.520 think we have

00:44:12.920 an academic

00:44:13.340 field that is

00:44:14.120 devoted to

00:44:14.620 developing things

00:44:15.300 like that

00:44:15.660 well I think

00:44:17.820 I think this

00:44:18.760 have been

00:44:19.240 shifting so

00:44:20.300 slowly but

00:44:21.160 steadily over

00:44:22.460 the last 20

00:44:24.040 years or so

00:44:24.620 I think like

00:44:25.700 there certainly

00:44:27.660 used to be an

00:44:28.420 extremely strong

00:44:29.280 double standard

00:44:30.500 if we take

00:44:31.200 say something

00:44:31.940 that slows

00:44:32.420 like doing

00:44:32.920 something about

00:44:33.500 aging was a

00:44:34.940 big no-go

00:44:35.460 because people

00:44:36.780 couldn't see

00:44:37.280 why you would

00:44:37.760 possibly want

00:44:38.420 to do that

00:44:38.800 but then at

00:44:39.220 the same time

00:44:39.800 there were

00:44:40.140 billions of

00:44:40.680 fundings to

00:44:41.180 try to fix

00:44:41.900 cancer to

00:44:42.800 fix heart

00:44:43.280 disease to

00:44:44.120 fix diabetes

00:44:44.940 like to

00:44:47.060 fix wrinkles

00:44:47.780 like all

00:44:49.180 these things

00:44:49.620 right but

00:44:50.020 then like the

00:44:50.400 sum total of

00:44:51.140 that is like

00:44:51.700 aging like

00:44:52.240 aging is what

00:44:52.880 makes you more

00:44:53.420 vulnerable to

00:44:54.080 cancer to

00:44:54.660 heart disease to

00:44:55.420 diabetes to

00:44:56.240 wrinkles and

00:44:57.420 so for exactly

00:44:58.140 the same reasons

00:44:58.880 that you might

00:44:59.480 not like these

00:45:00.740 diseases and

00:45:02.180 symptoms for the

00:45:03.580 same reason you

00:45:04.100 might not like the

00:45:04.800 underlying thing

00:45:05.340 that is creating

00:45:06.100 a huge fraction

00:45:07.560 of all of that

00:45:08.200 so it's not

00:45:09.140 as if there

00:45:09.560 was some weird

00:45:10.500 special reason

00:45:11.260 you would have

00:45:12.080 to have for

00:45:13.220 favoring life

00:45:14.200 extension or

00:45:14.840 anti-aging it's

00:45:15.600 just exactly the

00:45:16.500 same reasons that

00:45:17.140 they play in all

00:45:17.680 its other cases

00:45:18.360 but for some

00:45:19.000 reason that was

00:45:19.480 this mental block

00:45:20.340 that a lot of

00:45:21.180 people had that

00:45:21.800 they I think

00:45:23.580 actually it was

00:45:24.240 because the

00:45:25.340 anti-aging thing

00:45:26.220 seemed unrealistic

00:45:27.740 and panciful so

00:45:28.820 it was not

00:45:29.360 evaluated by the

00:45:30.580 same standards

00:45:31.220 that like a pill

00:45:32.840 that is likely

00:45:33.540 better for treating

00:45:34.320 some cancer

00:45:34.860 would be

00:45:35.120 evaluated it

00:45:35.840 was more

00:45:36.280 evaluated by

00:45:37.100 reference to

00:45:37.720 some traditions

00:45:39.980 or like some

00:45:42.620 kind of spiritual

00:45:43.260 wisdom where

00:45:43.940 you're like you're

00:45:44.620 supposed to accept

00:45:45.700 the fact that 0.96

00:45:46.880 you're gonna die 0.97

00:45:47.580 and that's a sign 0.98

00:45:48.260 of wisdom and

00:45:49.020 you should come to

00:45:49.600 terms with it

00:45:50.220 rather than fight

00:45:50.860 against it

00:45:51.400 so it was kind

00:45:54.500 of the issue

00:45:55.160 was placed in

00:45:55.980 that mental

00:45:56.760 bucket rather

00:45:57.420 than this is

00:45:58.560 something we can

00:45:59.100 actually work on

00:45:59.800 yeah but you

00:46:00.780 raise a good

00:46:01.120 point if that's

00:46:02.000 your mindset

00:46:02.800 then why get

00:46:04.520 chemotherapy

00:46:05.020 why not eat 0.74

00:46:06.840 as many trans 0.99

00:46:07.920 fats as you

00:46:08.500 want right

00:46:08.920 like we all

00:46:09.460 take steps to

00:46:10.280 prolong our

00:46:10.980 lives even when

00:46:12.280 we get bad

00:46:12.700 diagnoses

00:46:13.220 even though we

00:46:14.580 may be people

00:46:15.080 of faith and

00:46:15.560 understand how

00:46:16.200 this is going

00:46:16.600 to end

00:46:16.860 eventually

00:46:17.280 yeah I mean

00:46:18.620 to take care

00:46:19.500 of your body

00:46:20.540 seems good

00:46:20.960 I mean it's

00:46:21.360 not to say

00:46:21.740 that in every

00:46:22.360 and all

00:46:22.700 circumstances

00:46:23.320 extending life

00:46:25.260 further I mean

00:46:25.880 if you're kind

00:46:26.600 of have a life

00:46:27.960 that's not worth

00:46:28.600 living you're in

00:46:29.260 some respirator

00:46:30.060 and you just

00:46:30.700 kind of kept

00:46:31.700 alive under

00:46:33.400 horrible conditions

00:46:34.220 like that I'm

00:46:34.760 not I'm not sure

00:46:35.360 it's a great

00:46:35.780 boon if that can

00:46:36.620 be like two

00:46:37.580 years of that

00:46:38.220 rather than one

00:46:38.780 year but but

00:46:39.600 if you have a

00:46:40.760 good quality of

00:46:41.520 life if you are

00:46:42.200 healthy you enjoy

00:46:42.860 life maybe you

00:46:43.680 can contribute to

00:46:44.560 society in some

00:46:45.240 way then that

00:46:45.740 seems something

00:46:47.660 that is definitely

00:46:48.380 worth preserving

00:46:49.180 or just if we can

00:46:49.860 improve our senior

00:46:50.560 years you know I

00:46:51.340 mean we've all

00:46:51.760 watched our parents

00:46:52.500 or grandparents

00:46:53.020 deteriorate mentally

00:46:54.180 some some would

00:46:55.380 argue we're

00:46:55.740 watching it right

00:46:56.240 now with our

00:46:56.600 president okay I

00:46:57.460 that was an

00:46:57.860 aside

00:46:58.160 but we've all

00:47:00.220 watched that and

00:47:00.860 thought to

00:47:01.160 ourselves oh we

00:47:02.380 you know I don't

00:47:02.920 want that for

00:47:03.480 them and if you

00:47:04.500 could create a

00:47:05.460 situation where we

00:47:06.220 could live to our

00:47:06.840 80s or 90s forget

00:47:08.080 beyond beyond would

00:47:08.900 be delightful but

00:47:09.660 even that old but

00:47:11.600 sharp with mental

00:47:12.940 acuity why wouldn't

00:47:14.620 we want that it'd be

00:47:15.440 great but can I

00:47:16.740 ask you um yeah

00:47:18.360 yeah go ahead

00:47:18.800 yeah and I'm like

00:47:20.140 if I think if and

00:47:21.560 whether it actually

00:47:22.620 is possible like if

00:47:23.620 there is actually a

00:47:24.300 pill on the market at

00:47:25.180 some point that does

00:47:25.940 this then I think a

00:47:26.880 lot of the people

00:47:27.440 who were previously

00:47:28.880 expressing skepticism

00:47:30.600 would kind of

00:47:31.200 quickly come around

00:47:32.080 how far away from

00:47:34.140 that are we

00:47:34.920 well I mean

00:47:39.200 from from curing

00:47:41.680 aging altogether I

00:47:42.660 mean I think it's

00:47:44.300 it seems quite hard

00:47:45.440 to do that I mean

00:47:46.820 maybe super

00:47:47.520 intelligence would

00:47:48.160 expedite that like

00:47:49.560 I think actually a

00:47:50.560 lot of these things

00:47:51.100 would happen soon

00:47:51.740 after super

00:47:52.320 intelligence but other

00:47:53.460 than that I mean

00:47:54.480 we've been kind of

00:47:55.420 on the verge of

00:47:56.060 curing cancer now

00:47:56.960 for the last 50

00:47:57.780 years right so we've

00:47:59.340 made small incremental

00:48:00.240 progress so the

00:48:02.080 super intelligence is

00:48:03.080 going to make us live

00:48:04.000 forever and then kill 1.00

00:48:04.860 us all off something 0.99

00:48:05.740 to look forward to

00:48:06.340 one or the other

00:48:07.360 yeah one of those

00:48:08.700 right they may have

00:48:10.540 buyers regret can we

00:48:11.760 talk about cryogenics

00:48:12.600 for a minute because

00:48:13.360 this is something you

00:48:14.260 know that I think most

00:48:14.920 of us grew up thinking

00:48:15.920 as I'm told now wrongly

00:48:18.180 Walt Disney had himself

00:48:19.680 frozen so he could be

00:48:20.700 brought back to life

00:48:21.500 but cryogenics is a

00:48:23.740 real thing and I

00:48:25.100 know you're in favor

00:48:25.940 of it can you talk

00:48:26.640 about it for a minute

00:48:27.240 well so it's the idea

00:48:28.980 that we know that

00:48:31.240 that's sufficiently

00:48:31.840 cold temperatures

00:48:32.860 basically all

00:48:33.960 physiological processes

00:48:35.740 stop so I mean you

00:48:38.640 know things last

00:48:39.300 longer if you put them

00:48:40.180 in your freezer right

00:48:41.100 but if you put them in

00:48:44.300 even colder temperatures

00:48:45.460 like in liquid nitrogen

00:48:47.200 then they can last for

00:48:50.300 hundreds of years

00:48:51.320 with basically not

00:48:52.460 change so so the idea

00:48:53.780 is if somebody dies

00:48:54.680 today if you freeze

00:48:57.460 them then you preserve

00:49:00.460 whatever is there and

00:49:02.160 of course if you thaw

00:49:02.960 them up they are still

00:49:03.920 dead because their

00:49:05.260 tissue well a that was

00:49:06.780 the original thing that

00:49:07.780 killed them and be the 0.99

00:49:08.760 actual freezing process 0.85

00:49:09.780 creates additional

00:49:10.660 damage but but if you

00:49:12.000 think that technology

00:49:12.720 will continue to improve

00:49:13.780 then maybe at some

00:49:14.680 point in the future

00:49:15.420 the technology will

00:49:16.760 exist to reverse

00:49:18.080 whatever originally

00:49:19.680 caused their death and

00:49:21.060 to cure frostbite that

00:49:23.160 the kind of damage that

00:49:24.080 happens during freezing

00:49:24.960 so if you if you just

00:49:26.660 preserve somebody long

00:49:27.880 enough there then there

00:49:28.780 is some hope that the

00:49:29.600 technology will one day

00:49:30.560 exist to to bring them

00:49:31.780 back to life unless you're

00:49:33.540 sure that that technology

00:49:34.520 will never be developed

00:49:36.280 then it seems like the

00:49:37.460 conservative thing to do

00:49:38.680 would be to to put them

00:49:40.140 in liquid nitrogen and uh you

00:49:43.260 know if if it doesn't work

00:49:45.220 well they'd be dead

00:49:46.000 anyway so uh the downside

00:49:48.200 is you're hedging your

00:49:48.820 bets yeah well would you

00:49:51.060 want to come back i mean

00:49:52.040 one of the fears i would

00:49:53.340 have is what if they wake

00:49:54.940 me up in the year 4000

00:49:56.140 and it's terrifying it's

00:49:58.960 like a caveman walking

00:50:00.100 into 2021 and nothing is

00:50:02.940 familiar no one he loves

00:50:05.300 is around anymore it's it

00:50:07.980 seems like a nightmare in

00:50:09.800 some ways yeah i mean

00:50:11.540 that's obviously very hard

00:50:13.700 to know what the future

00:50:14.600 would be like that that

00:50:16.280 you would be brought back

00:50:17.360 into um and so i think

00:50:19.760 yeah that probably mostly

00:50:22.600 comes down to to whether

00:50:24.160 some somebody's like more

00:50:25.720 of an optimistic person or

00:50:27.020 a pessimistic person i guess

00:50:29.380 that's true and and what

00:50:30.800 about what what is the

00:50:31.940 likelihood that the people

00:50:32.880 in the future who may have

00:50:34.120 these prolonged lifespans

00:50:35.580 are going to want to come

00:50:37.040 back and get people like

00:50:38.040 you and me well forget me

00:50:39.360 but you i mean you they

00:50:40.420 should want but seriously

00:50:42.200 what what is the given

00:50:43.400 overpopulation concerns and

00:50:44.980 the limited size of the

00:50:46.020 earth why would they want

00:50:47.560 to be bring people back

00:50:48.600 well i mean it would be

00:50:50.840 very cheap for them given

00:50:52.160 the resources in the future

00:50:53.820 this would be like a drop

00:50:55.480 in the pocket so if they

00:50:56.980 have sentimental reasons or

00:50:59.580 ethical reasons for doing

00:51:01.640 this i mean i i think

00:51:02.980 certainly if if there were

00:51:05.120 somebody who lived a

00:51:06.980 thousand years ago and i

00:51:08.360 could you know with the

00:51:09.660 snap of the finger or you

00:51:11.160 know bring them back i think

00:51:12.740 that that would be a very

00:51:13.760 nice thing to do and so

00:51:15.200 yeah like what if it were

00:51:16.200 he blinked you know what if

00:51:17.360 it were this historical

00:51:18.160 figure who would be right

00:51:19.300 i mean or even if it were

00:51:20.460 just some bumpkin right i

00:51:21.680 mean like you could save a

00:51:24.320 life even if it were a life

00:51:25.840 that started a long time ago

00:51:27.240 and had been sort of in

00:51:28.280 suspension that that would

00:51:30.120 be it would be nice do you

00:51:32.480 think do you have reason to

00:51:33.600 believe that in that

00:51:34.620 scenario the brain would

00:51:36.220 still have the information

00:51:37.280 you know upon revival that

00:51:40.460 it had on the way out that

00:51:42.420 you know the brain retains

00:51:43.400 information i i think so i

00:51:46.200 mean it depends obviously

00:51:48.420 somewhat on how you die so

00:51:49.680 if you die in a fire or

00:51:51.380 you're lost at sea right then

00:51:52.760 it's kind of gone but in a

00:51:55.480 reasonably good scenario i

00:51:57.580 think it's likely that uh the

00:51:59.820 information is basically

00:52:00.780 preserved it's quite hard to

00:52:03.760 destroy information so i think

00:52:05.300 if you have a book right okay

00:52:06.660 and you tear up all the 0.56

00:52:08.020 pages in small pieces so now

00:52:10.420 you can't read the book

00:52:11.240 anymore but the information

00:52:12.340 is still there like in theory

00:52:13.780 you can put the pieces

00:52:15.000 together in game if you have

00:52:16.460 enough patience and i think

00:52:19.180 it's similar with the brain

00:52:20.120 like the freezing damage will

00:52:21.880 kind of their ice crystals

00:52:23.880 forming actually they are trying

00:52:25.560 to use various anti-freeze

00:52:28.680 agents and they are not

00:52:29.860 literally freezing it they are

00:52:31.140 vitrifying it setting aside

00:52:33.260 these technical details i think

00:52:34.500 there is like some kind of

00:52:35.540 shoving around of different

00:52:37.200 pieces that happen in this

00:52:38.740 process but the information is

00:52:40.420 still there most likely and

00:52:42.700 with sufficiently advanced

00:52:44.340 technology it should be able to

00:52:45.680 to to to put it together i think

00:52:48.300 so but but i i my guess is if it

00:52:51.020 happens at all it will happen

00:52:52.320 after super intelligence as a

00:52:54.580 consequence of super

00:52:55.420 intelligence wow um the the last

00:53:02.580 sort of question i have for you

00:53:03.600 is a practical one which is

00:53:05.020 given that we may be looking at

00:53:08.500 total unemployment if there is

00:53:10.340 super intelligence total

00:53:11.440 unemployment what do you see as a

00:53:14.000 an important area for kids today

00:53:16.420 for young people today kids in

00:53:17.640 college to be looking at or even

00:53:20.220 younger you know i have i have a

00:53:21.980 almost 12 10 and eight year old

00:53:24.140 it used to be it would definitely

00:53:27.280 robotics and we talk about that a

00:53:28.700 lot in schools today but like

00:53:30.280 where where do we steer our kids if

00:53:32.640 they want to stay on the cutting

00:53:34.100 edge of technology and future jobs

00:53:36.820 and you know i know that one of the

00:53:38.340 fields that that artificial

00:53:40.900 intelligence may take over is

00:53:42.360 actually doing pretty pretty good job

00:53:43.740 of right now is radiology they can

00:53:45.600 read x-rays in certain settings so i

00:53:48.480 don't know maybe you don't want your

00:53:49.380 kid to be a radiologist but what what

00:53:50.820 is your thinking about sort of the

00:53:51.880 wave of the future and the likely

00:53:53.320 good industries to be in

00:53:54.840 yeah i mean so so for for kids i mean

00:53:58.360 i think it depends a lot on the kid

00:53:59.660 like you want to build on their unique

00:54:02.300 strengths and not everybody have to be a

00:54:04.680 computer programmer right that's a

00:54:06.300 small part of the economy um i i think

00:54:11.540 that i mean already today we are at an

00:54:16.020 unprecedented time of of wealth and

00:54:19.800 prosperity relative to any other time

00:54:22.300 in all of human history and so in

00:54:25.260 addition to having a focus on you know

00:54:29.780 finding a profitable career for your

00:54:32.360 your kid i think also equipping them to

00:54:34.920 actually enjoy life and to find do

00:54:37.700 something meaningful in their life would

00:54:40.600 be worthwhile because if not now then

00:54:42.260 like when i mean i get i guess after

00:54:44.380 the singularity maybe but um um so i think i

00:54:51.640 think that would be important like in

00:54:53.000 in terms of areas i mean i i think

00:54:54.840 computer stuff will continue to be

00:54:56.720 important but so will many many other

00:54:58.380 uh areas in the economy um i think

00:55:03.080 maybe inculculating certain habits like a

00:55:09.240 habit of continual learning like a

00:55:11.880 flexibility to to to be able and willing

00:55:14.300 sometimes and feel empowered to kind of

00:55:16.600 there's a field you don't know anything

00:55:17.800 about some skill you don't yet have well i

00:55:19.940 could try to learn it like there's so much

00:55:21.780 information online or training courses or

00:55:24.060 it's easier than ever to like get access to

00:55:26.940 new areas so giving them that sense of kind

00:55:29.440 of personal agency that i can take

00:55:31.120 responsible for what i want to do and i

00:55:33.400 can figure out how to learn it or if i

00:55:35.560 can't figure that out i can figure out who to

00:55:37.660 ask i think i think that would be useful

00:55:40.260 across a very wide range of different

00:55:41.940 scenarios they're gonna have to stay

00:55:44.480 nimble in the world that's coming their

00:55:46.320 way that's for sure thank you so much for

00:55:48.620 your expertise it's been an absolute

00:55:49.880 pleasure oh no i enjoyed it all right up

00:55:53.300 next andrew ang this guy was super high

00:55:56.380 up uh both at google and at china's google

00:55:59.540 it's called baidu he was their chief

00:56:01.020 scientist and has led huge teams when it

00:56:04.100 comes to developing the ai of both groups

00:56:06.340 what does he think about all of this and

00:56:09.060 is your computer already spying on you is

00:56:12.520 your government spying on you this is a

00:56:14.800 guy who's been at some very well-known

00:56:16.820 big uh leading corporations when it comes

00:56:20.000 to artificial intelligence and data

00:56:21.800 uh amassing what does he have to say

00:56:24.100 you're gonna love this guy he's next

00:56:25.580 before we get to him however want to bring

00:56:27.960 you a feature we have here on the mk show

00:56:29.460 called from the archives uh this is where

00:56:32.400 we bring you a bit of audio from our

00:56:33.780 growing library of content now nearing 150

00:56:36.420 episodes hard to believe today we're going

00:56:38.780 to go back to our 69th episode and one of

00:56:41.220 our most popular with tulsi gabbard a

00:56:44.000 veteran former congresswoman who shared

00:56:46.200 with us some stories of her time in the

00:56:47.520 military in washington dc and in the media

00:56:50.880 ringer here's just a bit on that on the

00:56:53.360 way she was covered during her 2020 0.63

00:56:55.800 presidential run against now president

00:56:57.800 biden and vp kamala harris take a listen

00:57:00.560 the ones who are writing about you of course

00:57:02.840 the mainstream press is left wing we're

00:57:04.960 writing bad things and the ones who

00:57:07.660 control the airwaves weren't giving you

00:57:09.780 any airtime that's exactly right and and

00:57:12.960 that's where that's where you know the the

00:57:16.580 evidence of this kind of facade of a

00:57:19.340 democracy comes to the forefront because

00:57:21.620 you really have these corporate media

00:57:24.080 interests who are are uh who most care

00:57:27.200 about ratings and entertainment and how

00:57:31.100 they can create conflict um you know on a

00:57:34.060 debate stage or uh push push a narrative

00:57:36.760 that they think will get more eyeballs to

00:57:39.920 their to their screens and i put social

00:57:42.820 media in this category as well uh combined

00:57:46.220 with a a party that uh pre-selecting who

00:57:52.560 they wanted voters to hear from and so

00:57:56.240 that's where you saw a lot of hey you know

00:57:57.840 they're changing the standards for the debates

00:57:59.660 as they go along

00:58:00.840 um you know just as you know hey okay we're

00:58:03.820 ticking up a little bit in the polls where we

00:58:05.580 think we're going to qualify for another

00:58:06.900 debate

00:58:07.300 oh sorry rules changed you know the day

00:58:10.680 before or right right when uh you know those

00:58:13.620 new polls were coming out and and just other

00:58:16.020 things you know the democratic uh the dnc

00:58:18.360 saying hey you know all presidential candidates

00:58:20.880 if you want to be featured in any of our our um

00:58:23.460 publicity that we're putting out then you got

00:58:26.480 to fork up i think it was something like

00:58:27.940 175 000 to the dnc just to be included in

00:58:32.760 their you know social media videos or whatever

00:58:35.360 and i'm just like no i'm not gonna do that

00:58:38.040 you know i got you know people across the

00:58:40.440 country who are giving you know five bucks

00:58:42.520 ten bucks contributing to my campaign because

00:58:45.340 they believe in the kind of leadership that

00:58:47.160 i'll bring in the message and the truth that

00:58:48.800 i'm i'm sharing with voters and and they're

00:58:50.820 certainly not giving me a whole bunch of

00:58:54.160 money to go and then pass it on to uh to the

00:58:57.200 dnc and and so ultimately that's where we saw

00:59:00.740 time and time again even even small thing it's

00:59:03.560 not that small but things that that went

00:59:06.240 unnoticed for example you know cnn had a

00:59:09.420 bunch of town halls where they featured

00:59:10.820 different candidates um they they only gave me

00:59:15.500 one most of the other candidates had more

00:59:17.880 than one and someone called me one day and

00:59:20.800 said hey you know i'm going through my um

00:59:22.760 cnn it's not dvr but if you go to cnn's i

00:59:27.320 guess digital library they had uh you could

00:59:30.840 replay the town halls of all the different

00:59:32.640 candidates they're like you're not on here

00:59:34.720 like it's just not it doesn't exist there's no

00:59:38.640 option to find your town hall but i can find

00:59:40.920 every single other democrat who ran for president

00:59:43.540 and so there there were there were things like

00:59:46.520 that and more more forward blatant things that

00:59:49.580 made it very clear that if the media makes a

00:59:52.020 decision not to allow voters to hear from you

00:59:56.760 then um a voters really don't have the ability to

01:00:02.120 make an informed decision in a true democracy and and

01:00:06.680 then b the reality is that if you want to if you want to

01:00:09.420 talk about issues if you want to get information to

01:00:11.360 people so they can make this informed decision then

01:00:13.340 clearly running for office is not the way to do

01:00:15.720 it gabbard has just struck a new deal with rumble

01:00:18.720 the video social network youtube competitor so i think

01:00:21.860 we're we're about to hear a lot more from her in the

01:00:24.420 weeks to come and good and we in the meantime will

01:00:27.580 keep bringing you more of our best episodes from the

01:00:31.020 archives up next andrew ang you'll love

01:00:35.000 thank you for being here i'm i'm excited for this

01:00:42.420 conversation we just wrapped up with nick bostrom uh who he

01:00:46.880 wasn't totally anti-ai right he he's pro-ai uh but has

01:00:51.900 some concerns about i i think what you call artificial

01:00:54.900 general intelligence agi the the the long-term game where

01:00:59.620 you develop a machine that develops super intelligence

01:01:03.040 intelligence so let's just start there what's your take on

01:01:05.820 the likelihood that we will develop super intelligent

01:01:09.700 machines in this century nick bostrom is an interesting

01:01:13.900 character um ai is the new electricity is transforming tons

01:01:17.340 of industries revolutionized the way we do things in the

01:01:20.020 united states and around the world uh as for artificial

01:01:22.840 general intelligence i think we'll get there but whether it

01:01:25.980 will take 50 or 500 or 2 000 years to make computers as

01:01:29.900 intelligent as you know you or me or other people i think

01:01:33.080 that's a really long-term open research project

01:01:35.720 it's exciting okay i like 2 000 2 000 makes me feel better than

01:01:39.840 by the end of this century when my kids are still

01:01:41.900 god willing alive you know i think that uh one of the

01:01:45.600 problems with the whole field of ai is um

01:01:48.540 is is is it's confusing in this way uh there's a one type of

01:01:52.360 ai called agi artificial general intelligence things to do

01:01:55.440 anything a human could do maybe someday

01:01:57.120 and artificial narrow intelligence which is the ai that does one thing

01:02:00.900 really really well and it's really valuable

01:02:02.740 turns out over the last you know 10 20 years we've had tons of progress in

01:02:06.600 artificial narrow intelligence those ai that do one thing really really 0.87

01:02:10.080 well so people say accurately there's tons

01:02:12.980 of progress in ai i agree with that but just because there's tons of

01:02:16.240 progress in ai doesn't mean from where i'm sitting i'm candidly not

01:02:19.360 seeing that much progress toward artificial general intelligence

01:02:22.900 uh so i think that's led to some of the unnecessary hype and fear-mongering

01:02:27.600 candidly about ai that makes me feel better i'm feeling better already

01:02:31.820 now you know a thing or two about narrow artificial narrow intelligence just so

01:02:36.180 the audience understands you've led teams at google and is it

01:02:39.180 is it pronounced baidu forgive me for not knowing

01:02:41.160 oh yes i i started led the google brain team also ran ai for baidu which

01:02:46.060 which is a large web search engine company in china

01:02:48.760 because china doesn't use google so this is china's google

01:02:51.680 uh china's leading web search engine was this is baidu and then i'm also

01:02:56.180 really proud of the work that i did leading the google brain team which is

01:02:59.360 a team that hopes a lot of google embrace modern ai so

01:03:02.300 if you use google you know you're probably using technology that that that my

01:03:06.220 former team wrote actually almost that's amazing

01:03:08.300 so now what are what are some of the fun things that you and your team have

01:03:11.560 introduced into my life that i don't even know i should be thanking you for

01:03:14.220 uh don't don't don't thank me thank the many millions well thousands of

01:03:18.880 people around the world building these technologies um i think that all of us use

01:03:23.380 ai dozens of times a day maybe even more perhaps without even knowing it uh thanks

01:03:28.460 to modern ai when you do an internet search you get much more relevant

01:03:31.780 results or every time you check your email there's a spam filter in there

01:03:35.560 kind of saving us from massive amounts of spam uh that's ai every time you use a

01:03:39.960 credit card it's probably an ai trying to figure out if it is you

01:03:43.120 or if you know someone stole the credit card and we should not let that transaction

01:03:47.180 through so all of us probably use ai albums many many dozens of times a day

01:03:51.700 maybe without even knowing it and what about the self-driving car

01:03:56.080 is it you know that makes the news every so often and it's interesting to me i

01:04:00.060 it's scary to me because you also hear some reports of crashes and you

01:04:02.900 understand that okay that the technology is not exactly where they want it to be

01:04:05.880 yet but what do you see when it comes to self-driving cars i think that many

01:04:11.240 people including me uh collectively um uh underestimated how difficult it will

01:04:17.220 be to get to you know true fully autonomous self-driving cars that could drive

01:04:21.700 the way that that the person can um i think we will get there but it's been a

01:04:26.720 longer road than any of us estimated when i drive these cars i'm happy for the

01:04:30.780 driver assistance technology i personally don't really fully trust them yet so i

01:04:35.060 keep an eye on the road you know when i'm driving and one of these technologies

01:04:37.780 is supposedly doing something well yeah so here's a dumb question i

01:04:41.960 understand why somebody if we perfect the technology somebody like my mom who's 80

01:04:45.720 and really not all that well physically mentally she's great but

01:04:49.300 physically um i could see why a self-driving car would work well for her 0.99

01:04:53.240 it's like you're a built-in chauffeur but why do young able-bodied people need 0.79

01:04:58.160 that why is it an improvement for people our age

01:05:00.440 um i think that it depends a lot on the individual

01:05:04.460 i sometimes find it fun to drive you know if i don't know take my daughter out on

01:05:09.620 the road drive around that's fun but sometimes if i'm driving to work in

01:05:13.160 traffic it's like boy i wish someone else could do the driving for me and if a

01:05:17.720 computer could do that so i could maybe even sit in the back seat

01:05:20.960 and you know play with my daughter i would rather do that than be stuck in

01:05:24.160 traffic so i think it depends a lot on the

01:05:26.100 individual it's funny because i asked this having

01:05:29.180 just yesterday i had to go to the city i'm in

01:05:31.680 new jersey for the summer i had to go to the city

01:05:33.360 it's a couple of hours and i had the choice of driving myself

01:05:37.080 or sometimes we use a driver and i said you know i'm going to use a driver

01:05:41.180 because i had a bunch of interviews to do today and i said i want to read all my

01:05:44.920 stuff and i and so it's a dumb question right it's basically

01:05:47.680 you can read everything if you have a self-driving car it's going to make your 0.51

01:05:50.640 life really convenient if it doesn't kill you or all the

01:05:53.580 people around you

01:05:54.300 and and i and again i i think i i think i know you have kids right my my kids are

01:05:59.900 really young part of me worries you know when they grow up

01:06:02.540 will they ever get in the car accident so with when my daughter grows up

01:06:06.520 uh if you know there's a computer that can drive her safer than if i were to 0.99

01:06:10.260 drive or she would drive herself i think it'll make all of us better off

01:06:13.600 you know how far away are we from that you know the ai world we keep on we've 0.74

01:06:19.400 made a lot of predictions and and sometimes we're not very good at

01:06:22.460 predicting the timeline uh by which on which this will happen

01:06:26.760 i think that self-driving cars are kind of getting there in limited

01:06:29.860 environments uh so i'm seeing exciting progress for example if you're driving

01:06:34.560 around the constrained environment of a port you know shipping stuff or in a

01:06:38.980 mine or sometimes on the farm that's actually kind of getting there

01:06:42.100 um if we're willing to rejigger some of the cities uh i think we'll be there

01:06:46.340 pretty soon um i don't know i think i think it'll

01:06:49.200 still be quite a few years i still be many many many years before

01:06:52.520 you know we can drive in downtown new york or downtown new jersey

01:06:56.280 yeah i understand that they're they're not as good at picking up

01:07:01.200 things like the hand signals that a construction worker might be issuing to

01:07:06.120 you that they don't totally understand those

01:07:08.460 things yet so they're they're not quite where they need to be

01:07:12.060 um okay so so let's talk about other ways in which

01:07:15.860 ai is going to be helping our lives and how you see it because

01:07:19.800 one of the things that nick said that concerned me was we're probably headed toward

01:07:23.660 total unemployment total eventually in the distant future once the

01:07:27.500 machines become as smart as we think they they're likely to get and that 0.62

01:07:31.400 concerns me you know i don't know what life looks like if nobody works for a

01:07:34.380 living and if the machines are in control of

01:07:36.420 everything so what is the journey from here to there look like

01:07:39.020 in terms of technological advances you know i think that um uh total

01:07:45.120 unemployment i i'm actually skeptical it'll

01:07:47.660 ever happen or if it does happen and maybe i don't know how many thousands of

01:07:51.460 years away um you know it turns out it's just just let's let's demystify ai

01:07:55.880 what can we make the idea what can we not it turns out um to get a little bit

01:07:59.600 geeky and technical almost all of ai today is about

01:08:02.460 input output mappings such as uh input an email

01:08:05.860 you know output is it spam or not or input a picture of what's in front of

01:08:10.120 your car and output the position of the other cars

01:08:12.700 um or uh input you know an x-ray image and

01:08:16.560 output a diagnosis does this person have pneumonia or not or some of the

01:08:19.920 condition so that's sort of the one idea

01:08:22.780 input output uh that is creating 99 percent of the value of the economic

01:08:27.680 value of today's ai system turns out this is a ton of economic value

01:08:31.480 uh the large ad platforms have an ai that inputs

01:08:34.680 an ad and some information of other user and outputs are you going to click on

01:08:38.700 this ad or not uh because you know if you can get people to click on more ads

01:08:42.300 this direct impact on the bottom line of the large ad platforms

01:08:45.020 so it's creating tons of economic value but frankly this input output thing if we

01:08:50.700 think about how much more people do uh uh this is so much more people could do

01:08:55.720 i don't think anyone in the world uh uh has a realistic roadmap for getting the

01:09:00.380 agi so i think sometimes the that that concept

01:09:03.320 has been has been over heightened fear mongered

01:09:06.060 um i do worry about unemployment with every wave of technology looking back

01:09:10.400 you know industrial revolution uh invention of electricity

01:09:13.180 i mean well all the people working on steam engines they unfortunately really

01:09:17.500 sadly lost their jobs or we used to have uh human operator

01:09:21.660 elevators right you know there was someone standing in the elevator

01:09:24.760 that would dial it up and down when someone invented automatic elevators those 0.90

01:09:28.820 jobs went away so i worry about that for ai that

01:09:31.940 creates some amounts of um disruption and and effect work

01:09:36.300 but complete total unemployment this this input output mapping i don't see that

01:09:40.700 piece of software replacing you know you anytime or me anytime soon

01:09:45.520 can you talk about the radiology things i read i read about the work being done

01:09:49.420 is it stanford with the ai and radiology but the conditions have to be just so can

01:09:54.860 you just talk about that sure so um i think that uh i'm sad about ai and

01:09:59.300 potential to improve health care but um actually my some of my friends and i um

01:10:04.560 worked on ai that can pitch input a picture of an x-ray and i'll put you know

01:10:09.080 what's the appropriate diagnosis and it turns out

01:10:11.220 we were able to show in the lab that we could diagnose or recognize many

01:10:16.540 conditions as accurately as a board certified highly trained

01:10:20.460 doctor radiologist but it turns out that it worked great

01:10:23.560 if we were to uh train on data we collected from you know our

01:10:27.580 research from stanford hospital uh and then you see if the system did

01:10:31.500 work well on data from the same hospital from the same set of x-ray machines

01:10:35.140 it turns out if you take that ai system and walk it down to a different hospital

01:10:40.340 down the street with maybe an older x-ray machine maybe the technician has a

01:10:44.400 slightly different way of imaging the patient

01:10:46.100 the performance gets much worse whereas any human doctor can walk down the street

01:10:49.980 uh and diagnose at this other hospital you know you kind of roughly equally well

01:10:55.120 so i think that um one of the challenges of ai is we have a lot of

01:10:59.780 prototypes in the lab uh that you didn't read about in the news you know you see oh ai does

01:11:04.520 us what diagnose that human radiologists or something you may be about in the news

01:11:07.460 but it turns out that we collectively in the ai field we still have a lot of work

01:11:11.260 to do um to take those lab prototypes and put them into production uh in in a

01:11:16.760 hospital setting it will happen it's just that this will be some additional

01:11:20.280 years of work before some of the things that you know have been promised right

01:11:23.740 come come to fruition well the medical field is is so ripe for uh help from this

01:11:30.420 kind of technology i can think of a million ways in which it could change lives and

01:11:33.860 save lives but it's really every industry i know you've been making the

01:11:36.800 point it's that it's every industry that's going to be touched by this

01:11:38.900 eventually but before we move off the medical field may i just ask you about a

01:11:42.500 report in the wall street journal that got my attention

01:11:44.480 um okay among other things they're talking about

01:11:48.040 what we should expect in the next few years

01:11:50.480 toilets that screen for disease uh it says researchers at stanford

01:11:56.040 have developed a prototype toilet that uses an artificial intelligence trained

01:12:00.280 camera to track the form of feces and monitor the color and flow of urine why

01:12:04.740 is this necessary because it could potentially analyze micro stool samples

01:12:08.840 to detect viruses like covet 19 and blood it could potentially detect irritable

01:12:13.600 bowel syndrome or colorectal cancer and here was the part forgive me because i'm

01:12:17.800 really just a 12 year old boy at heart um that i wanted to ask you about so that

01:12:21.220 the toilet could identify individual users by scanning their anuses unique 0.98

01:12:25.980 characteristics or anal print now we no one wants an anal print going off to 0.78

01:12:32.440 some ai researcher but this is happening this is actually they're saying these 0.80

01:12:38.280 units could cost between 300 and a thousand bucks

01:12:40.640 they could be rolled out in the next couple of years is this what life is going

01:12:44.500 to hold for us yeah let's hope not a lot of that description was i i think uh a lot

01:12:50.380 of that you know the description you read sounds disturbing uh having said that i

01:12:55.600 think uh there are you know doctors that have to do many disturbing things for the

01:13:00.320 good of the patients but i think i i i don't i think a lot of us will not want

01:13:04.740 this in our homes anytime soon but we'll see you know okay doctors got to innovate we'll

01:13:08.920 we'll we'll see what the fda proves and what seems to be appropriate for patients

01:13:12.580 that may need it even if it doesn't seem like the right thing for everyone

01:13:16.060 because you know that's going to turn into one of these things where you get false

01:13:19.220 alarms every other day and you're in the doctor saying oh my anal print suggested i've got

01:13:24.020 colorectal cancer i don't know sounds like there's a ton of internet means to be created

01:13:29.660 off what you just said like it and i i listen as somebody who's on camera for a living uh a lot of

01:13:35.840 my life i there there are limits to how far i'm willing to go and i think i speak for a lot

01:13:39.740 of people so what about the other industries like how else could ai improve or negatively

01:13:45.560 impact our lives over the next 10 25 years

01:13:48.180 one of the challenges i see is uh ai as of today has clearly transformed the computer

01:13:54.960 uh software the consumer software internet industry where you use a website to launch website

01:14:00.780 operating app operating companies almost all of them i mean all of them use ai to great effects

01:14:06.060 um one of the challenges that still faces us is ahead of us is figure out how to use ai

01:14:10.960 to improve transform create value for all of the other industries out there um so for example one

01:14:18.240 thing i'm personally passionate about is manufacturing i think that for um american manufacturing to be more

01:14:24.400 competitive the road forward is not you know to try to just try harder to do the jobs that were around

01:14:31.440 20 years ago i think it's america and frankly all nations around the world should um raise a head

01:14:38.220 a head to figure out how this technology can work for manufacturing and for all those other

01:14:43.000 industries so for example uh it turns out that in many factories around the world today there are tons

01:14:48.640 of people standing around using their eyes to inspect you know manufacture things like automotive

01:14:54.000 component or pill bottle or food and beverage you know like a food component to see if there's a defect on

01:15:00.740 i think ai is clearly going to be able to do a lot of that work in the near future in an automated way

01:15:07.200 and if we in america want to embrace this technology you figure out how to use ai for automatic vision

01:15:14.080 inspection is coming now i'm i'm working on it my friends are working on it i think that's how many

01:15:19.280 industries become competitive but it turns out getting you had to work for manufacturing for health care

01:15:24.180 for agriculture these industries there's actually a different recipe it turns out the stuff that i was

01:15:28.480 doing you know like google and other internet companies it doesn't quite work so there's

01:15:32.400 something a little bit more needed but again a bunch of us in the ai view are working on this i hope we'll get

01:15:37.760 don't leave me now we got more coming up in 60 seconds

01:15:42.340 when we talk about baidu for a minute and just talk about china and its approach to data because

01:15:51.140 i know that they they really want to be leaders in the ai field and the united states is watching them

01:15:55.880 and they're watching us do you think that the chinese are any better than the googles of the 0.99

01:16:02.160 world where you were also the top guy at collecting information um synthesizing it keeping an eye on

01:16:09.100 people's habits and so on yeah i think that uh uh i think that uh china is phenomenal at some types of

01:16:16.900 technology the u.s is phenomenal at some types of technology um we do i think we do live in a you know

01:16:23.000 multi-polar world where i see innovations in the u.s and europe and china really frankly all around

01:16:28.300 the world and the ai community tends to be very global uh there is a global network where researchers

01:16:35.280 you know in research in singapore may publish a paper and then like two weeks later it's running in

01:16:41.240 some you know site in the united states and then someone in the uk will read it too and figure out

01:16:46.280 something to to apply and deploy in europe so i think we live in a global world where different

01:16:52.260 teams sometimes collaborate and different teams sometimes compete um uh i i i think actually one

01:16:58.920 thing i will say a lot of people underestimate um the importance of government support uh in the early

01:17:05.720 days of ai so not many people know this when i was running ai way back uh before modern ai deep

01:17:11.700 learning became popular a lot of reasons i was able to do my work was because doctor you know

01:17:16.980 of the defense agency in washington dc was willing to fund some of my work so i think without doctor

01:17:22.280 you know funding some of my research work i don't know that i would ever have gone to google

01:17:26.600 to propose starting the google brain project so i think i think just ensuring american competitiveness

01:17:31.840 is something i would love to see where are we on the scale are we are we the world leaders you

01:17:36.960 know you look at sort of the military superpowers and we know where we are but where is america when it

01:17:41.000 comes to ai i think that uh the two leading countries in the world in ai are quite clearly um uh uh the

01:17:48.300 u.s and china i think the u.s is the world leader in um a lot of basic research innovations but this is

01:17:56.140 not a lead that we should take for granted and we just got to keep on working really hard and what

01:18:01.340 about the creation of super intelligence because i i i read something about you creating something about

01:18:08.300 where a computer can recognize a cat i don't know you can tell me what it was but to me that sounded

01:18:14.900 like working toward developing super intelligence you know a computer that can learn on its own and

01:18:21.020 you know develop its own intelligence and improve its own intelligence but can you talk about that

01:18:25.440 about where we are on it what you've done on it and whether you think well you know how far along we

01:18:30.280 are um yeah the the the the cat result uh what was the google brain team one of the early results we

01:18:35.880 had was uh we built an ai system called a neural network and had it watched tons of youtube video

01:18:41.360 basically had it you know sit in front of the computer and watch youtube video for for like a

01:18:45.500 week and then we all said hey what did you learn and to our surprise one of the things that learned

01:18:50.720 was it had figured out or had uh learned to detect this thing which turns out to be a cat because it

01:18:57.900 turns out when you have an ai system watch youtube videos for a lot it learns to detect things that

01:19:02.540 occur a lot in youtube video so people faces occur a lot in youtube figured out you know how to detect

01:19:08.020 that there are also a lot of cats right there's another internet meme on youtube so it also figured

01:19:12.160 out how to detect that um it wasn't a very good cat detector uh but but the wrong thing about that

01:19:17.820 was that it had figured out that you know there's this thing it didn't know it was called cat cat

01:19:22.220 but there was this thing so they just learned to boy see a lot of this thing whatever it is i don't know

01:19:27.140 what it is so so it was it's pretty remarkable the ai system your network had figured that out uh by

01:19:32.720 itself now but again you know between that and and super intelligence or agi i i think it's very far

01:19:38.980 away i think that's worrying about um ai super intelligence today is a bit like over is like

01:19:45.820 worrying about overpopulation on the planet mars um i should hope that we will you know manage to

01:19:52.160 colonize mars and and maybe someday we'll have so many people on mars that we have children dying

01:19:57.960 because of pollution on mars and and you may be saying hey andrew how do you be so heartless to

01:20:03.280 not care about all the children dying on mars and my answer is well you know we haven't even kind of

01:20:08.140 landed people on the planet yet so i i don't know how to productively defend against overpopulation

01:20:13.320 there so i feel a little bit about i i think it's fine if academics study it you know publish some

01:20:18.380 theories on what to do when we have agi but it's so far away uh i personally don't really know how

01:20:24.780 to productively work on that problem now you are the co-founder of a group called corsera is that how

01:20:32.140 you pronounce it yes corsera and i feel like this this dovetails very nicely with one of the things

01:20:37.620 that nick was recommending when i talked to him about the future our children and so on and he was saying

01:20:42.320 the one thing the kids of the future are going to need to be able to do is understand that learning

01:20:46.240 is a lifetime process right that nothing is as static as it used to be the world is changing so

01:20:51.940 rapidly and our kids are going to need to be able to handle information at an even more rapid pace

01:20:57.100 than it now comes into their life which is already faster than ever and i feel like this is one of the

01:21:02.040 missions of corsera is to to nurture lifelong learning can you talk about it because it sounds

01:21:06.880 really interesting and it's been hugely successful yeah so yeah through corsera um hope we can give

01:21:13.140 anyone the power to transform their lives through learning um i was teaching at snappy university

01:21:19.300 about a decade ago actually over a decade ago and uh put my class on machine learning type of ai on the

01:21:25.580 internet and kind of to my surprise uh a hundred thousand people uh signed up for it and and i kind

01:21:31.820 of did the map you know i was teaching 400 people 400 students a year but when i did the map i realized

01:21:36.880 that uh for me to reach a similar audience a hundred thousand people teaching 400 people a year i would have to

01:21:42.500 teach at stanford university for you know like 200 years um and so so so based on that early traction

01:21:50.100 uh i got together uh with with uh with a friend uh to start corsera uh to create a platform that now

01:21:56.840 works with uh over 200 um uh universities and other institutions and companies uh in order to create

01:22:04.120 online learning courses that you know pretty much anyone in the world can access that's so great i mean

01:22:10.880 so it's like for those of us who didn't go to stanford or harvard or what have you uh but want access

01:22:15.940 to that kind of education though not full-time we can go here yeah in fact you know to to actually

01:22:23.420 i want to share two thoughts relevant to to to you to you know all of you watching this uh if you want

01:22:29.440 to learn about ai and not you know and cut through the hype one of the classes i'm most proud of is

01:22:34.900 ai for everyone uh on on corsera and i think i tried to uh give a non-technical presentation of

01:22:42.100 ai so if you want to know how will ai affect your life in the future how will ai affect your job

01:22:47.120 your industry you know there's a several hours of video that i hope will give anyone that's interested

01:22:52.520 uh a non-technical introduction to ai so you can think about this strategically and know how it will

01:22:57.820 affect you but also learn to recognize ignore some of the hype um there's one of the trend i'm

01:23:04.760 excited about which is you know with the rise of tech um i think we may i hope will eventually

01:23:11.300 shift toward the world uh and it's just irrelevant to all of you you know with children for example

01:23:15.980 but i hope we'll shift toward a world where almost everyone will know a little bit about coding and i

01:23:22.500 say this because many many hundreds of years ago we lived in a society where you know some people

01:23:28.060 believe that maybe not everyone needs to read right maybe there are just a few priests you know and

01:23:33.120 monks they had to learn to read so they could read the holy book to the rest of us or something and

01:23:37.980 and the rest of us we didn't need to read we just sit there and listen to them fortunately society

01:23:42.500 wised up and now with widespread literacy uh we've figured out that it makes human to human

01:23:48.260 communications much better i think that with the rise of computers in today's society you know for

01:23:54.200 good and for ill this is very powerful force i would love to see a lot of people able to just

01:23:59.840 learn to code so not just like not all of us need to learn to be great authors right you know i can

01:24:05.040 write but i'm not a great author i don't think everyone needs to be a great programmer but for

01:24:09.860 many of us there will become a time where um if if you're uh uh uh you know if you could write a few

01:24:16.880 lines of code get your computer to do what you want um just like literacy has created much deeper

01:24:22.340 human to human communications i think if everyone can kind of learn you know a little bit of coding or

01:24:28.000 computer literacy then all of us can have much deeper interactions with our computers and that'd

01:24:32.760 be a very powerful tool for all of you in the future well it certainly had a massive impact on

01:24:36.940 your life just reading your background um how did you get into it at such a young age it was your dad

01:24:41.860 i understand oh yes uh so my my dad's a doctor and uh when i was a teenager i was born in the uk but i was

01:24:50.840 living in uh singapore at the time but so my dad was interested in ai for healthcare so you know he kind of

01:24:56.200 taught me about uh his attempts to use you know frankly like 1980s ai which is not that advanced

01:25:03.420 to do medical diagnosis so that sparked off a lifelong interest in you know i do remember when

01:25:09.500 in high school i once had a internship i once had a job uh as a as an office admin and i don't remember

01:25:16.120 much from that job i just remember doing a lot of photocopying and and even i was like whatever you

01:25:21.440 15 16 years old i remember thinking boy if only why am i doing so much photocopying if only we could

01:25:28.360 write some software have a robot or something do all this photocopying maybe i could do uh something

01:25:34.340 something even more interesting and more valuable and i think that for me was part of my uh lifelong

01:25:40.120 inspiration to just write software that can help you know automate some of the more repetitive things

01:25:46.000 so that all of us collectively can tackle more challenging and exciting things well it's so great

01:25:50.920 because i tell you i went out to google and i i spoke to a bunch of executives there a couple years ago

01:25:56.040 and i know that they try to give the coders some stress relief some like a break because it can be very

01:26:05.060 intense work and one of the one of the stations on campus was sword fighting i'm like this is so great

01:26:11.220 you know they're just because you know you spend all day doing that it's very intense and you do need

01:26:15.220 a mental break a break for your eyes a break for your for your body so it's it's just a totally

01:26:21.700 different way of approaching the workplace yeah i think yeah i i find that uh i think coding is hard

01:26:29.060 work but i find that almost you know when i look across our society i think uh almost everything is

01:26:35.540 hard work right when when i walk into a manufacturing plant some of the work that you know my company

01:26:39.820 landing ai does for manufacturing i see the men and women on the manufacturing shop floor and they're

01:26:44.540 really smart uh at you know what they do and then i uh meet up with my friends from google and i think

01:26:51.720 they're really smart at at what they do i i i think that uh the world is a challenging has has lots of

01:26:59.180 challenging intellectually simulating or physically challenging work for us to to to do and and hopefully

01:27:04.360 ai tools can help make things a little bit better for everyone well i like that you sort of decide where

01:27:09.000 you're going to put your energies because i understand looking at you today and your blue

01:27:11.880 shirt it is no accident you are wearing that blue shirt and it is one of the areas of your life in

01:27:16.040 which you've chosen to simplify and streamline your decision making yeah i think uh yeah a few friends

01:27:23.700 have asked me yeah there's actually a four of those i think someone actually asked publicly why does

01:27:27.580 andrew wear a blue shirt all the time uh uh so i used to wear either blue or like a light purple

01:27:34.060 but then i realized every morning is like oh don't wear a blue shirt or a purple shirt

01:27:38.100 i can't decide it's like forget it i'm just buying a full stack of blue shirts and do that

01:27:43.740 i don't know so you don't have to think about it in the morning vera wang does the same thing vera

01:27:47.860 wang who dresses you know the most beautiful successful you know prominent people in the world

01:27:52.000 just wears sort of a black a column of black every day that's her uniform i did not know that 0.85

01:27:57.320 because she doesn't want to think about it same as you 0.58

01:27:59.700 yeah turns out there is a downside to this uh one day one of my friends was working on ai for

01:28:04.920 fashion thing uh and i tried to express him in a pinet said well you want to do ai for fashion

01:28:09.700 how about this how about this and she said andrew you have no credibility whatsoever when it comes

01:28:14.380 to fashion so okay i have to ask you one other personal question now i understand you married

01:28:19.360 somebody who's in robotics and i read that you you used a 3d printer to make your wedding rings

01:28:26.120 which brought up a lot of things for me which is number one i do not understand the 3d printer

01:28:31.340 at all my kids are using it at school it scares me i don't get it what is it what is it and how

01:28:36.580 does it print out a wedding ring how does it produce a wedding ring yeah so i think uh so carol

01:28:42.860 um she's from a prince michigan but we now are in washington state um so 3d printer takes you know

01:28:50.140 one way that 3d printing works is it uh takes little bits of um metal and melts them and kind

01:28:56.060 of you know deposits little drops of metal until gradually you end up building a ring i'm not

01:29:00.760 wearing a ring now i have it i have enough sense uh uh and then you end up with this you know

01:29:04.900 incredible shape uh whatever whatever you can almost anything you can imagine and program into

01:29:11.660 computer it can just by putting little drops of plastic or little drops of metal or some of the

01:29:17.760 substance uh just create this you know incredible 3d shape that's that's maybe difficult to manufacture

01:29:23.880 via other ways so i don't know actually this is one fun thing about technology or 3d printing

01:29:28.420 were still on really really cutting edge technology but now we have high school students able to use it

01:29:35.520 um i hope we like that for ai too frankly i i find that today ai seems a little bit mysterious maybe a

01:29:41.180 little bit overly so but uh actually last week i was chatting with a few high school students in

01:29:46.040 different parts of the country uh talking about you know they're taking online ai classes from

01:29:50.760 coursera or from or from whatever and now we have um high school students able to do things that

01:29:57.560 if done just five or six years ago would have been a chapter in a phd thesis at a place like stanford

01:30:04.380 right so really like what like what uh actually so so one thing happened to me i was attending a fair

01:30:10.460 make a fair um uh where i met you know this students that was demoing his robot that was taking

01:30:15.980 pictures of plants trying to figure out if they were um diseased if they had a you know disease on the

01:30:20.940 on the leaves or not so uh i looked at his work and i thought boy if this had been done five or six years

01:30:26.940 ago uh this would have been a chapter in someone's phd thesis at stanford university and you know what i

01:30:33.020 asked him how old are you and he said oh i'm i'm 12 years old uh so this is today's world no no no so

01:30:39.980 this is today's world i think anyone in the world you know can go and learn this stuff and then

01:30:45.260 implement this uh and and even though some technology seems so cutting edge i think that if someone out

01:30:52.220 there is you know watching this and wants to learn it a lot of tools are now on the internet right go go

01:30:57.500 go go learn it online from deep learning the article sarah and then on your computer you could

01:31:01.980 actually start developing stuff that while not the cutting edge stuff right that's actually still

01:31:06.620 pretty difficult you could actually do stuff that was kind of state of the art just a few years

01:31:10.940 all right i had on this subject i have a confession to make to you i were moving into a

01:31:15.420 new home or moving towns and i decided to not make my new home a smart home because my old smart

01:31:21.980 home was annoying me my dishwasher was yelling at me and my microwave was yelling at me and i was

01:31:27.340 walking around my apartment all day saying you are not the boss of me i am the boss of you shut up 0.97

01:31:32.140 i will unload you when i'm good and ready and uh the tv required 40 000 buttons to turn on and it's 0.98

01:31:38.460 like i i just want a dumb home for me because maybe it says i'm a dumb person but it it seemed easier to

01:31:45.420 me um and yet all of these appliances are getting smarter by the day and they're saying there's going to 0.97

01:31:51.340 be a refrigerator that's going to tell you whether things are spoiled on the inside and so on

01:31:55.660 so do you have a smart home do you recommend a smart home and how if at all concerned

01:32:03.260 should we be about people spying on us for lack of a better term you know i think people

01:32:09.020 they distrust google they think google's amassing information on them they distrust the government

01:32:12.780 they think the government could possibly hack into one of these appliances you know these are real

01:32:16.780 concerns you hear from people yeah so i know i i think that a lot of people are concerned about privacy

01:32:22.780 so in my voice web i have friends at many of the large internet companies and i know uh they're my

01:32:29.020 friends i trust them to tell me the truth many of my friends are genuinely uh a concern but also very

01:32:35.340 respectful of privacy so a lot of the large internet companies you know some better than others really do

01:32:40.700 have stringent privacy controls and makes it incredibly difficult for anyone to just spy on you now having

01:32:46.620 said that i i actually would be um disappointed i have no reason to think the u.s government can hack

01:32:51.660 into these devices but frankly i'd be a little bit disappointed if they can't um uh but yeah yeah so

01:32:58.460 you know by the way i i used to work on speech recognition right so i worked on this voice activated

01:33:03.020 devices uh one thing i'm not proud of for a long time even as working on these devices i had exactly one

01:33:08.780 light bulb in my home that was connected to my smart speaker because the configuration process is so

01:33:13.740 annoying so i got through you know configuring one light bulb so i could turn it on with a voice

01:33:18.300 command but after that i couldn't be bothered so i i think we still we still gotta make these things

01:33:23.820 better you know i we tend to inject the margin of a lot of things sometimes it's really great and i

01:33:28.940 love it uh but sometimes you know you do wonder right if if we're really helping solve people's problems

01:33:35.340 hey if we have more people working on it maybe even all collectively make all this tech much better

01:33:39.580 yeah no i've said in in this day and age it's not enough to pretend you actually have to be a

01:33:44.460 good person because someone's probably always listening watching amassing data they're gonna

01:33:49.100 know uh one way or another it's disconcerting but i don't know if you're not a criminal and you're not

01:33:54.860 you know dealing with terrorists and so on how worried do you need to be i don't know i'll give you

01:33:59.180 the last word yeah you know um yeah i think uh ai is the new electricity much of the rise of electricity

01:34:06.060 starting about 100 years ago transform every industry i think ai is now on a path to do the

01:34:11.420 same so i think really to anyone wondering if it's worth learning about it jumping in trying to help

01:34:16.540 all of us collectively navigate the future i think every citizen every government all of us individuals

01:34:23.180 should jump in and play a role in shaping a better future for everyone in light of this amazing

01:34:29.020 technology wonderful talking to you thank you so much for your expertise and your insights

01:34:34.140 you know thank you thank you it was really really fun to do this with you

01:34:42.540 so as i mentioned in our other episode this week we're we're scaling back a little for this week and

01:34:47.580 next week on our episodes just as we get ready to launch on serious my team especially my team has a

01:34:52.540 lot they need to be doing so we're going to launch five days a week starting on september 7th

01:34:57.500 but in the meantime we're a little bit of scale backs uh schedule for those of you wondering and

01:35:01.660 but our next guest who's going to be coming up on monday is one we've really wanted to have on for a

01:35:07.340 while controversial guy because he worked for trump and you know he's been completely excoriated by the

01:35:12.220 mainstream media but fascinating and really smart dude steven miller is going to be here you used to

01:35:17.420 have him on the kelly file all the time then you saw what the press did to him when he went uh on

01:35:22.460 inside the trump team but even just you know i've spent years talking to him there is no better person

01:35:29.820 to talk to if you want to understand what's happening in this country with our southern border

01:35:34.860 our northern border and our approach toward immigration in general uh so i'm really looking

01:35:39.020 forward to the conversation steven miller monday don't miss it in the meantime go ahead and subscribe

01:35:42.940 so you don't uh miss it download give me a five-star rating while you're there and give me a review

01:35:48.460 let me know what you think uh what do you think of ai are you in favor and uh what would you like

01:35:53.420 me to ask steven miller taking your thoughts right now in the apple review section uh or wherever you

01:35:58.700 download your podcasts thanks thanks for listening to the megan kelly show no bs no agenda and no fear

01:36:09.180 the megan kelly show is a devil may care media production in collaboration with red seat ventures

01:36:18.460 so

01:36:25.580 so

01:36:28.940 so

01:36:29.420 so

01:36:30.220 you

The Megyn Kelly Show - August 26, 2021

The Benefits and Dangers of Artificial Intelligence, with Nick Bostrom and Andrew Ng ｜ Ep. 151

Episode Stats

Summary

Transcript