The Mikhaila Peterson Podcast - 144. Is Artificial Intelligence A Threat To Human Existence？｜ Opposing Views

00:00:00.000 It's almost like the can't beat them, join them, but they're still trying to beat them

00:00:03.920 to a certain degree.

00:00:05.120 And it's by going even further.

00:00:08.740 What?

00:00:09.540 Okay.

00:00:10.740 What do we do about this?

00:00:12.020 Is there anything?

00:00:12.820 And how can you be on the pro AI side here?

00:00:17.000 AI does have a bunch of benefits when it's designed for human values and not for the

00:00:23.280 tech, the technocracy and the oligarchs who are trying to control everything or sorry,

00:00:27.720 mostly do control everything.

00:00:30.000 Now, this question of whether AI rules us is scientific and headline grabbing, but actually

00:00:34.460 kind of uninteresting.

00:00:35.640 The really interesting question is the tech of AI, which is much more mundane and down

00:00:42.280 to earth, but actually changing the world as we speak.

00:00:48.480 Welcome to episode 144 of the Michaela Peterson podcast.

00:00:52.420 I'm Michaela Peterson with tree allergies.

00:00:56.520 Sorry, I didn't give you guys a podcast last week.

00:00:59.140 i had no voice now my voice is kind of back this might be what i sound like forever we will see

00:01:04.480 this is an opposing views episode where i interviewed two people with vastly different

00:01:10.140 views on a subject this episode was about artificial intelligence in this episode of

00:01:15.460 opposing views we discussed the existential risk of ai job automation mass surveillance

00:01:21.160 deep learning government ai and the ex-ceo of google pushing for facial recognition

00:01:26.860 and social credit systems in America.

00:01:29.960 My first guest was the entrepreneur, filmmaker, podcast host, and author, Andrew Keene.

00:01:35.340 Andrew also hosts the award-winning documentary and TV show, How to Fix Democracy, although

00:01:40.780 he wasn't exactly optimistic about fixing America.

00:01:43.740 My second guest was Stephen Umbrello, the managing director at the Institute for Ethics

00:01:48.680 and Emerging Technologies, and editor of Ethics and Robotics.

00:01:52.540 Stephen holds a PhD in the ethics and design of AI.

00:01:55.620 I, he's the one who told me about the ex-Google CEO and what America has planned for its citizens.

00:02:01.800 Dun, dun, dun.

00:02:03.060 Let me know who you found more convincing in the comments section.

00:02:06.460 Or perhaps both guests are right.

00:02:10.220 This episode was sponsored by Schwenk Grills.

00:02:12.980 Do you remember the best steak you've ever had? 0.94

00:02:14.960 Or are you one of those poor suckers that thinks steaks are supposed to be gray?

00:02:19.180 Well, I'm here to tell you there's hope for you. 0.97

00:02:20.920 Let me introduce Schwenk Grill's 1500 degree portable infrared grill.

00:02:26.200 These things make anyone look like a natural.

00:02:29.020 They're made in America and the parts are all stainless steel that slide out for easy washing.

00:02:33.440 It even comes with a pizza stone for people who still eat pizza.

00:02:37.340 No, pizza isn't healthy, even in moderation.

00:02:40.400 Sorry, but you're still going to do it.

00:02:42.420 So they sent a pizza stone.

00:02:44.440 These grills are perfect for any deck, patio or backyard,

00:02:47.120 and they cook with the same infrared technology used in professional kitchens like morton's

00:02:51.760 steakhouse if you want to impress people or if you make your steaks gray anyways get one of these

00:02:58.860 they're amazing schwank has a great deal for my listeners get your very own american-made 1500

00:03:04.840 degree infrared grill today at schwankgrills.com all you need is promo code mp at checkout to get

00:03:11.840 $150 off your purchase. Again, that's $150 off at schwenkgrills.com with code MP, schwenkgrills.com, MP.

00:03:26.180 Andrew Keen, welcome to my podcast.

00:03:28.700 Well, thank you so much, Michaela, for inviting me.

00:03:32.040 Before we get started, can you give a brief background about who you are and what it is you do?

00:03:36.800 uh i'm a internet entrepreneur based in san francisco california on the edge of silicon valley

00:03:43.900 i have uh probably the the world's most viewed daily podcast about books and new writing called

00:03:52.100 keen on uh which you can find on the lit hub network uh i'm also the host of how to fix

00:03:58.540 democracy uh which is uh my award-winning show and movie about the future of democracy i've

00:04:05.680 written four books uh and numerous startups so i do lots of different things okay well you're on

00:04:13.560 here to talk about ai good well always happy to talk about ai and i am the real andrew key and

00:04:21.960 there's no nothing artificial about me or you i hope mikaela oh you never know actually i have

00:04:28.880 an artificial hip and an artificial ankle but they're not learning anything so good yeah

00:04:34.800 uh okay so what's your view on ai is it a positive is it a negative

00:04:41.420 i think you have to refine that question a little bit more it's like me saying to you

00:04:46.980 michaela what's your view on the internet or what's your view on life it's too big a subject

00:04:51.760 you need to sharpen the question okay i like you if we have self-learning robots is it a possibility

00:05:02.040 that they're going to become smarter than us and then view us as a hindrance.

00:05:06.620 I don't think it would be hard, Michaela, for robots to be smarter than us. We're a fairly

00:05:11.760 dumb species. Whether we'll become the slaves of them is a really interesting one. I think it's 1.00

00:05:19.520 a technological issue, which quite frankly is kind of beyond me. I'm not a technologist.

00:05:27.460 The idea, it's an ongoing debate amongst the hardcore scientists who study AI of whether

00:05:35.720 we can essentially, by giving AIs, giving robots, giving smart machines the ability to think,

00:05:46.900 whether they can outthink us. And so I, you know, philosophically, of course, it's a very sexy idea.

00:05:53.700 It's been the subject of many different science fiction writers, the idea of right from the 19th century.

00:06:01.660 In fact, even before the 19th century, the idea of us creating machines, Frankensteins, which come to rule us.

00:06:08.480 So philosophically, it's fascinating. 0.96

00:06:10.660 I think it would be an appropriate end to our species.

00:06:13.200 I think we get what we deserve.

00:06:15.580 But in technological terms, I'm not quite sure.

00:06:18.380 I just don't understand the intricacies of the debate in terms of whether or not we can build

00:06:28.920 machines which then go on to think for themselves and then think beyond us. If we do, then we are

00:06:34.600 the footnote, as somebody famously said, they are our last invention. So it's impossible to say,

00:06:43.020 for me at least. And I think the scientific community is very divided. And when you talk

00:06:48.220 to serious scientists rather than people who just seek headlines, most of them will say,

00:06:55.360 well, we really, really don't know it. It's such a complicated subject and perhaps a subject which

00:07:02.180 will surprise us because we'll only know when we know. So as I said, maybe you're an AI,

00:07:10.320 Yeah, maybe I'm an AI.

00:07:13.100 We'll only know when we know.

00:07:16.320 Maybe we'll never know, and that will be the interesting narrative 0.91

00:07:20.180 that we'll get ruled by AIs and we won't even know it. 0.84

00:07:23.120 Maybe we already are.

00:07:25.780 Well, that's horrifying.

00:07:28.080 Well, I don't know.

00:07:29.080 Aren't you happy?

00:07:29.900 You seem quite a cheerful person.

00:07:31.780 Maybe it wouldn't bother me in the world today if AIs are ruling it,

00:07:35.820 not doing a bad job.

00:07:36.940 They could do a lot worse.

00:07:38.440 Oh, sure.

00:07:38.840 I don't know.

00:07:39.240 i just had to move from canada i wasn't very happy by how that was being run but maybe it needs more

00:07:44.920 ai oh you're where are you now i'm in nashville tennessee so you went from canada to nashville

00:07:50.500 yeah isn't that going from the frying pan to the fire oh it depends what you identify the fire as

00:07:58.340 oh okay okay so ai it's so it's in the military now you know they're using smart machines to

00:08:08.280 track faces and everything. And then each side has their own AI. Um, do you think, so it's already

00:08:16.400 implemented in some areas. Do you think when it gets implemented in factories, this is going to

00:08:20.880 be an issue for factory workers or people who have more automated jobs? Yeah, that's an interesting

00:08:27.320 question. It's already being implemented. I mean, it's when people talk about new tech, when people

00:08:32.180 talk about the internet, when people talk about the next generation technology, some people use

00:08:38.060 the term now, Web 3 as the next chapter after Web 2.0. It's everywhere. It's ubiquitous.

00:08:45.200 So whenever you're talking about tech or big tech or big tech companies, you're talking about AI.

00:08:51.320 So Amazon, for example, is essentially an AI company. Google is an AI company. Facebook is

00:08:56.440 an AI company. Even Apple is an AI company focusing on hardware. So it's already in factories. If you

00:09:04.920 have the ill fortune of working in an Amazon factory, you're being watched all the time by

00:09:09.900 smart machines. Your ability or inability on the shop floor is being watched. And increasingly,

00:09:15.960 there'll be technologies which are often being championed by tech investors like Jeff Bezos,

00:09:21.640 the founder of Amazon, technologies of surveillance, which are designed to make

00:09:27.860 sure that we do our job and to evaluate us and measure us. So I don't think of that in futuristic

00:09:34.020 terms. It's already arrived. The really interesting question is when it begins to affect

00:09:38.340 the higher end workers. And I use that term carefully. I'm not sure we're so high end,

00:09:45.600 but whether it's commentators like you or I, whether it's lawyers or doctors or engineers

00:09:50.500 or politicians, but we're in the midst of this revolution. And it's a very profound one in terms

00:09:55.660 of the workplace. So for example, there's a big debate, as you know, about whether or not people

00:10:01.700 should go back to the workplace and the office, which is an interesting debate. And it doesn't

00:10:06.160 affect me because I've always worked from home, but a lot of people, it does affect.

00:10:09.620 And they think, well, they have more freedom at home, more independence, but actually they don't

00:10:13.500 with all this new technology, observing people's work. It doesn't mean just a camera, but it's

00:10:18.400 smart tech, which can determine whether or not we're actually working or doing something else.

00:10:25.540 So all this technology is designed to measure our performance in the workplace and perhaps

00:10:31.300 elsewhere. And it's already arrived. We're in the midst of this revolution. It's a very profound

00:10:34.960 one. It's probably the most profound one of the 21st century. So that's why, um, you know,

00:10:40.520 this question of whether AI rules, this is scientific and headline grabbing, but actually

00:10:44.900 kind of uninteresting. The really interesting question is the tech of AI, which is much more

00:10:51.600 mundane and down to earth, but actually changing the world as we speak.

00:10:55.480 so are you talking the tech are you referring to things like ai that's able to answer

00:11:02.500 well that's different that's ai that's used for medicine uh so there's lots of ais that help

00:11:12.720 engineers and doctors and journalists i mean there's ai um for me as a writer i mean it's very

00:11:19.660 basic AI, but it helps me spell. Um, so that's a different issue. What we're talking about AI in

00:11:27.040 the workplace, it's AI that evaluates our performance. So it goes the other way. AI can

00:11:32.460 be very useful too. I mean, there's no doubt about that. No one would argue against AI that can help

00:11:36.880 fix cancer, but it goes both ways. I mean, the thing with AI is it's so ubiquitous. It's like

00:11:45.780 me saying, are you for or against oxygen? Or are you for or against the internet? Clearly,

00:11:52.660 there are arguments for and against, but it's so ubiquitous and essential

00:11:56.120 that it's very easy to make arguments for or against.

00:12:00.860 Do you think there's any way going forward that the government or somebody, I don't know who,

00:12:06.040 could implement policies to stop some of the negatives people are concerned about,

00:12:10.280 about AI in the workplace? Yeah, there will be. But remember, the American government,

00:12:16.640 you're a Canadian immigrant. The American government is profoundly dysfunctional. It 0.95

00:12:22.780 has one of the least effective governments in the world, staffed by generally incompetent people, 0.96

00:12:29.620 because smart Americans don't go into government. It's badly paid and frowned upon in our social

00:12:36.180 hierarchies and culture. So the American government won't be the government that

00:12:41.000 addresses this in the same way as the American government hasn't been the government that

00:12:45.000 addresses the dangers of big tech companies. This will likely be done in two ways, by two

00:12:51.620 kinds of government, by the authoritarian technocracy in China, where they will just

00:13:00.800 announce regulation and legislation of AI, according to some wise man precept, Chinese

00:13:10.080 culture, which has been inherited now by the Chinese Communist Party, and then by the democracies

00:13:17.100 of Northern Europe, the German government, the Scandinavian government, and the EU government,

00:13:23.140 which I think will do a reasonably good job. American government is always second or third

00:13:29.080 in this. So eventually, as these threats are regulated and legislated against, so the American

00:13:36.260 government will pick it up. They've done that with big tech. They're only beginning to address

00:13:41.140 the dangers of big tech now after the Europeans and the Chinese have addressed it.

00:13:46.420 What did the Chinese do to address the dangers of big tech?

00:13:50.120 Well, they shut them down. They fined them. They put their leaders in jail. I'm not saying it's 0.98

00:13:54.300 necessarily a good thing, but that's one effective way of doing it. Chinese have a very strong, 1.00

00:13:59.840 effective, functional government, whereas the Americans don't, which is why the Chinese economy

00:14:06.400 is stronger. It's why Chinese society is more perhaps successful. I would prefer, I think,

00:14:13.980 personally to live in the United States, but certainly China is a more successful society

00:14:17.760 than the united states richer happier people less crime less prisons all the indicators all the

00:14:29.300 indexes which indicate happiness and success is there less poverty there much less than in the

00:14:36.760 united states when's the last time you're in san francisco mikaela it's become a giant homeless

00:14:41.420 camp yeah i was just in new york yesterday it's becoming like san francisco

00:14:47.760 awful toronto was like that that's why i moved not yeah not as bad as new york but getting there

00:14:53.120 right well how's nashville amazing yeah no homeless no poverty not really i mean there

00:15:01.160 are there are areas like any you know american city but compared to toronto it's much better

00:15:07.200 people are calm and happier good i'll have to come and visit yeah you should it's pretty nice down

00:15:13.160 here in the middle of America. Okay. What about the fact, so I want to go back to China for a sec.

00:15:25.540 There's an issue with the fact that government has as much power over citizens. Don't you think

00:15:30.680 AI implemented at a government level surveilling everybody? Isn't that an issue, collecting all

00:15:36.300 the data? Oh, it's definitely an issue. I mean, I've written and spoken about this a lot. The

00:15:41.600 Chinese are refining what I would call a digital Orwellian state. Orwell, of course, in 1948 wrote 0.99

00:15:51.840 a book called 1984, which was a warning about, I guess, an analog surveillance state, a sort of

00:15:59.260 Stalinist analog surveillance state, which he thought was coming into being after the Second

00:16:03.680 World War, when Stalinism was the dominant form of government in Eastern Europe and in the Soviet

00:16:09.920 Union and imminently in China. What the Chinese are doing are refining a kind of techno-capitalist 0.95

00:16:17.700 surveillance state where everyone is watched all the time and where the government is controlling 0.95

00:16:22.840 the levers, the technologies of surveillance. Yeah, it's terrifying. I don't want to live there.

00:16:28.680 But it's effective. Whereas in contrast, the American government is

00:16:33.340 is profoundly dysfunctional and is resulting in a society that's failing, which is another

00:16:42.120 subject. Your fellow countryman, Stephen Marsh, I'm interviewing him next month on my Keen On

00:16:48.500 show. He has a new book out about the imminence of civil war in America. There's going to be a

00:16:53.100 whole rash, and I use that word carefully, of books in the early 2022 about civil war in America,

00:17:00.720 the breakdown of government. The government's already broken down, but that's not justifying

00:17:05.680 the Chinese system. It's not either or. There are alternatives. As I said, the European model 0.87

00:17:12.880 of a stronger, more functional state with politicians and regulators who are respected, 0.98

00:17:20.760 that's probably the best compromise between Chinese techno-authoritarianism and American

00:17:27.680 dysfunctional government don't don't areas of europe have an immigration problem that's

00:17:33.820 potentially destroying parts of it though you mean um americans coming over no i meant places

00:17:42.680 like paris and from north africa coming over specifically that's going to do with the

00:17:48.640 government i mean immigration policy wouldn't that be the government

00:17:52.320 yeah i mean that's another question i don't think that has anything to do with surveillance

00:17:58.280 culture or the effectiveness of government um and uh i i don't think that um you know i think

00:18:06.140 they're different parts of the world so the issue of immigration in europe and in in north america

00:18:13.020 is different but yeah i don't think that's got anything to do with it i don't think it has

00:18:17.320 anything to do with surveillance but i was commenting on how effective the government was

00:18:22.280 for building a society yeah but the european model still works and it's still trying to be

00:18:29.000 emulated around the world whereas the american north america the united states one isn't america

00:18:34.100 is a joke around the world the american the dysfunctionality of american society and

00:18:39.240 government now is is perhaps the dominant theme of early 21st century the global world they're

00:18:46.560 laughed at around the world because the system doesn't work and the country's on the brink of

00:18:51.120 civil war. That's a different issue from having an immigration policy problem, which of course

00:18:56.820 it does have. You said you've written books on how to fix democracy. I've made a film about how

00:19:04.020 to fix democracy. And so do you have a solution for America that can avoid civil disaster?

00:19:10.540 No, I think America is beyond hope.

00:19:13.700 Actually, I'm afraid it's I'm about to do an interview on my show with Eric Topol, you know, the leading one of America's leading doctors.

00:19:21.620 And I think America is the equivalent of that corpse or almost corpse that's about to die.

00:19:28.260 And there's not much you can do about it, I'm afraid.

00:19:32.240 Well, do you have a plan? Are you going to stay here and watch the collapse? Are you moving?

00:19:35.880 no it doesn't bother uh coastal elites like myself we fly in and out doesn't make a difference

00:19:44.100 it's not going to be full civil war full-scale civil war like in the 19th century it will be

00:19:50.580 outbreak i mean it's already kind of i mean this is a this is separate from ai but it will be a

00:19:58.900 low-level kind of civil war that sort of lebanese style or it's not gonna i don't think it will be

00:20:04.820 like Syria, but basically a government breakdown, sort of eruptions of violence around the country.

00:20:11.620 It's not going to affect New York or San Francisco or Los Angeles, but it still will result in the

00:20:19.860 breakdown ultimately of order. It's this book by Stephen Marsh. It's really good. He does an

00:20:26.000 excellent job. I read it a couple of weeks ago. It's out in early January. You should get him

00:20:30.640 on your show even marsh i'm writing that down m-a-r-c-h-e okay that's fascinating okay you're 0.95

00:20:40.240 very interesting character i'm kind of upset i brought you on to talk about ai why well i don't

00:20:47.580 know we could be talking about the crash of civilizations that's what we are we are maybe

00:20:53.680 i'm an ai designed to upset you ai is part of it all as well you know ai is um there's a piece of

00:21:02.920 the narrative it's a chapter in the breakdown relying on machines which don't do a very good

00:21:08.260 job generally yeah you know i had to talk to one on a phone yesterday you did credit card oh yeah

00:21:15.700 it was awful yeah yeah i always put the phone down whenever you talk to if you i think we all have to

00:21:21.720 uh agree that whenever you're talking to an afv put the phone down enough times eventually their

00:21:26.800 masters and they remain humans will um will get the message and put real people on yeah or press

00:21:34.080 zero enough times that sometimes works although if you you never know michaela if you do it enough

00:21:38.780 times you may get one of those immigrants from europe one of those north africans who you don't

00:21:43.140 like they'll be on the phone with you they'll know who you are they'll be calling you all the time

00:21:49.360 i don't know if muhammad from tangiers will be on to michaela

00:21:55.200 i still think that there's an immigration policy problem in some another show have me back on we

00:22:02.300 can have a serious chat about that okay okay i will for sure so do you so the guy i talked to

00:22:11.060 who was more you know this is going to happen ai is going to become a thing but i'm not saying it

00:22:15.680 It isn't. I've said it's going to happen. It's inevitable.

00:22:18.840 But do you think they're actually going to be able to make machines that can outthink humans?

00:22:24.280 Well, you asked me that before. I think you've got to rephrase the question because

00:22:31.140 from a technological point of view, I'm not able to answer that. So maybe if you rephrase it,

00:22:35.800 would I like to have machines that can outthink humans? Maybe ask me that.

00:22:41.680 Okay. Would you like to have machines that can outthink humans?

00:22:44.360 i i shouldn't say this but i but since you have a pretty big audience i'm not sure if anyone's

00:22:51.160 going to be watching this as soon as they see me they'll probably switch off but um 0.82

00:22:54.320 i would like to have machines that outthink humans i think we're such a pathetic species

00:23:00.800 that it would actually be amusing i'm not sure how we'd react maybe it would get us to grow up

00:23:05.500 and think a bit more responsibly but i don't think it would necessarily be a bad thing it'd

00:23:09.740 interesting. I mean, maybe they'd be wiser, more humane, who knows? I mean, the idea of the kind

00:23:16.680 of the hardcore AI community is we could fill these algorithms with our best qualities. So we

00:23:23.980 could make them human. We could make them sympathetic. We could make them smart. And I

00:23:31.500 guess that's possible. The problem is whenever you hear those arguments and maybe there's some truth

00:23:38.360 in them. They're always going to be owned by humans. So it's like the Google argument or

00:23:44.660 the Facebook argument. Oh, we're going to make the world a better place. We're going to liberate.

00:23:48.800 We're going to democratize. We're going to decentralize. We're going to empower. And all

00:23:54.440 those things may be true of social media and ubiquitous search and Amazon, but ultimately,

00:24:00.420 these products and services and platforms have human masters. So the problem with making an AI

00:24:07.540 that's smarter than us, more sympathetic, more responsible, wiser, is it will still have human

00:24:14.620 masters who will profit economically, politically, culturally. That's the really scary thing about

00:24:22.040 dominant AIs. You have it in Silicon Valley. I've been living here for the last 30 or 40 years,

00:24:31.260 and I've seen it time after time after time. Technology is saying, oh, now we have the fix.

00:24:36.600 now we have the solution now we have technology that will essentially make us better that will

00:24:42.340 take our best bits and and liberate us from our bad bits and sometimes that's true we've seen it

00:24:50.620 with social media we heard all the arguments oh well big media is bad is controlled by these

00:24:57.940 people no one has a voice so we gave everyone a voice we gave them a platform we even made it free

00:25:03.820 and look what's happened. We've had the destruction of democracy. We've had

00:25:08.040 basically online civil war, the breakdown of civility. And the biggest problem of all is there

00:25:16.200 is a young man, and he remains a young man who owns this AI. His name is Mark Zuckerberg.

00:25:21.200 And he's made tens, hundreds of billions of dollars off this AI. The same is true of Google

00:25:26.880 and Amazon and Facebook and all the rest of these companies. So the real problem with AI is not the

00:25:32.740 AI itself, it's who owns it. So the vision that some people have, which I don't believe,

00:25:40.480 I don't know how it works, is to have an open source platform, a Bitcoin blockchain style AI

00:25:49.020 that's owned by everyone collectively. But those things never work. Ultimately,

00:25:54.580 it's like those free credit card offers or those people who are calling you when you talk to your

00:25:59.420 credit card companies. The promise and the reality are entirely different. So whenever there's a

00:26:05.860 promise of everything's free and everything's perfect and we're in control, I'm always deeply

00:26:10.660 suspicious. I was suspicious 20 years ago about Web 2.0, and I was proved right. I was one of

00:26:15.920 the first people to raise a warning about social media and Web 2.0. And I think the same is true

00:26:23.940 of AI. The same is true of Bitcoin and blockchain. When you listen to these promises, oh, well,

00:26:28.260 we're going to liberate currency from the central banks. Everyone now will earn currency. There'll

00:26:34.080 be no more money. It will be a P2P economic system. The truth is that they're still speculators and

00:26:41.540 owners of these products. The same will be true of AI. I don't see any way that AI can really

00:26:48.100 become a public good unless it's controlled by government. Then we come back to our earlier

00:26:53.760 conversation. We have the truly dystopian Chinese model where the government owns AI and controls 1.00

00:27:02.340 it for its own benefit. Or we have the American model where government is just dysfunctional, 0.72

00:27:09.000 nothing happens. So everything is controlled by private companies. Or we have a European model

00:27:14.780 that tries to mix the two and have an authoritative government, which still acts in a responsible

00:27:21.160 democratic way. So that's the challenge. You have all these new AI platforms and they're all

00:27:29.320 promising some sort of openness. When anyone, Michaela, comes on your show and uses the O word,

00:27:35.500 open, be suspicious because nothing is ever truly open. There's always someone owning it behind it.

00:27:41.200 So the open word, the O word, open is always really about ownership. And my fear with AI

00:27:50.340 is we're never going to get beyond ownership. There will always be someone who owns it.

00:27:56.000 There'll always be some smart human, maybe a James Bond evil character, a Zuckerberg style

00:28:02.620 character, a Jeff Bezos and Elon Musk who own this stuff and are getting enormously rich off

00:28:08.400 it and enormously powerful. And that's from my point of view, it's chilling, particularly when

00:28:15.840 they articulate the double speak of Silicon Valley and promise to make the world a better 0.76

00:28:22.380 place. But in truth, they're only making the world a better place for themselves and their

00:28:26.440 investors and workers. Okay. I think we're pretty much in agreement. My concern- We are? I didn't

00:28:33.900 come on to agree with you. Well, I'm sorry. What about those Muslims? Those dangerous Muslims

00:28:39.960 on the border? You said we have to do that in another podcast. Okay. We'll do that in another 1.00

00:28:44.900 we should um i think my concern with ai would be somebody controlling it who thinks that they're

00:28:52.400 doing something good for the world and really truly believes that yeah aren't because you don't

00:28:57.200 have evil people you don't have i don't think you really have evil people i think you have people

00:29:01.920 who think they're doing good and then do evil which is yeah there was a wonderful book one

00:29:07.560 perhaps the most profound 20th century american thinker on tech and big tech companies was a

00:29:14.040 a man who taught at New York university could, um, Neil Postman. And in 1984, which of course

00:29:21.160 was the year that Orwell predicted his 1984 would come into being, uh, Postman wrote a book called

00:29:27.800 Amusing Ourselves to Death. And in that introduction, it's a wonderful introduction,

00:29:33.200 it's only about four or five pages. He says, we all got it wrong. The real danger in 1984 wasn't

00:29:39.140 Orwell's 1984. It was Brave New World. It was Aldous Huxley's dystopia of a techno-authoritarianism.

00:29:48.340 And I think he's right. And I think the book, for your viewers, if you want to really worry about

00:29:55.260 the future, don't read 1984. Read Brave New World. Huxley's, I think it was 1934, warning about a

00:30:03.680 world saturated with technology, a world in which government uses technology to make everyone happy.

00:30:11.640 Yeah, I read that in high school. That was a scary book. Too much pleasure, right?

00:30:17.120 Too much pleasure. And then the force of no pain. It's a dystopian utilitarianism.

00:30:24.460 And of course, it's one in which everyone is intoxicated, a government which gives out pills

00:30:31.360 to make people forget um and and i'm afraid you know that's what social media is and that's what 0.93

00:30:37.660 that's the world we're living in it's very different from say the chinese model

00:30:42.060 china isn't doing that i mean china is kind of mixed the chinese digital totalitarian system

00:30:50.680 is a mix of orwell and huxig as if the leader of china maybe like you read in high school both 0.82

00:30:57.820 brave new world in 1984 and say, well, I don't want, I'm going to get them both. And that's 0.99

00:31:02.460 what they're doing in China. You think that kind of system is going to outlast the U S

00:31:08.680 yeah, it's a much more successful system. If you're, if you're a, um, a government leader or,

00:31:17.840 um, want to be politician in central or Latin America, in Africa, in, in, in most of South

00:31:25.340 Asia, and you think, well, which model are we going to choose? Are we going to choose the

00:31:29.580 dysfunctionality of America on the brink of civil war for a city like San Francisco, which is 0.88

00:31:33.940 dominated by homeless people, or the Chinese model that works? It may not be very pleasant, 0.96

00:31:38.960 may not be very moral, but it works. I've done a lot of speeches around the world on tech.

00:31:44.040 I was at the UN just before COVID doing a speech to the assembly there on the dangers of Web 2.0

00:31:52.120 and the dangers of big tech in America. And the people who were most sympathetic to what I was

00:31:57.780 saying were many of the delegates from Africa. Back in the old Cold War, Cold War 1.0 in the

00:32:04.100 late 20th century, the Western model worked. No one wanted to go and live in the Soviet Union. 0.91

00:32:10.820 It was dysfunctional. It was dark. It didn't work. Everyone lived in poverty. Everything was 1.00

00:32:16.500 unpleasant. But the Chinese model is a much more effective model than the US model. So that's the 0.99

00:32:21.720 challenge is we've got to come up with a model that can compete with China, which isn't as

00:32:26.900 autocratic, isn't as evil, isn't as unpleasant.

00:32:33.100 And is it going to be the business owners who do that considering the government's dysfunctional?

00:32:38.800 It may be the business owners. I've had a lot of people on my show talking about social

00:32:44.000 capitalism. It may be reforming democracy itself, citizen assemblies, reforming government. One of

00:32:49.180 the biggest problems in America is just the political system is archaic. It's built on an

00:32:54.540 18th century rural country that no longer exists. So we need profound reform. Maybe we need to

00:33:02.020 unleash the citizen. They're doing that again in Europe. They're doing it with citizen assemblies.

00:33:07.780 They're doing it in Canada. They're not doing it in the United States. So America, for all its

00:33:13.280 um forwardness and innovation when it comes to tech is profoundly reactionary and conservative

00:33:23.220 when it comes to any kind of institutional or constitutional reform okay well i'd love to keep

00:33:31.640 talking to you i don't want to keep you on because i know you have something to do yeah we're lovely

00:33:36.420 talking to michaela and we'll have to come on again and talk about those immigrants in europe

00:33:41.140 i would really enjoy that uh if if people want to find you if they don't want to find me um i'm

00:33:48.300 everywhere uh and nowhere like ai you know you i'm aj keen on twitter uh uh follow uh follow my keen

00:33:56.860 on show it's um uh it's of course you can find it anywhere on apple it's on lit hub it's the

00:34:03.400 one of the the the featured shows on on lit hub.com i have a sub stack page called keen on

00:34:10.860 I'm not on Facebook because I don't like Facebook.

00:34:14.400 I'm on LinkedIn.

00:34:15.600 So I'm easy to find them, ajkeen.com on the web.

00:34:18.760 So it's a real pleasure.

00:34:19.860 I enjoyed this conversation, Michaela.

00:34:21.600 And thank you so much.

00:34:23.180 Thank you very much for coming on.

00:34:24.440 This episode was brought to you by NordVPN.

00:34:27.180 Have you ever worried about things like,

00:34:29.300 is my data secure when on public Wi-Fi?

00:34:31.760 Or what if somebody steals my identity?

00:34:35.160 Don't be that person.

00:34:36.640 That would be a neurotic person.

00:34:38.380 I'm certainly not.

00:34:39.360 but i may be so unconcerned i go too far in the wrong direction to get the best protection against

00:34:44.860 cybercrime nordvpn's new threat protection feature blocks trackers ads and malware lurking online

00:34:52.480 it's a brand new feature of nordvpn and comes absolutely free most people know about nordvpn

00:34:59.180 i probably don't have to tell you guys it's the fastest vpn on the planet but did you know it's

00:35:04.880 the safest as well nordvpn uses the same encryption standard as top secret nsa files and they have a

00:35:12.620 strict no log policy on your data which is why they're headquartered in panama it's a panama law

00:35:18.680 thing apparently but most importantly nordvpn can be used to bypass countries and companies

00:35:24.120 that are stopping you from accessing videos and content online like serbia blocking disney

00:35:29.240 hotels blocking certain YouTube videos, and Canada blocking anything remotely good on Netflix.

00:35:35.780 Grab your exclusive NordVPN deal by going to nordvpn.com slash tmpp to get a huge discount

00:35:44.260 off your NordVPN plan plus free threat protection plus one additional month for free. It's completely

00:35:52.800 risk-free with Nord's 30-day money-back guarantee. That's at nordvpn.com slash tmpp. I hope you enjoy

00:36:01.700 the rest of this episode. Stephen Umbrella, welcome to my podcast.

00:36:10.320 Thank you for having me.

00:36:12.220 Before we get started, can you give my audience a brief intro about who you are and what you do?

00:36:17.700 Sure. I'm right now the Managing Director for the Institute for Ethics and Emerging Technologies, which is a 501c3 nonprofit think tank out of Boston.

00:36:29.060 And our general aim is to promote ideas about how technological progress can increase freedom, happiness, and human flourishing in democratic societies.

00:36:39.020 And I'm also a research fellow at the Delft University of Technology, where right now I'm working on creating stakeholder meetings for technologies like geoengineering, neurotechnologies, and digital extended reality.

00:36:53.560 And aside from that academic stuff, I'm the editor of the ethics and robotics series with Tribe in Publishing, where my forthcoming book, Designed for Death, will become a note soon.

00:37:06.740 Designed for Death. Okay.

00:37:09.020 That sounds intriguing. So you're of the opinion that AI isn't going to take over the world and

00:37:15.400 then destroy us eventually? No, absolutely not. Well, first of all, we have to, I think we're a

00:37:24.500 lot of source where a lot of confusion comes from in both the academic and particularly in public

00:37:29.740 conversations on AI is, well, what is AI exactly? Cause it's kind of like a buzzword right now that

00:37:35.740 everyone likes to use. And the issue is that there's no agreed upon definition of actually

00:37:42.020 what AI is like globally. So I'd argue if I were to give a definition that it would be best

00:37:48.340 understood as a class of technologies that are autonomous, they're interactive, so we can interact

00:37:54.580 with them. They're adaptive, so they can learn to new situations, new challenges in their

00:37:59.820 environments, and they're capable of carrying out human-like tasks. And in particular,

00:38:05.680 when we use the word AI and that definition, it's mostly those technologies which are based

00:38:10.680 on machine learning, which allows the system to learn on the basis of their interaction with

00:38:16.120 and feedback from their environment. So I would say that that would probably be the

00:38:20.220 simplest way of understanding what AI systems are, but there's actually the European Commission

00:38:26.180 actually recently came out with, and they're well aware of this issue, a draft legislation

00:38:32.540 for artificial intelligence. And it's for promoting uptake of AI and addressing the

00:38:38.920 risks associated with the new technology. And they actually came up with a pretty good

00:38:43.020 definition of AI. And that's probably what would be the benchmark from now on of what AI is.

00:38:49.460 What I want to start with, I think, is if AI continues to be implemented, is it going to end

00:38:55.460 up taking jobs away from people who do kind of not to call it menial tasks, but repetitive tasks.

00:39:01.820 And is that going to be an issue? Yeah, I think this will undoubtedly be the case. And we've

00:39:08.980 already seen how the various industrial revolutions from mechanization and electricity to telecommunications,

00:39:15.480 electronics, and more recently, obviously, big data, AI and the internet. And these revolutions

00:39:20.720 naturally come with huge shifts in labor and production. We can think back to the English 0.72

00:39:26.180 textile workers of the 19th century that destroyed their machinery, in particular, the cotton and

00:39:31.060 wool mills that they believed were threatening their jobs. And these people are called the

00:39:36.320 Luddites. And now we use that term to refer to someone who's opposed to new technologies and

00:39:42.500 associated ways of working with those technologies that are substitutive. And the IET where I work

00:39:48.000 is actually doing a lot of research on this particular topic. And actually in collaboration

00:39:52.120 with the University of Massachusetts, Boston, the IAT hired actually its first postdoctoral fellow

00:39:57.540 in the future of work to explore questions surrounding technological unemployment,

00:40:02.940 which is essentially the question that you just asked, and how to address it. And looking at

00:40:07.240 questions like if automation is liberating us from the drudgery or just exacerbating inequalities.

00:40:13.500 And I think the next few years of this research is going to be pretty interesting.

00:40:17.660 So anybody who's interested in that should just keep an eye out on the work we're doing because we'll publish it.

00:40:22.920 Interesting. Okay.

00:40:24.360 I suppose the argument might be that if some of those jobs go away, humanity will adapt, right, in some way that maybe we can't even imagine at the moment.

00:40:35.140 The not imagine at the moment, perhaps.

00:40:37.280 I think that that comes with an underlying presumption that AI needs to substitute.

00:40:48.300 Now, of course, it will substitute to a certain degree because we've already seen that in the other industrial revolutions, how that's changed.

00:40:55.160 And, you know, there's some jobs that simply do not exist anymore.

00:40:57.880 And unfortunately, that means there's certain jobs that may not exist anymore going into the future.

00:41:03.160 and that's going to be a serious question that we don't really have a solid answer to is what are

00:41:07.520 we going to do with the people that are going to be displaced because there will be a certain

00:41:11.300 portion of the population that will be displaced okay well i guess we'll see what happens do you

00:41:20.040 when do you think that's going that kind of thing is going to happen i mean it's already happening

00:41:23.300 but is there going to be a period in the next 10 years when that's going to pose a significant

00:41:27.460 problem? I don't think it'll pose a significant problem if we readjust the way that we look at

00:41:34.260 AI. So if we have this active attempt at creating artificial intelligence systems with the purpose

00:41:40.800 of substituting humans, rather than trying to empower people, rather than trying to augment

00:41:48.520 what they're already capable of. So these are systems that work alongside humans in order to

00:41:54.960 increase their capabilities, then perhaps we will increasingly face the issue of technological

00:42:00.920 unemployment. But if we shift, if we have a paradigm shift in the way we actually look at AI

00:42:06.360 and understand it as being supplementary to human labor in order to increase our productivity

00:42:12.760 and also perhaps our wellbeing, then perhaps we won't face as much of that particular element of

00:42:21.700 AI substituting people of technological unemployment. Yeah, I guess that would have

00:42:26.860 to be implemented at the government level, probably. Otherwise, corporations are going

00:42:30.320 to be like, nah, we don't need workers. I'll just buy this program. It'll do everything.

00:42:35.780 We see that already, don't we, to a certain degree? Yeah, definitely. Cool. Sounds like a

00:42:42.440 fun job, kind of. Very interesting. Do you think there are issues with artificial intelligence

00:42:49.220 impacting privacy and data? Oh, wow. I would say that we can rephrase that a little bit and say,

00:42:59.360 what do we do about the privacy issues that we have now given in terms of how governments control

00:43:06.600 AI systems and systems more broadly, because there's already issues that perhaps they will

00:43:10.920 be exacerbated. I think they will be, but there's already issues with governments controlling AI

00:43:15.480 system. So China is the classic example. And naturally, the corporate media is more than

00:43:21.260 willing to saber rattle and push the narrative, despite the fact that the Chinese state does

00:43:26.580 employ these methods. Still, it shouldn't distract us, the very real example of China and what

00:43:33.000 they're doing from what's actually going on right now on the home front. I'm not sure if you've

00:43:37.860 heard. But back in 2019, the organization called EPIC, which is the Electronic Privacy Information

00:43:46.120 Center, filed, strangely enough, successfully a freedom of information request to the US

00:43:51.520 government. And they got a bunch of documents back revealing how the National Security Commission

00:43:56.100 on AI, which is a relatively little known body, I never heard of it before this, which was created

00:44:02.900 with the 2018 NDAA, the National Defense Authorization Act. And these documents revealed

00:44:10.740 how artificial intelligence can address, and this is quote, the national security and defense needs

00:44:17.720 of the United States. And it discusses even further the structural changes that need to be

00:44:23.300 made in both American economy and society to ensure that the U.S. maintains a technological

00:44:29.460 advantage over China. So, you know, you go, okay, that makes sense. Given what China is,

00:44:38.040 do we want to stay on top? And that's what America does. But the document is actually 0.93

00:44:42.920 pretty scary in terms of the suggestions that are made. They argue that the U S should adopt

00:44:47.860 a China style mass surveillance program in order to lead or even surpass China concerning AI driven

00:44:54.380 technologies like facial recognition and citizen scoring, which is actually what's prohibited in

00:44:59.820 that European regulation that I mentioned before, that type of technology. And funny, although

00:45:05.280 unsurprising, that U.S. body responsible for these proposals that was gotten through this FOIA

00:45:11.920 is chaired by the former head of Google, Eric Schmidt, where it further goes. This is crazy.

00:45:18.300 Yeah. And it further goes into how the regular, this is what the document says, that the regulatory barriers need to be removed on emerging AI technologies.

00:45:28.640 And it even goes so far as to boldly say that existing privacy laws need to be changed so that the U.S. government can do all of that.

00:45:36.920 What I just said more openly, whereas up until now, they've been doing it secretly to citizens.

00:45:43.880 Wow, that's crazy.

00:45:46.140 welcome to the dystopian nightmare that people have been talking about for years now

00:45:50.680 oh my gosh okay what do you think is going to happen like how much power does that tiny unknown

00:45:58.580 what would you what would you refer to that not bored what did you call it what was the acronym

00:46:04.820 yeah the that that that government body the yeah the national security commission on ai

00:46:11.380 Okay. And so is that actually going to happen? Like, is that already happening?

00:46:18.960 If what Edward Snowden revealed is any indication of what's been going up and on until now,

00:46:25.460 it seems that's exactly what will probably happen. We're already seeing the kind of

00:46:30.680 infrastructure being put into place that allows for this. I saw someone from one of the central

00:46:36.480 banks in canada discussing digital id uh and you know saying how secure it is nothing is fully

00:46:43.280 secure but it allows the government to close your bank accounts we see what the tyrant in canada

00:46:49.620 has already done to certain citizens in their bank accounts so i i think the precedent has already

00:46:55.760 been set oh my gosh well this was more dramatic than i thought it was going to be so is there

00:47:01.960 i mean i guess i can understand if i not that i agree with it but i can understand if the

00:47:06.420 government is concerned about China just overtaking. And they're like, wow, we'll just do it, but

00:47:10.960 better, which is not. It's almost like the can't beat them, join them, but they're still trying

00:47:17.640 to beat them to a certain degree. And it's by going even further. What? Okay. What do we do

00:47:25.380 about this? Is there anything? And how can you be on the pro AI side here? AI does have a bunch of

00:47:33.340 benefits when it's designed for human values and not for the tech, the technocracy and the

00:47:38.860 oligarchs who are trying to control everything or sorry, mostly do control everything.

00:47:44.200 What can be done about that right now? Well, first of all, I think most people need to know 1.00

00:47:49.400 what's going on. And I think to a certain degree, people like yourself who spread these kinds of

00:47:54.240 messages or allow people to discuss these things, for example, I would suggest that people who are

00:47:59.760 interested in this particular issue right here, what's going on in the United States,

00:48:04.700 they should check out the great journalism being done by Whitney Webb, who's taken a lot of care

00:48:10.000 to documents like this and these issues. She's watching the watchers, essentially,

00:48:15.320 despite the risks that come with this kind of journalism. She's definitely an excellent resource.

00:48:22.700 Okay. Wow. That's terrifying. Okay. So we're basically screwed then though, aren't we?

00:48:28.500 I mean, because it's not going to be designed to help people.

00:48:32.640 That's not what's going to happen.

00:48:34.440 What do you think?

00:48:35.480 Not the systems that they're discussing.

00:48:37.600 No, not unless there's a real populist resistance against these things, but you cannot have

00:48:45.480 that resistance unless people know about it.

00:48:49.840 Wow.

00:48:50.440 George Orwell will be rolling in his grave if you heard this, because this is something

00:48:54.180 beyond 1984.

00:48:57.720 Wow.

00:48:58.500 Oh, that's terrifying. Okay. Okay. So how could, let's go in a more positive direction for now,

00:49:06.180 although we'll probably go back to that because that's kind of fun. But how could AI be designed

00:49:11.380 going forward so that it doesn't end up just destroying everyone's privacy and we don't turn 0.82

00:49:18.220 into China? Okay. Well, that's actually like two different particulars because there's what's

00:49:23.880 already being done at the government level in harnessing the power of AI. And then there's

00:49:28.200 more of that, the design level in how can we design safe, reliable and trustworthy and these

00:49:33.680 separate, slightly separate domains. If you're asking me about the former of what do we do about

00:49:38.640 what the government wants to do? Well, I think that that's at a more at a national level of what

00:49:44.140 do the people want, right? And like I mentioned, they have to know about what's going on before

00:49:48.420 they can actually say what they want. So these things are currently not transparent. I'm actually

00:49:53.160 surprised that the FOIA request was granted. I think that's almost crazy. But then again,

00:49:57.820 it does state that they want to do this openly. So it's not even one, it's not a secret anymore,

00:50:01.760 nor is it a conspiracy when we have the FOIA documents, you know, they're published online

00:50:06.420 and it even says within them, they want to do a publicly. But if we go to the design side,

00:50:11.600 I think there are ways of designing safe, reliable, and trustworthy AI. And I can give

00:50:15.940 you just like a few examples if you want. So there's this, there's something called Z-inspection.

00:50:23.160 And it's a process that has the potential, I think, to play a key role in the context of the new European Union's AI regulation that I mentioned.

00:50:32.000 And essentially, it's an initiative of a team of international independent researchers, and I work with them, which takes real-life case studies of AI, such as certain AI systems in the healthcare domain, and they apply this process as a way of achieving trustworthy AI.

00:50:49.300 And it's a general inspection process.

00:50:51.860 So we look at something that already exists, an AI system that already exists, and we look

00:50:56.380 at how can we apply ethical AI to them in a variety of domains like business, health

00:51:03.480 care, public sectors.

00:51:04.980 And we use the definition of trustworthy AI that's given by the high level European Commission's

00:51:10.900 expert group on AI.

00:51:12.160 But we actually even go beyond the key elements that the European Union asks for.

00:51:17.380 and we provide an additional two more. One of them is assessing if the ecosystems respect

00:51:22.980 the values of Western European democracy and whether we can avoid a concentration of power.

00:51:30.320 That's what we just discussed, the concentration of power. So for example, we took a close look at

00:51:37.300 a deep learning based skin lesion classifier to see how we can use the process to co-design

00:51:43.560 in such a way as to make it trustworthy. So the system explains decisions made by

00:51:48.120 deep learning networks, analyzing images of skin lesions. And this is an example of what we call

00:51:54.460 computer aided diagnostic systems or CAD systems. And they've been shown to yield higher sensitivity

00:52:01.600 and specificity in diagnosing malignant melanoma by analyzing dermatostopic pictures compared to

00:52:09.720 a well-trained dermatologist. So they actually show efficacy even beyond. Yeah. But, and this

00:52:15.500 is usually the cause of hesitancy when employing these types of AI systems like this is the

00:52:20.380 acceptance of these CAD systems in real clinical setups is actually quite limited. So it hasn't

00:52:25.780 been uptaken at large scales. And that's because their decision-making processes, how it actually

00:52:31.240 makes decisions are more often than not largely obscure. And that's because there's a lack of

00:52:38.440 explainability it's what we call black box we don't know what's going inside the black box

00:52:43.100 cool yeah but you know we argue that in order for in order to code design trustworthy ai with

00:52:51.300 a holistic approach means that we need some unique aspects aspects in the structure and

00:52:57.600 design process or more specifically we need an interdisciplinary team of experts coming from

00:53:03.060 the domain of the ai application so in this case with the cad system that would be the field of

00:53:08.280 healthcare, um, as well as other fields like the technical side, the legal side, the social and

00:53:13.260 ethical side. So we need an interdisciplinary team working together to uncover what the issues are

00:53:19.280 and then working with the design team in order to address these issues. So it's a collaborative

00:53:23.300 co-constructed effort. And that's just one of the many frameworks that has been proposed for

00:53:28.400 actually achieving reliable, safe, and trustworthy AI. Okay. So that makes sense.

00:53:34.780 That's interesting. Yeah. So AI can be used in medical set us medical settings much more

00:53:42.340 effectively, right? Because they probably have better pattern recognition. They can just contain

00:53:46.240 more data than people. Yeah. So AI systems in healthcare definitely can be used to assist

00:53:52.100 healthcare workers in their day-to-day work, what you would consider perhaps to be slight drudgery,

00:53:57.080 right? Or routine things. And ideally it could be leveraged to enhance preventative care and

00:54:02.020 quality of life, as well as produce more accurate diagnoses and treatment plans.

00:54:07.540 But recently, there's been issues regarding this. The proper design of systems could

00:54:11.980 significantly increase the quality of medical care. So for example, back in 2017, there was

00:54:17.000 actually a study in the Journal of the American Medical Association where an AI model was able

00:54:22.400 to accurately diagnose breast cancer at a higher rate than 11 top pathologists. So it definitely

00:54:29.280 is possible great and there's significant benefits yeah yeah yeah okay so is that

00:54:36.500 why hasn't that been put in you said this you talked about it a little bit but why hasn't that

00:54:41.740 been put in the medical system is it just slow unfortunately there has been um some pushback

00:54:50.060 from the the medical field given their the general hesitancy of ai systems more broadly and although

00:54:58.600 it is true that medical, um, although it's true that AI systems, particularly within popular media

00:55:07.220 have shown to be less than, uh, accurate in certain circumstances, uh, certain medical experts are,

00:55:15.080 you know, less than happy to fully trust the system. And when, when you don't, when you ask

00:55:20.760 the system or when you look at the system, when a medical expert, like a doctor goes, okay,

00:55:24.260 it showed me that there's this malignant melanoma i want to know why why the system told me this

00:55:31.360 particularly if it's not 100 accurate we're never going to get that 100 actually what progress not

00:55:36.800 perfection so there's always going to be a certain amount of doubt and when there is a amount of

00:55:42.680 which is fine and you can't figure out why the system made the decision that did the black box

00:55:47.400 then you're far less likely to actually trust the system okay that makes sense

00:55:53.980 do you view the human brain as something that can be replicated with ai

00:55:58.700 so replicated fully no because it depends exactly what you mean our brain is wetware

00:56:09.880 but if we're talking in silicone not exactly so what you're perhaps asking is about deep learning

00:56:16.440 and deep learning is actually a further subfield of machine learning and it's usually understood

00:56:22.380 and inspired by how the human brain functions and uses what is often referred to as artificial

00:56:28.700 neural networks. So, you know, in order to simulate how our brain works, but in silicone.

00:56:34.600 So essentially it works by learning from examples in a similar way that actually humans learn new

00:56:41.180 things. They're actually, yeah, they're actually pretty good systems and they work great at

00:56:46.360 classifying images actually quite accurately in text and sounds, and they can even perform at

00:56:51.640 levels better than humans can at doing that task. So one of their main benefits, like machine

00:56:57.060 learning more broadly, because deep learning is a type of machine learning, is that these systems

00:57:01.300 are capable of computing huge, vast quantities of data that would be impossible to compute with

00:57:08.080 traditional methods in a coherent way. And more traditional techniques of machine learning are

00:57:13.560 pretty linear, whereas deep learning neural networks are stacked in hierarchies of different

00:57:19.200 levels of abstraction. So imagine we want to teach like a deep learning system, what a car is.

00:57:25.040 Well, how does a kid, like a child learn what a car is and what a car is not? Because that's also

00:57:30.900 part of learning what a car is, what it isn't. Well, the child can, you know, point at the car

00:57:35.580 and say, that's a car. And then the parent can either say, yes, that's a car or no, that's not

00:57:40.340 a car. And as a child continues to point at different cars, they're able to abstract from

00:57:46.980 all those experiences, the features of what a car isn't and what a car is. And I guess you say

00:57:53.740 that's kind of like the platonic car-ness in essence that they're extracting. Deep learning

00:58:00.740 systems work in this way, very similar to the way we would learn or we tried to mimic that in

00:58:07.080 silicone. Okay. And do you see that progressing to the point where it's comparable to a human?

00:58:16.420 No. Uh, I, I have a hard time with, uh, the concept of AI, uh, being anthropomorphized

00:58:26.700 too far. Uh, the classic Terminator, you know, AI, a threat to humanity, right. Taking over,

00:58:33.360 We see the images all the time on magazines and movies.

00:58:36.160 I love those movies, right?

00:58:37.880 But I have a really hard time with that particular way of conceiving of AI.

00:58:46.300 And if you want, I can tell you why.

00:58:47.960 Yeah, yeah.

00:58:48.300 I do want to know why.

00:58:50.140 Well, you know, if we take that definition of AI that I gave at the beginning as our starting point,

00:58:55.260 it's hard to imagine that the thing, that that type of thing is taking over anything

00:59:00.140 or being a significant global threat.

00:59:02.380 And usually the discussions of AI as being as a threat to humanity or, you know, on an existential level is referring to speculative artificial general intelligence or artificial super intelligence, something out of Hollywood.

00:59:15.780 And Nick Bostrom, actually, a professor at the University of Oxford, in his book called Super Intelligence, talks about this.

00:59:21.900 But despite, you know, the popular media having their fun with with this, it's not a really realistic picture of what AI is.

00:59:29.200 It's just not. It puts the wrong idea of what AI is in people's minds. And this kind of embodied

00:59:35.900 mechanistic God that can do anything because it's more intelligent than us. But we're so far removed

00:59:41.820 from that kind of scenario that talking about it amounts to nothing more than having fun. And I

00:59:47.040 think it's actually worse than that. Talking about AI like that does more harm than good.

00:59:52.940 Like for one, there's the risk that the public will see these futuristic ideas of AI, whether

00:59:58.080 good or bad, and create pretty big expectations in their minds about what AI is and what it can do.

01:00:04.820 And this is actually pretty bad for AI research because it can lead to another, what would be the

01:00:09.020 third AI winter. And these are freezes in serious AI research and development, hence the winter part

01:00:17.020 and the freezes that come as a consequence of overly ambitious promises that AI can yield

01:00:22.880 with the actual results being pretty humbler than expectations. And this happened in the mid-70s

01:00:28.680 and in the late 80s, early 90s. And it's the perceptual loss in faith of what AI can do

01:00:35.400 that leads to the actual loss in funding to do this research. And we're living in kind of like

01:00:41.820 this renaissance of sorts where everything is AI, which is, of course, nonsense. But what it does

01:00:47.500 reveal is that there's, you know, pretty big global interest in the capabilities that AI can

01:00:52.500 offer us. Still, you'd think we'd, you know, learn our lesson by now, but these kinds of exaggerated

01:00:57.540 illustrations of AI in popular culture can eventually blow up in our faces. And there's

01:01:04.820 also another danger and worrying about the risks of artificial general or artificial super

01:01:10.300 intelligence leads us to be distracted from the actual issues that AI systems pose. And the most

01:01:17.920 common, of course, is bias in these systems. And bias can exist across all levels and all stages

01:01:23.820 of how an AI system is conceived, trained, built, and used. So we have to address these concerns

01:01:30.120 head on if we want to build AI for social good, AI for people, rather than have to constantly

01:01:36.860 rated headlines about how AI systems reinforce discrimination or existing spheres of disproportionate

01:01:44.100 influence and power. Okay. Well, I feel somewhat better with that answer. I can get rid of the

01:01:51.820 Arnold Schwarzenegger taking over the world type robot. That's good. But the real problem is people

01:01:58.280 who get their hands on this and then everything's destroyed because of what they do with it,

01:02:03.020 which is equally as scary. Yeah. It's that dual use of technology.

01:02:08.160 Not everyone uses a technology the way the designers originally intended.

01:02:13.240 Yeah. Like the thing about the government that I mentioned.

01:02:16.920 Yeah. We're going to get back into that. Cause I want to talk about exactly what China is doing.

01:02:20.960 But before we do that, what, what did you mean by built in bias? Can you give some examples

01:02:26.960 of what that could look like? Sure. Yeah, definitely. So let's take the mortgage

01:02:32.980 industry, for example. So, you know, obviously how an AI functions in the real world is a

01:02:38.880 consequence of, well, the people who made it, the designers, its creators. So there's actually

01:02:43.600 numerous examples of how a flawed algorithm leads to, well, faulty outputs, right? If it's faulty

01:02:50.560 from the beginning, it's going to be faulty, you know, when it's actually being used. So when the

01:02:54.440 designers create software programs, these programs are constrained within specific parameters.

01:02:59.780 And it's the nature of those parameters to be the subject of interpretation.

01:03:04.920 So, for example, if someone from a low income neighborhood submitted a mortgage proposal, there's good chance that that proposal would be classified by the system as high risk.

01:03:14.560 And as a consequence, the system will recommend a high interest rate in order to account for that risk that the firm is going to have to shoulder.

01:03:22.900 And the unfortunate thing is that how these neighborhoods are divided actually has little

01:03:29.100 to do with how income is generated and more about the social economic factors.

01:03:34.000 So things like race, for example.

01:03:36.240 And these systems aren't capable of recognizing this bias.

01:03:39.940 It can only account for and compute the information based on its design.

01:03:44.700 And this usually ends with a firm's agent lacking proper justification when they have

01:03:50.680 to reject the client.

01:03:51.520 And this is the black box again. Why was I rejected? I don't know. And, you know, this can

01:03:56.940 lead to what I would say is deleterious impacts on the firm's reputation, given that people can

01:04:04.060 quickly spread the word of how they were rejected, but they weren't given a clear reason as to why.

01:04:10.240 Yeah, that wouldn't be good. If you want to see some great bias and facial recognition,

01:04:15.040 that's also a fun one. There's a whole host of examples. So like a few years ago in New Jersey,

01:04:20.380 It was a man that was arrested by police and actually spent time in jail when they followed what they called a high profile comparison from a facial recognition scan that they extracted from a fake ID that they found at the crime scene.

01:04:34.740 And the machine learning algorithm that they used in the facial recognition system learned to identify a person based on their facial features.

01:04:43.180 And these programs have been shown in a number of studies to be less than reliable, to say the least.

01:04:49.020 And they're certainly less than accurate in their ability to properly identify particularly black and Asian faces.

01:04:55.780 I remember there was one where Google want to Google facial image classifier classified black males as gorillas in the same category.

01:05:04.840 Oh, my gosh. Yeah, that is not what we want.

01:05:08.560 No, definitely not.

01:05:09.660 okay so could that go i want to get back into china what china is doing but could we end up with

01:05:19.120 ai systems that are completely you know the population is completely entrenched in this and

01:05:26.640 then you end up getting these kind of mistakes where you identify people improperly and then

01:05:32.740 that ends up ruining lives because ai said oh no you know with 99.9 certainty this is the guy you

01:05:39.340 want. And then that's it. Life over. This is the classic example. You essentially just defined

01:05:46.760 what we call the Colin Ridge dilemma, which is we can't understand the full impacts of a technology

01:05:53.840 until it's ubiquitous and therefore pervasive. So it becomes kind of like part of the social

01:05:58.580 technical infrastructure around us. So it's like normalized. But once those things emerge,

01:06:04.300 it's so entrenched that how do we actually go back? Because it's so fundamental because

01:06:08.440 technologies don't come from nothing, right? They build on previous iterations and then they

01:06:13.600 slowly become normalized as they become diffused in society. So that's one of the reasons why we

01:06:18.820 have to adopt a design approach from the get-go that does at least two things. So one of them is

01:06:26.140 that we have to design technologies for human values. And that means we have to enroll as many

01:06:30.680 stakeholders as possible. That's not only the designer who goes, I'm going to build this system

01:06:35.080 I mean, you know, the CEO goes trying to make profits.

01:06:37.480 Those are important stakeholders.

01:06:38.740 Those are direct stakeholders.

01:06:40.280 We also have to bring the publics in because in the end, the public is going to be the

01:06:44.000 users, the environment, whatever it may be, are going to be the ones that are fundamentally

01:06:48.720 impacted by these technologies.

01:06:50.940 So they have to also be part of the design program itself in designing these technologies.

01:06:55.500 What are their values?

01:06:56.460 How can we design for them rather than sidelining them or as an after fact?

01:07:00.660 And I think that's a good rule for most technologies, if not all. But AI, we have to

01:07:06.780 add something even further, is a good way to conceptualize how to design AI for social good

01:07:12.880 is to get away from this idea that AI is simply a product. A designer makes something, then they

01:07:19.320 sell it, and then they just wash their hands of it. No more of that. We have to build in

01:07:24.680 full lifecycle monitoring. So because many of these technologies are black box, we don't know

01:07:30.220 why they do what they do in many cases, even if they could be trustworthy, that means that when

01:07:35.000 we throw them out in the world, they're most more likely than not to display emergent behaviors that

01:07:41.260 were unforeseen or even unforeseeable. And if some of those behaviors that are unforeseen or

01:07:48.060 unforeseeable aren't ones that we like, we need a way to one, recognize them and to have the

01:07:53.120 mechanisms to kind of rein things in, to pull them back into the design space, to do another

01:07:59.000 iteration of design, a firmware update or a software update, whatever it may be. It's no

01:08:04.200 more, let's build this, let's sell it and we're done. That's not going to work anymore given that

01:08:08.560 these are technologies that really do change over time. I'm less confident about humanity's

01:08:16.840 ability to do this than I was at the beginning of this conversation, just so you know.

01:08:22.000 Okay. Let's get back to China. I've got a couple of questions, but one, if the US

01:08:28.700 doesn't do this kind of competitive thing they're doing where we're going to do that, but better.

01:08:35.460 What's your opinion on the likelihood of China just being so effective at this that they do

01:08:40.040 become more and more and more powerful? So are you asking me more about what's going to happen

01:08:48.860 with the future of China? Because essentially you described what happens when China has

01:08:53.740 essentially global dominance uh well if the u.s uh if the u.s doesn't like if they decide okay no 0.91

01:09:01.460 we don't want to go the same route as china do you think china is just going to become world

01:09:06.340 dominant yeah yeah yeah for sure we see that already in europe you know europe uh we've been

01:09:11.720 criticized uh that's where i am right now we've been criticized as having the religion of privacy

01:09:16.260 Right. Yeah. You know, the the the new European Commission's regulations on AI once passed and ratified does ban certain types of AI, things like facial recognition, particularly in high risk environments like law enforcement because of issues of bias and so on.

01:09:34.520 But there's different levels of risk assessment. And then with those different levels, there's also different levels of regulation or ways to actually address and govern those technologies.

01:09:43.780 So we're taking a very kind of middle road, cautious approach.

01:09:47.380 The United States, if we are to believe these documents, if they're actually going to act on these documents that we've seen on the suggestions, it seems that they do want to surpass China. 0.81

01:09:56.040 And the hypothetical that they don't, I imagine just the continual fall into decadence of the United States and Europe becoming essentially the playground for China, where they come to visit because it's beautiful. 0.90

01:10:09.120 And then they go back. 0.95

01:10:11.380 Hmm.

01:10:12.580 Okay.

01:10:13.100 Okay. So then what, that's tricky.

01:10:16.580 It's very tricky because when you have someone who doesn't play by the same rules, what do you do?

01:10:22.240 I don't have an answer for that. It's a really tricky one. I'm not going to pretend to solve the China issue.

01:10:28.240 Yeah, definitely not. Okay. Then. Yeah. So what exactly, can you give me some details about what they're doing over there?

01:10:36.500 Yeah. So the Chinese state ain't no stranger to the powers of huge bodies with equally huge influence like we have back in the United States, working with the intelligence communities, Google.

01:10:49.680 So it's no wonder that seeing the influence that big tech has on the West, China would crack down on how big tech uses its weight to influence its users, like using recommender systems and other techniques that big tech giants have used to rake in the money and monopolize their power.

01:11:09.920 But China won't suffer any other monopolistic powers. That's for sure. And they state that

01:11:16.880 they have zero tolerance for corrupt behaviors, particularly those giants that actively try to

01:11:22.300 consolidate power and capital, so those trying to become monopolies. Obviously, the thresholds for

01:11:28.440 this are super vague, and we don't know how it's actually being implemented in China on a day-to-day

01:11:33.440 basis. It's quite secretive, and then we can't really trust what comes out. Still, we see that

01:11:38.820 the consequences in, in the West of the unhindered growth of these giants, we already talked about

01:11:44.760 that and how they become political actors with their own agendas, uh, using the users as currency.

01:11:51.160 You know, there's the same, uh, if, uh, if you're not paying anything, for example, Facebook, well,

01:11:55.020 then you're the product. And, you know, that's not to say that we should adopt the China,

01:11:59.160 the Chinese style of crackdowns, whatever that entails. Uh, I'm sure that would lead to a sort of 0.98

01:12:05.700 equally if not more detrimental overcorrection that's usually what happens when people react

01:12:11.460 as an overcorrection but i you hit on it something has to be done i just don't know exactly what the

01:12:19.400 best ways to separate these powers in the west are without risking that you know extreme overcorrection

01:12:25.200 of just you know entirely dissolving them hmm okay wow i'm gonna let other people deal with

01:12:35.140 that problem i think i'm gonna i'm gonna deal with something else that sounds a little too

01:12:38.680 complicated definitely that yeah that seems like a real problem can you after can you send me over

01:12:46.000 links to those documents we were talking about earlier of course you said they're online okay

01:12:50.620 i'll post those in the show notes yeah that's a great way this is this is the only way right

01:12:55.100 it's we got to get people to realize this is to push back small acts of rebellion right not

01:13:00.420 expecting revolutions, but people say, no, I, I won't continue to use Facebook. I won't be,

01:13:05.480 uh, their currency, right. I won't give them the data that they'll then sell, uh, then obviously

01:13:11.460 to the intelligence agencies also who have full access to these, uh, to these systems. So it's,

01:13:17.020 you know, making, getting people informed. And this is, this is one way.

01:13:21.580 Is there a way that you can, like, you can give somebody your data and have it not be a bad thing?

01:13:28.200 Like, is there any future where the government actually does something good with all the data instead of becomes tyrannical and something terrible happens?

01:13:38.260 Like, I mean, I assume that what they're doing and I assume that they don't think they're evil in China.

01:13:42.320 Right. They're like, no, this is for the benefit of all our citizens and there's going to be less suffering this way.

01:13:47.280 Right. But we're freaking out about it. So how do we know that it's bad?

01:13:51.020 Yeah, it's really, I guess, the point of view, you know, no other country in history has raised so many people out of poverty in a single generation. So you have to you have to ask the question, do the ends justify the means?

01:14:04.800 I'm highly skeptical of the centralization of power in general, particularly when everything is tied together in terms of our personal data, when it's also tied to our ability to be financially independent and autonomous.

01:14:21.380 And these are the discussions that are being had right now of centralizing our ability to transact with our vaccination status, with our perhaps what happens when you go to protest and the government doesn't like that protest.

01:14:37.800 Well, we see the consequence of that. Can you imagine when all these things are far simpler

01:14:42.200 because they're unified into a single system that they say it's going to be highly secure? 0.98

01:14:47.920 That's horseshit. Nothing is that secure. Someone can always get in, but that's not the, 0.97

01:14:53.280 I'm not worried about the people getting in. I'm worried about the watchers who are

01:14:56.420 actively using that. Yeah, no, that's definitely scary because what if things need to change,

01:15:03.200 right? What if as a society we're doing something wrong and you know, how do you even

01:15:07.500 discover something new like you probably know i've done a lot of work with diet which goes

01:15:11.920 completely against mainstream right so how and just using that as an example if we're all

01:15:18.360 centralized like this what happens when there's an outlier and they're like oh wait but maybe we

01:15:22.880 should do this differently it's like no cut that off easily does not does that not seem exactly

01:15:29.660 the point of these kinds of systems they keep saying it's going to be so much more convenient

01:15:34.660 as if it's so difficult for me to pull my license out of my wallet. It's so heavy to carry around

01:15:42.160 that I need a digital version of it. So now you have all my documents, you have the ability to

01:15:47.680 access and to block my bank accounts. And then that's also tied with whatever new vaccination

01:15:52.300 status you have, whatever that is. And then obviously, in terms of identifying people who

01:15:58.180 hold unacceptable views, right? Being targeted when you have the switch in your hand so much

01:16:07.380 easier. It seems like that's exactly the point. And we have the power in the East as essentially

01:16:13.660 an example of how that's being played out. Hmm. Okay. So say these documents do get

01:16:21.020 implemented. Is there, I mean, you must've thought about a timeline, like how long do we have before

01:16:26.620 we turn into another China. Wow. I would say that if it's implemented the Chinese style system with

01:16:37.360 these structural changes, as soon as those structural changes are made, you've passed

01:16:43.980 the event horizon, everything becomes centralized. Now, can we actually make that transition?

01:16:49.980 i'm not too sure i'm not too sure people realize no we're not going to do that i think that there's

01:16:57.780 a certain consciousness growing that's becoming skeptical of these times of measures it seems

01:17:03.340 that perhaps governments within the last year or so have pushed too far rather than taking an inch

01:17:09.180 they took a meter right so that kind of works against them you take one step at a time and

01:17:15.480 you're able to do that. This was hidden. Like I said, I'm not sure if the FOIA request was granted

01:17:20.400 purposefully in order to acclimatize people to the idea, right? You saber rattle the people

01:17:27.520 in the East and you go look at if we don't even try doing this against them, well, then we just 0.99

01:17:33.620 become the playground for China. No simple answer, but it is scary to consider what would happen 0.99

01:17:40.960 because, uh, I don't trust the, the next seas of power. Hmm. Okay. Well, thank you very much

01:17:51.200 for coming on my podcast. That was much more interesting. I appreciate it. It was fun.

01:17:56.840 Yeah, that was fun. Um, if people want to find you or any of your information online,

01:18:01.900 where should they go? Yeah. You can just toss my name in Google. You'll probably find, uh,

01:18:05.960 My website, I probably use DuckDuckGo even better.

01:18:08.940 And my institutional sites, if you're interested in the academic work.

01:18:12.320 But if you're interested in my forthcoming book, which was written for a more general

01:18:15.860 audience, like I mentioned before, it's called Design for Death, Controlling Killer Robots

01:18:20.100 About Autonomous Weapon Systems.

01:18:21.760 And you can check that out on TriVent Publishing's Ethics and Robotics book series.

01:18:26.880 But if you want to follow me on Twitter, you can do so with the handle at Stephen Umbro

01:18:30.440 and the IAT with the handle at IAT.

01:18:33.420 okay did you say your book was about what did you say it was about autonomous killing machines

01:18:39.540 yeah autonomous weapons yeah okay can you just describe what that is right before we end it

01:18:45.420 what does that mean that sounds awful yeah it does sound it does sound awful so for years now

01:18:50.800 this is kind of the terminator scenario but the book tries to get away from that terminator

01:18:55.340 scenario more generally because it gives a bad idea of uh what our autonomous weapon systems are

01:19:01.820 And there's this trend towards more and more autonomy within the warfare theater.

01:19:07.800 And we've seen that since the Middle Eastern wars, the trend towards more and more drone strikes.

01:19:13.520 It seems that's the that's the norm rather than the exception.

01:19:17.740 So the question of what is a fully autonomous weapon system is, well, no longer does the human actually physically control the system in terms of target selection and engagement.

01:19:27.800 but the system is able to identify targets and then engage them autonomously.

01:19:32.720 So that is what an autonomous weapon system is.

01:19:35.620 And of course, there's different levels of autonomy,

01:19:37.400 just like there's different levels of autonomy, even within AI.

01:19:40.480 So these are the questions that I explore.

01:19:43.160 I try to find a middle path because the argument in the book is that

01:19:47.740 if we just try to ban these outright, like go no autonomous weapon systems whatsoever,

01:19:52.460 then those states that are designing them simply won't sign the international agreements

01:19:57.360 and use them anyway.

01:19:58.900 But if we find the middle path of saying,

01:20:00.380 maybe these ones we shouldn't do

01:20:01.880 and these ones are acceptable,

01:20:03.520 if we design them for values,

01:20:05.740 like the laws of armed conflict,

01:20:07.460 then maybe those state actors

01:20:09.420 who have an interest in developing

01:20:10.840 these types of systems

01:20:11.820 will be more likely to sign on

01:20:15.080 to these international treaties

01:20:16.400 and respect them.

01:20:18.900 Okay, I like that.

01:20:20.300 It seems like a good thing to write.

01:20:22.720 Hopefully the bad guys read it.

01:20:25.880 Okay, thank you again

01:20:27.180 for coming on. That was enjoyable.

The Mikhaila Peterson Podcast - April 05, 2022

144. Is Artificial Intelligence A Threat To Human Existence？｜ Opposing Views

Episode Stats

Transcript

The Mikhaila Peterson Podcast - April 05, 2022

144. Is Artificial Intelligence A Threat To Human Existence？ ｜ Opposing Views

Episode Stats

Transcript

144. Is Artificial Intelligence A Threat To Human Existence？｜ Opposing Views