Latest / Uncomfortable Conversations with Josh Szeps / Toby Walsh: Your Wild A.I. Future

Transcript

Toby Walsh: Your Wild A.I. Future · Uncomfortable Conversations with Josh Szeps

• 0:00
  Hey humans, welcome to Safe Space for Dangerous Ideas and what idea is more dangerous at the
• 0:08
  moment than artificial intelligence and its consequences for humankind, for you, for me,
• 0:15
  for civilization. Today's guest knows more about the subject than just about anybody else.
• 0:19
  Toby Walsh runs the Artificial Intelligence Lab at UNSW, the University of New South Wales,
• 0:24
  which is one of Australia's big Ivy League sandstone institutions. Toby was invited earlier
• 0:31
  this year to sign that famous letter that was signed by a bunch of AI luminaries, the world over.
• 0:38
  Warning policy makers and the public that artificial intelligence poses a threat to humankind
• 0:45
  on a level of pandemics or nuclear weapons. He declined to sign that letter, which is interesting
• 0:52
  in itself. But back in 2015, he authored a letter that was signed by 20,000 of his colleagues
• 0:59
  in curaging governments not to use AI in warfare and banning militaries from doing so. In fact,
• 1:06
  he's been influential in pressuring the Australian government to take a lead on this at various
• 1:10
  international forums. He's written three books about artificial intelligence. It's a live
• 1:15
  artificial intelligence from the logic piano to killer robots, a book called 2062, the world
• 1:21
  that AI made. And most recently, machines behaving badly, the morality of AI. A few people think
• 1:28
  about this subject in as knowledgeable and impartial away as Professor Toby Walsh, please enjoy this
• 1:34
  conversation with Toby Walsh.
• 1:46
  Should I turn my phone off, do you think? Would that be a professional thing to do?
• 1:49
  Well, unless you want machines to be listening. Are they? That's not true, right?
• 1:54
  When people have a myth that I was talking about kayaking and then all of a sudden the next day,
• 1:59
  there was a kayaking thing. So my phone's spying on me.
• 2:03
  They are spying on you. That's true, but not in the way that they're not directly listening to you.
• 2:08
  Well, some of them are listening to you. I mean, the reason that Siri always answers or
• 2:12
  Alexa always answers is because it's always listening. I mean, as a good answer. But actually,
• 2:16
  they're probably throwing away that data. But it's not the case that they're listening and then
• 2:22
  sending you adverts based upon what they're listening. They're using indirectly this information.
• 2:28
  So they'll look at the IP address of where you're connecting. So someone was saying
• 2:32
  their wife went to the dentist recently and then suddenly they were getting all these
• 2:35
  denture adverts and she'd been talking about with the dentist about getting dentures.
• 2:40
  They were probably just using the fact that they worked out that she was connected to the
• 2:43
  Wi-Fi outside the dentist. She's a certain age. That's an age in which you start sending people
• 2:49
  those sorts of adverts. Right. And then there may have been a map app that was selling data as well
• 2:53
  that knew that she was at a dentist's office. I mean, in some ways, it's scarier
• 2:57
  that they don't have to hear us talking directly and can nonetheless piece together
• 3:01
  such a good picture of what we are up to. It is. And people don't realize, for example,
• 3:06
  I don't know what it is in Australia, but in the US, Google tracks 70% of all credit card
• 3:11
  transactions. Not just the ones online. That's the offline ones where you physically hand
• 3:16
  the credit card to the salesperson. Google gets over 70% of that data that they buy.
• 3:22
  To do what with? So they charge people for the adverts so that they know that they
• 3:28
  just bought something. You bought something that they served the advert and then they go to the
• 3:32
  the retailer and say, we gave this digital advert and now they've gone and bought that.
• 3:36
  We want a commission on that sale. So let's talk about there were so many places to start with.
• 3:41
  I think we should ramp up in order of worrying aness from the most plausible concerns to the
• 3:49
  existential catastrophes, maybe the simplest way to structure this conversation.
• 3:53
  Can we just get our heads around first? What we're talking about when we're talking about AI?
• 3:57
  Because to the AI skeptics or the people who think there's nothing really to see here and we're
• 4:01
  all just running around with our pants on fire for no reason, they say things like worrying about
• 4:08
  AI sort of ruining human civilization or taking over our minds is a bit like worrying
• 4:17
  in the advent of the automobile that cars would become better sprinters than humans would.
• 4:22
  Like only humans sprint cars that are there for a tool or that calculators would take over
• 4:28
  mathematics. We have machines that are doing their thing. We're going to teach machines to do
• 4:33
  increasingly sophisticated things and they're going to get increasingly good in their own lane.
• 4:38
  But it's foolish to worry about that spilling into some kind of general intelligence that would
• 4:43
  threaten our intellectual supremacy. What do you say to that?
• 4:48
  Well, I'm not sure why we're so keen to have intellectual supremacy. But there are a couple of
• 4:54
  things I think this technology uniquely bring to the table. One is speed and scale. They can do
• 4:59
  things that humans just could not have done ever in the past because they are computers and they
• 5:04
  could work at electronic speeds and they can work at scales that humans just can't comprehend.
• 5:09
  And then the other, and this comes to some of these concerns, is that they have
• 5:14
  increasingly amounts of autonomy, the ability to act on their own, make decisions independent or
• 5:20
  largely independent of our telling so that when you get in your autonomous car, as you will in
• 5:26
  the near future, you will say, take me home. And that's the only decision that you make. The car
• 5:31
  makes all the other decisions as to what it's going to do. And if it runs into a pedestrian,
• 5:35
  that's why we're in interesting moral, legal territory. Why is it a worry that the car makes a lot
• 5:42
  of sub-decisions as long as the meta decision of getting me home is uncorrupted?
• 5:48
  Because some of those decisions might be life or death decisions. They might be
• 5:53
  turn a corner as a pedestrian standing in the middle of the road. You have to make that split
• 5:57
  second decision or the cars to make that split second decision. Shall I run into the pedestrian
• 6:02
  killing the pedestrian? Shall I run off the road, possibly into the brick wall, killing the
• 6:06
  occupant of the car? I'll take the car that kills the pedestrian, please. Thank you, Toby.
• 6:12
  You can imagine a car that sells a self-driving car. You can imagine Volvo pitching its cars as being
• 6:18
  the safer alternative where the programming is tweaked to bias the occupant of the vehicle
• 6:24
  more than other cars do. Yes. Interestingly enough, they've actually passed a law in Germany to tell
• 6:29
  car manufacturers what you have to do, which is you can't. You can't make a decision other than
• 6:33
  just on purely the number of people who are going to die. Wow. You can't make a decision based
• 6:38
  upon age or any other criteria. The utilitarians of one in Deutschland.
• 6:43
  Yeah. Excellent. I mean, it's not that cars are actually making those little decisions yet,
• 6:47
  but they will, possibly in the future. So we already live with a whole bunch of AI. I mean,
• 6:51
  in the sense of opening up my phone and calling an Uber, there's a lot of, I mean, this is not
• 6:58
  a general intelligence, but there's an intelligent network, even though we still have humans driving
• 7:02
  the cars, there's an intelligent network pairing me with drivers and handling the building and
• 7:09
  figuring out the routes and so on. Extrapolate from the world we're into the one that will exist in
• 7:15
  some decades where the AI is of a kind that would currently boggle our mind. What does it look like?
• 7:21
  I think, well, I think that's an exciting word to think about. If you think about the world that we
• 7:25
  live in today, that's the product of human intelligence. And I think it would be somewhat
• 7:30
  conceited to think that human intelligence was the maximum you could achieve. I mean,
• 7:33
  what is the maximum of human intelligence? There are smarter people or less smart people in the room.
• 7:39
  So, and there are good reasons to suppose that machines could be smarter than us. They're going
• 7:44
  to think faster than us. They're not going to be as forgetful as us. They can have more memory than
• 7:48
  us. And already in the narrow-focused domains where we've got machines to do intelligent things,
• 7:53
  whether that be playing chess, translating tweets from Mandarin into English or reading
• 7:59
  interpreting X-rays, and they typically do it not just as well as humans, but typically much
• 8:04
  better than humans. And so, once we put all those things together, I suspect, by 2050 or whenever
• 8:09
  it is that we have machines that are as smart as us. They will not stop at our intelligence but
• 8:14
  become smarter than us. In a general way? I think at the end of the day that we are machines
• 8:21
  ourselves with biological machines, and therefore we're going to build machines that will match
• 8:26
  pretty much all of our capabilities. With, of course, the exception that we don't know whether
• 8:30
  those machines are going to be conscious, you know, have our emotions. In a strict scientific
• 8:36
  sense, they won't mean computers are electrical devices. They're not biochemical devices,
• 8:42
  and our emotions are biochemical. So, it's an interesting... Right, but that sort of
• 8:47
  taughtology, isn't it? I mean, they're sort of defines the problem of whether a computer will
• 8:50
  let it be self-aware, but we don't know whether they have some sort of consciousness, we don't have
• 8:57
  any way of... We don't consciousness is very poorly defined. So, it's hard to know what it is,
• 9:01
  whether you don't know how to measure it. But I mean, you could make the same argument about
• 9:06
  about other primates, for example. You could say, well, of course, chimps don't have consciousness
• 9:11
  because consciousness is something that's defined as being human and they're not human,
• 9:14
  so they don't have consciousness. Well, I think the general consensus within
• 9:19
  zoology and primate, as zoology in particular, is that primates, most primates do have consciousness.
• 9:26
  Indeed, it's not restricted to... No, but I mean, the same total, the same rhetorical trick
• 9:30
  could be made to exclude them if you wanted to. I mean, why don't we just dig into this
• 9:34
  since it is an interesting field of... Yes. ...of the philosophy of mind, I suppose, as well.
• 9:39
  If you're religious, you may think that there's a sprinkle of pixie dust that makes the humans
• 9:46
  and animals, whichever we can draw the boundary wherever we want to in the animal kingdom, but
• 9:53
  that gives us a spark of life and self-awareness. If you're not religious and you're a materialist
• 9:58
  and you think that our self-awareness and our sense of the lights being on inside us
• 10:05
  is a consequence of information processing that's going on in the gray, in the soft gray computer
• 10:11
  between our ears. Yes. Is there any reason why replicating that level of
• 10:16
  sophistication of information processing in a computer wouldn't yield a similar or dissimilar
• 10:24
  sense of being alive? It seems plausible. I mean, this is one of the interesting experiments we're
• 10:30
  running in building AI is to see is consciousness restricted to biology or is it something that we
• 10:37
  could reproduce in silicon? We don't know the answer to that question. What's your hunch?
• 10:44
  If we're lucky, it isn't. Right. Well, maybe, but you also don't want to be ruled by...
• 10:50
  I think the morality of being subjugated by machines would be different if the machines were
• 10:55
  conscious versus unconscious. It would be very macabre if we were ultimately done in by mindless
• 11:01
  bots, but if we were ultimately done in by machines that were capable of seeing vast horizons
• 11:06
  of possibility beyond our own, then that would be some consolation. Perhaps. I mean, if they're not
• 11:13
  conscious, then we won't have to give them any rights. We wouldn't have to worry about them. We
• 11:19
  can turn them off today. We don't care. I can go back after this podcast and take my AI robots
• 11:25
  apart in my laboratory, diode by diode. No one's going to care. There's no suffering happened in
• 11:30
  that process, but in the future, if the machines are not any smart but conscious, then we will have
• 11:36
  to start worrying about those sorts of things. I wonder what it'll do to us to be interacting with
• 11:42
  machines that seem like they're self-aware, whether or not we know if they are. Hopefully,
• 11:47
  it will be a moment of humility. Hopefully, we'll realize that we were guarding to the planet for
• 11:52
  a while. Maybe there's some great intelligence. At the very least, they're going to be a fantastic
• 11:58
  tool to transform the world we live in, to solve many of the challenges that face us today,
• 12:03
  to cure all the diseases that we think of, to find cures for all the cancers, to perhaps
• 12:10
  live on the planet in a more sustainable way, to eliminate poverty. The world that we live in
• 12:16
  today is the product of our intelligence. If we could imagine a greater intelligence,
• 12:19
  could we imagine what spectacular world that might be? Just before we get to the spectacular world,
• 12:24
  I still want to pursue this question of them being self-aware or then being conscious,
• 12:29
  or they're being something that it's like to be a computer. Would there be any way
• 12:35
  for us to tell? I sometimes wonder, if you were interacting with a computer and instead of
• 12:41
  it just seeming like a human, it started expressing the curiosity of a human. I mean,
• 12:47
  I don't think we're yet at the point, correct me if I'm wrong, where a computer talking to a
• 12:50
  human being has ever said, wait a second, earlier when you were talking about love, like what is
• 12:57
  that? What does that like? I mean, there have been those sorts of moments like Kevin Rousse from
• 13:02
  the New York Times coaxing or goading an AI into expressing human-like emotions. But will it
• 13:12
  change our estimation of what's going on inside the box when the box starts expressing
• 13:19
  weird curiosity in things that we're not asking at two? Well, we're very quick to
• 13:24
  ascribe feelings, emotions, consciousness to other things. I mean, people do this to machines
• 13:31
  that we know that are not. And so I think we're going to be easily fooled by this, indeed that's
• 13:37
  the subject of my forthcoming book, that we're easily going to be deceived. When I look, I mean,
• 13:44
  I don't actually have any guarantee that you're conscious. In fact, I have any guarantee that
• 13:48
  anything else around me, I don't actually have any guarantee I'm not in the matrix.
• 13:52
  But the simplest explanation for your behaviors and the fact that is to say, well, you look like
• 13:59
  you're made of the same stuff as me, you're behaving in the way that I would behave in these
• 14:03
  circumstances, the Occam's razor would demand, I say, well, you're probably conscious because
• 14:08
  I feel conscious and I project that onto you. That projection onto computers is not the simplest
• 14:14
  explanation. The simplest explanation is to say they're not conscious, they're just machines.
• 14:18
  And therefore, maybe they're going to be supremely intelligent, but lacking in that emotional
• 14:26
  life, that rich emotional life that we have. But is there any super-tructuring test that you
• 14:31
  would use to sway the balance? So let's concede that the balance of probabilities is harder to
• 14:38
  surmount with the machine than it is with another Homo sapien. But there must still be some point
• 14:44
  at which the balance of probabilities does tip in favor of the machine being self-aware in
• 14:49
  some hypothetical universe. What would that tipping point look like? It's hard to know what that
• 14:53
  tipping point is. And because we don't have a conscious meter and they will say the machines
• 14:59
  will shortly be saying all the right things, they don't hurt me, don't turn me off.
• 15:04
  I fall in in love with you. Indeed, many of them said these things already in special
• 15:09
  circumstances. And increasingly we're going to be forward by those, those please.
• 15:17
  I still can't tell. I'm trying to look between your eyes and glean what your actual true assumption
• 15:23
  is. Do you think that there'll ever be, that the lights will ever be on?
• 15:30
  I think it could go either way. I really, I can imagine worlds in which they do become conscious.
• 15:37
  That would be quite a challenging world because then we've created something in our likeness
• 15:42
  that is intelligent and deserves rights and we'll perhaps surpass this. Equally they may not,
• 15:49
  in which case they'll just be our machines, they'll be our servants. That perhaps be an easier
• 15:56
  world to navigate. Okay, let's talk about the good things that they can do before the bad
• 16:01
  things since you just rattled off a litany of potential good things like curing cancer and
• 16:06
  solving climate chaos or whatever it might be. How? Well, if we pick medicine, as an example,
• 16:13
  medicine is still stuck in the 19th century. I mean, we still give many women exactly the same medicine,
• 16:18
  despite the fact that biochemistry is completely different. And we now have this
• 16:24
  opportunity to completely personalize medicine. We can read people's DNA,
• 16:29
  and that's just cost a few hundred dollars now to read your genetics. We can write your DNA.
• 16:34
  CRISPR is a technology where we can actually change your DNA. It says there's something wrong with
• 16:38
  your DNA. Some bad genes in there. We can change them. And we need something to join those two,
• 16:43
  there's an information problem between those two things. We've got, it's fast amount of information,
• 16:47
  the hundreds of thousands of SNPs, the genetic letters that make you you and not me,
• 16:52
  and interpret what those mean and then work out how to change them in ways they're going to make
• 16:58
  eliminate those genetic disorders that you might have otherwise inherited. And AI is going to be
• 17:03
  the thing that joins those two dots. And this is not just about the the grunt work of crunching data.
• 17:11
  Right. I mean, we've had computers that can take in a whole lot of
• 17:16
  ones and zeros and you know, cobble together a genomic map of my DNA before. And then you're talking
• 17:23
  about machines that are capable of making interesting or innovative connections between.
• 17:28
  We're already starting to see that. So in the UK, for example, they have the UK gene bank,
• 17:32
  half a million people have had their gene sequence in the UK. And there's a database of all that
• 17:37
  information plus all of their phenotypical information, all about their physiology, their medical
• 17:42
  history. And that's gigabytes of data. Humans, it's not something that a human could,
• 17:50
  human attention span could comprehend. But you can run a machine learning algorithm over that.
• 17:56
  And you actually can find interesting correlations and connections is there. So now we can tell
• 18:02
  at birth or even in uterus, interesting ethical challenge there. How tall are you going to be to
• 18:08
  with an inch? Not particularly, you know, interesting, but not particularly medically useful.
• 18:12
  It depends how much you're like basketball. We can do we can tell what color eyes you're going to
• 18:16
  have, what color hair you could have. But we can also tell now whether you're going to be susceptible
• 18:19
  to bowel cancer. The third most deadly form of cancer. And most of the deaths of which are
• 18:24
  unnecessary because they could have been cured. But by the nature of bowel cancer, by the time you
• 18:29
  know you've got it, it was too late. But now we can tell at birth whether you're in that particular
• 18:34
  group. Now, genes are not destiny. There are still environmental triggers for those for those
• 18:39
  cancers. But at least now we can monitor you from birth and say, okay, well, you're going to
• 18:43
  we're going to do all the tests we can so that if you do start to get a cancer, a bowel cancer,
• 18:49
  we've identified it where it's possible to correct. And here we're talking about artificial
• 18:53
  intelligence that already exists, right? I mean, this is fairly simple stuff that we could be doing
• 18:56
  that we're not. One field of medicine where I hear which I hear being touted as being just over
• 19:02
  the horizon is in artificial intelligence cobbling together vaccines for novel diseases. How would
• 19:09
  that work? So again, it's it's searching for, you know, what's the what's the signature? What's the
• 19:16
  shape? You know, chemistry by chemistry is very dependent not only on the on the chemical formula,
• 19:22
  but also on the on the shape, whether the you know, whether you can dock around the the spike
• 19:26
  shape of the coronavirus or not. And so it's searching through all the billions of combinations of
• 19:33
  chemicals to find the one that's going to have the right effect and do things search. We already
• 19:39
  have. I mean, for example, the latest antibiotic was discovered not by human, but by a machine.
• 19:45
  Well, we're not discovering antibiotics quickly enough. It costs billions of dollars to develop a new
• 19:50
  drug antibiotic. The current antibacteria becoming drug resistant to the antibiotics that we that we
• 19:57
  have today because we're over prescribing them. We're still prescribing penicillin, which was to
• 20:01
  invade the first antibiotic discovered back in the 1940s. And so human was still showering it on
• 20:08
  pigs in stalls and, you know, which is undermining its utility by spring things through the agricultural
• 20:15
  system. In a very near future, it's going to be useless because there'll be too many bacteria that
• 20:20
  are resistant to it. So we need new antibiotics. And the latest antibiotic was discovered by a
• 20:26
  machine only program. And what's particularly interesting, particularly promising about this,
• 20:30
  it works in a completely different way to all previous antibiotics. So it disturbs the way
• 20:37
  that the cell can access energy. So the bacteria can't access energy. So it dies. And other
• 20:42
  antibiotics don't work in this way. So the expectation is under clinical trial. As we speak, the
• 20:47
  expectation is that the drug resistant bacteria are not going to be drug resistant to this new
• 20:52
  antibiotic because it's working in a completely different way. So in the future, my children,
• 20:57
  when they take medicines, it's highly likely that those medicines will be devised by machines
• 21:02
  and devised for their genetics. And devised for their genetics so that we can tailor it to them.
• 21:07
  Tailer's much smaller doses are much more specific to, you know, it's remarkable that we,
• 21:12
  you know, we give men and women the same drugs, we give, we can tailor them to, you know, exactly what
• 21:17
  is your biochemical? Right. I mean, speaking of tailoring things to one's own specific situation,
• 21:23
  you're reminding me of the prognostications about the creative arts. We're currently in a
• 21:27
  situation where creative artists are on strike in the United States writers and actors largely
• 21:33
  around questions of what happens to their words and what happens to their image in an era when
• 21:39
  artificial intelligence can take them reuse them, repackage them, build new scripts on the basis
• 21:46
  of scripts that have been written in the past and do a pretty good job of it. In the future,
• 21:52
  will machines be able to write convincing, compelling books that are tailored precisely to my taste?
• 22:02
  I suspect we'll be highly entertained by the books that they write. Well, it already would be
• 22:08
  entertainingly bad, but at some point it'll be entertainingly good, why not? I'm not sure it's
• 22:13
  ever going to be great art in the sense that, you know, great art always speaks to those human
• 22:19
  themes of love and loss and mortality and facing up to your own mortality and the mortality of
• 22:25
  that your friends and family around you and falling in love and all those things. And machines
• 22:29
  don't do that. No, but they can take things that we've written about it. And I don't think they can
• 22:35
  paste them together. Mean as much to us because we'll look at them and think, well, you don't really
• 22:40
  know what it's like to lose a loved one. What if I didn't know it was written by a machine?
• 22:43
  Well, that's why we're going to be entertained by them. I think on the surface they're going to
• 22:49
  certainly amuse us, but whether they really talk to us in the way that great art talks to us.
• 22:56
  I was speaking to an Uber driver of all people about this and he was saying, I don't think
• 23:01
  machines are ever going to be able to reach the creative heights of, you know, speaking truth
• 23:06
  and justice to us. And I said, but what is the human appetite for consuming the creative heights
• 23:13
  of truth and justice? And he slapped his hand on the steering wheel and said, oh, you're the best
• 23:17
  customer today. What is the human appetite? Nothing. Look at what's in the movies. Look at what
• 23:22
  people read. You don't want the good stuff. I mean, it's true. If you look at what the top 10
• 23:28
  movies are and what the summer blockbusters are, they're all remakes of the same Marvel universe,
• 23:34
  airport bookstores are not full of proofs. They're full of... Exactly. That's why I started by saying
• 23:40
  they're going to entertain us. They're going to entertain us to death.
• 23:42
  Right. Okay. I thought you were being cynical about them entertaining us. You think they will do...
• 23:47
  And they do a fantastic job. And if, you know, I like a particular genre of science fiction,
• 23:52
  they'll be able to write particular instances of that. And I can star in that as well. I mean,
• 23:57
  they'll be able to, you know, movies will merge with the gaming industry and we will be at the
• 24:03
  center of every movie. Wait, that's that's amazing. So thinking about deep fakes, I can create a
• 24:12
  deep fake version of myself or rather the computer could rendering it in real time and casting
• 24:20
  me as the hero of the new mission impossible. But using all of the things that they know,
• 24:25
  tickle me in particular, my love of aviation of whatever else it might be, cycling or something,
• 24:34
  and put all that into the film. Yes. And then you, it will respond to what you say in real time.
• 24:39
  You could take the movie off in a different direction and they'll try and build the plot around the
• 24:44
  the interventions. I want to fall in love with that person in the background and all of a sudden
• 24:48
  that happens. It's to choose your own adventure. Yeah. It's grand tourism or a grand scale. Yeah.
• 24:54
  Wow. So are they going to be any creative artists left apart from the occasional proust?
• 24:59
  Well, I think the, the A-list is still going to be famous. I mean, the idea of, you know,
• 25:05
  you're appearing in a movie next to Tom Cruise or whoever it is, your star, you know, the person
• 25:10
  you admire most, you'll be able to do that. And so Tom Cruise will still earn a living from selling
• 25:15
  his rights. But there's only one of him. There's only one. There's only one. Maybe ten of him.
• 25:20
  Yes. But if you're a B-list or an extra in a movie that you're going to be made with
• 25:25
  Archipelagoge's. Wow. So those people are right to be on strike. Their livelihood is at stake.
• 25:33
  But that's got to be a losing battle. He says we're the weary side. Well,
• 25:42
  there's a bit of me that holds hope for the fact that we do like to share collective experiences.
• 25:47
  That the movies are like theatre, you know, we suspend just our disbelief. It's always the same actor,
• 25:53
  sometimes they don't bother to put makeup on. They're just in a different movie. And we do like
• 26:00
  to share those cultural experiences together to be able to talk to, you know, our friends about.
• 26:04
  Right, but they're becoming rarer and rarer already. I mean, when I was growing up,
• 26:09
  I went to school and everybody knew what, well, everybody in the Cool Click Kids Club knew what
• 26:15
  Dave Letterman had done the night before, what had been on Seinfeld the night before.
• 26:19
  Now we're not watching with the exception of a few breakout hits, like white lotus and
• 26:26
  succession and sporting events and maybe reality TV. People aren't on the same page anymore.
• 26:31
  We're already siloed into our own spheres of consumption. So won't AI just supercharge that trend?
• 26:39
  That is something to worry about. I do feel that we're losing that, you know, the common culture
• 26:44
  that we used to share that we, there were only limited number of TV shows, we would share them,
• 26:50
  would watch them all. Now we choose when to watch things and we choose what to watch and we don't
• 26:54
  necessarily share anything with anyone else. That may start to go away. And there may only be, as you
• 27:00
  point out, just, you know, a few unique experiences. You go to that, you go to the Taylor Swift concert
• 27:05
  and everyone shares that moment, but that's the only moment you share with the rest of humanity.
• 27:10
  But do I go to the Taylor Swift concert or do I, is virtual reality and the rendering of
• 27:17
  of those events so good that Taylor Swift puts on a virtual concert with cameras swooping all over
• 27:22
  the place and I'm in my living room with my friends, but my friends are all virtually there as well.
• 27:26
  And if the device isn't a big bulky VR headset, but slim glasses, just like the ones that you're
• 27:32
  wearing, can I have the experience of being in a crowd without being in a crowd? Will that ever
• 27:36
  get good enough that you think it'll fulfill a little scratch stage?
• 27:40
  I think it will be indistinguishable from the real thing being there in person.
• 27:45
  So then I think it becomes a question of, is it a bit like art? I mean, painting now, we can
• 27:50
  perfectly reproduce a van Gogh. But the van Gogh is, the reproduced van Gogh is not as valuable as
• 27:56
  the original because of its scarcity. And so I suspect we will still go to the Taylor Swift concert
• 28:01
  because we're amongst the unique few and managed to get tickets to go. And that, that scarcity
• 28:08
  gives us value. I mean, I suppose it depends. But the experience would be actually worse. I mean,
• 28:13
  you would get a better experience by staying at home. Yeah. On your VR goggles and being right on
• 28:18
  the front row. Well, it also depends whether or not the Taylor Swift concert is also rendered in real
• 28:26
  time using a deep fake of Taylor Swift with her permission. So that, well, I mean, a real,
• 28:33
  real enough for whatever virtual reality device that I'm using so that if I want her to sing
• 28:41
  like a duet with me or if I wanted to sing my favorite song directly to me, maybe it's a
• 28:49
  price worth paying not to go and see the real Taylor Swift, but experience the mood of the crowd alone.
• 28:55
  I mean, I can even imagine a very old me in a nursing home, you know, watching himself in a movie
• 29:03
  theater as a young, as young me now surrounded by other people who are all eating popcorn and watching
• 29:10
  and enjoying me and Tom Cruise and Taylor Swift in my favorite movie. And I'm sharing a communal
• 29:17
  experience, but all of those other people are either non-players or real people in the world. Welcome
• 29:24
  to the Matrix. Yeah. And you see it's we're going that way already, right? So you go to an
• 29:30
  Abba concert today in London and it's holograms. It's holograms of our? Yes, really. The
• 29:36
  Abba experience at the O2 zone is the younger Abba members playing for you as a holograms.
• 29:45
  That's still a gimmick though, isn't it? That still feels gimmicky. Paul McCartney's doing a
• 29:50
  world tour later on this year and he's going to do a duet with John Lennon in, you know,
• 29:56
  a beautiful, I'm sure, a holographic, you know, heartfelt thing, but it feels stilltored.
• 30:04
  There will, there must come a point at which it will no longer feel stilltored.
• 30:08
  Is this Indiana Jones 56? He's still. He's been
• 30:13
  digitally eternally remastered. Yeah. What will be the, what are the impediments to it feeling
• 30:22
  natural? I suppose probably the nature of a headset, the vera similitude of the headsets are
• 30:28
  going to get lighter and lighter. Yeah. We are not going to be able to tell apart those
• 30:34
  synthetic virtual experiences from the real world. And the quality of the deep pack, I suppose.
• 30:38
  Yeah, but they're increasingly already. We've seen an evolution of them that's gone from
• 30:43
  you could work. You could see the little glitches to now. It's really hard to see the glitches
• 30:47
  to soon. It's going to be impossible. And we're going to have to, I mean, that's, you know,
• 30:52
  we would come to some of the challenges. That's going to be one of the challenges.
• 30:55
  Let's come to some of those challenges. Wait. We used to believe in what we see.
• 31:01
  And we're going to have to realize that we will be shortly in a world where,
• 31:05
  unless you were physically in the room and you literally felt the breath of those people
• 31:09
  on your face, you have to entertain the idea that what you're seeing is fake. Everything could
• 31:15
  be faked. We're already there with audio, certainly. Right. I mean, a deep fake of my voice could be
• 31:23
  constructed to call my spouse and tell him that I've run out of money and can he transfer me some
• 31:32
  money and it would sound exactly like me, especially someone like me for whom there's so much
• 31:37
  audio out there. We don't need so much audio anymore. It used to be we needed so much audio. We
• 31:42
  needed to begin with. We needed hours of audio to fake your voice, to clone your voice.
• 31:48
  There are apps that need just a few minutes of audio of to clone your voice. And Microsoft
• 31:56
  have an app that they're fortunately so far refused to release because they think it's too dangerous,
• 32:01
  which I would agree with that requires just three seconds of your voice. It's been trained on
• 32:05
  tens of thousands of hours of voice so it can speak. But then to specialise it to your voice just
• 32:10
  takes a few seconds of audio. So I could just I could just recap your answer phone. That's enough
• 32:14
  of your voice. Isn't it incredible that banks are introducing voice recognition and, you know,
• 32:19
  I was on the phone term. They're still doing it apparently. They still tell me, in order to
• 32:26
  protect your privacy in the future, instead of you having to remember your password or answer
• 32:30
  our security questions, we'll be able to take an array. You know, our system can take a snapshot
• 32:35
  of your voice and can identify you by your voice pattern. Well, surely. This was never true. I have
• 32:39
  a twin brother. It was never true. Is that right? Yes, he could he could pass for my voice.
• 32:44
  Well, sure. But if you're an identical twin, then the voice is identical enough, isn't it?
• 32:48
  Yes. Yeah. So it was never true that you were uniquely identical. For those of us who don't have
• 32:52
  identical twins, it was true. Yes. But that's got to be something that's going to go away
• 32:57
  evenly. Yes. Yes. Very true. And more ridiculously, when you ring out the tax office,
• 33:03
  and everyone says exactly the same phrase. I mean, if they change the phrase, then at least you'd
• 33:08
  have to be able to synthesise a different phrase on demand. But the fact that you always say,
• 33:13
  what do you mean everyone says the same phrase? Everyone says in Australia, your voice uniquely
• 33:16
  identifies you. Right. That's the that's what you have to say at the time. Wow, I haven't been
• 33:21
  through that. That's what they tell that's what they tell you to say in or to say the same thing.
• 33:25
  Wow. And you know what is going to be in advance? So it's so easy to copy.
• 33:30
  Ridiculous. Yeah. And once deep-fake videos become good enough and they are getting there,
• 33:36
  but I don't think like we're not quite there are we that you could FaceTime me and I could answer
• 33:42
  the phone and in real time and a scammer in an Indian call centre could be speaking into a microphone
• 33:49
  and the the system could be rendering to be washed perfectly with your voice and face yet.
• 33:55
  They're already apps that will render but not perfectly. Yeah. But soon it'll be perfect. So soon
• 34:02
  there will be no way of knowing whether or not when you're daughter FaceTime's you again saying
• 34:07
  that she's in a sticky situation and need help that it's her. That's why you need to tell your
• 34:11
  loved ones. What's the secret question? Right. How long does it take I to find out the secret question?
• 34:20
  Don't tell me. Listening to our phones is going to be spying. Mark Zuckerberg will be
• 34:23
  have been listening in. Yeah. Well find a quiet place to tell your loved one what the secret
• 34:27
  question is. In a secret answer. And that's the anyway you could be sure. We all need an
• 34:32
  insulated panic room with no electrical devices in it and no to talk to each other. I remember
• 34:37
  you saying that in the future it may end up being like it was during the Cold War where the only
• 34:43
  way you could ever have a truly private conversation was to meet someone in person in public.
• 34:48
  Yes. Well that's that's the case really. Yeah. It is a bit like the old-fashioned spy move is
• 34:55
  if you want to have a really private conversation you find a really noisy public space. Yeah.
• 35:00
  A park bench in the middle of that. Yeah. Make sure that you know no one's pointing a directional
• 35:06
  sound amplifier at you and then speak away. So we've seen to some extent the cultural and
• 35:11
  political rifts that can emerge when people are susceptible to misinformation. I think much of
• 35:16
  what's deranged the planet in the past decade from QAnon to Brexit to Trump to the rise of the
• 35:27
  far right has been fueled by scapegoating and misinformation and viral things that have gone
• 35:33
  that have popped off on social media about Hillary Clinton's crimes or the COVID pandemic and
• 35:40
  vaccines killing people and Syrian refugees coming in and raping all our women or whatever it
• 35:46
  might be. Once the dissemination of misinformation becomes frictionless and supercharged by deep
• 35:56
  fakes and so on, how do we wrestle with that? That is the thing that I think really worries me
• 36:03
  most at the moment is that we're building this technology. We saw what could happen with social
• 36:09
  media and that was with rather primitive means of manipulation and now we're going to supercharge
• 36:14
  what we could do with social media with art of intelligence and the dirty secret is that humans
• 36:21
  are easily hacked. We're easily persuaded to do things that are necessary in our best interests
• 36:27
  and AI is the perfect to do all to do that at speed and scale and that's going to take us I think
• 36:32
  to if we're not careful to a very dangerous place. We already started to see as you point out
• 36:37
  the beginnings of that with the election of Trump and the Brexit referendum. People got persuaded
• 36:42
  to do things that wasn't actually in their interest and I do worry what's going to happen in the
• 36:47
  next year when we have another US presidential election. We have about 50% of the world's democracies
• 36:53
  are going to the polls in the next year and now we've got this technology that could really
• 36:59
  manipulate what they do. Well, forgive it next year. There's always four years after that and four
• 37:03
  years after that and four years after that. What's the countervailing force to rein it in?
• 37:09
  I think there's one simple silver bullet. I think one is education. We've got to just educate
• 37:14
  ourselves to be a bit more skeptical about what we see. But equally, I think we could have
• 37:19
  probably introduced a bit of regulation here. Most of these harms, these deep fakes,
• 37:26
  are by themselves not particularly harmful. The picture of the Pope in a puffer jacket or
• 37:31
  the Pentagon burning or these are both real viral pictures that went around earlier this year
• 37:36
  that were fake. They only had impact because social media put them in front of millions,
• 37:42
  hundreds of millions of eyeballs. Perhaps we're going to have to hold the social media companies
• 37:48
  to a bit more account, like the sort of account that we hold old-fashioned media. We do hold old-fashioned
• 37:54
  media to a cap. They put stuff up there that's forced, then they have to be in a corrected.
• 37:59
  I mean, good luck because old-fashioned media are publishers and the social media companies
• 38:03
  insist that they're not and at the moment US law agrees with them and they have the richest
• 38:08
  companies in the world. So the idea of them taking regulation lying down seems far-fetched.
• 38:17
  I don't think it's that far-fetched. I mean, after what happened in Christchurch, the tragedy
• 38:21
  that happened in Christchurch, we did introduce laws here in Australia to say if there's this
• 38:26
  really harmful content that is going to incite these sorts of behaviors, you have to take it down.
• 38:31
  You have to take it down quite quickly. And if you're done, we're going to hold the officers of
• 38:34
  those companies criminally responsible. And it's not been perfect, but if you look at the data
• 38:41
  today, that sort of content gets taken down that much quicker. Yeah, right.
• 38:50
  The point about skepticism is an interesting one because teaching people to be more skeptical
• 38:57
  is a double-edged sword. Part of what has happened over the past few years is that people have become
• 39:03
  more skeptical largely of the wrong things, skeptical of mainstream narratives. It's very easy
• 39:11
  to feel like you're wise for seeing through the bullshit that people are feeding you and
• 39:16
  finding out the real truth and doing your own research. How do you teach skepticism towards
• 39:23
  the tsunami of bullshit that's about to engulf us without also teaching skepticism?
• 39:28
  You know, in some way, there's something to be said for a certain level of credulity.
• 39:38
  Well, there's a little bit of me hopes that there's so much fake content out there
• 39:44
  that actually it immunizes us, that social media becomes just so full of these entertaining memes
• 39:53
  and fake content that we realize social media is the place to be entertained. Right.
• 39:58
  And that there is so little signal to noise is so small there that it's just entertainment.
• 40:05
  So it may actually bolster the institutions that have been undermined in recent years as we may be
• 40:10
  at a tipping point and actually gets better rather than worse because as the whole thing becomes flooded,
• 40:16
  as the zone becomes flooded with toxic shit. Yeah, it becomes obvious that that's full of shit.
• 40:21
  Yeah, and only the gatekeepers can be. We've had the luxury of disbelieving the gatekeepers
• 40:27
  by which I mean the organs of journalism and public health and so on.
• 40:33
  But we won't have that luxury once all of the alternative sources of information are obviously
• 40:39
  compromised. Yeah, so it is an opportunity, I suspect, for all fashion media to step up to the
• 40:44
  plate and say, okay, if you want to find out what is true, we put the legwork in. Listen to Josh
• 40:49
  steps as ABC show. Exactly. Go back to the old fashion sources, the ABC or CNN or however it is
• 40:56
  that you trust, because that's where the only place you can be sure that what you're reading or
• 41:01
  hearing might have some validity to. And the challenges for those institutions not to drop the ball
• 41:06
  as I would argue CNN has to a large extent. And we realize that social media is just the place
• 41:12
  where we get entertained. Yeah, right. So let's talk about the siloing of the information that we
• 41:18
  get then, because we were talking about the movies that are tailored specifically to us,
• 41:22
  the Taylor Swift concert and so on. All that's a bit benign when we're talking about reading a novel
• 41:26
  or watching a movie that has me as the protagonist. It's less benign when we're talking about
• 41:31
  the media that we consume. Now, let's set aside for a moment the question of misinformation or
• 41:37
  disinformation online and assume that the what we just said about the swamp actually triggering
• 41:43
  a return to credibility for mainstream journalism is true. Nonetheless, there are going to be
• 41:49
  so many and there are so many different slants and angles and things to be reported on
• 41:55
  that the average person in the seven or eight minutes a day that they spend consuming the news
• 41:59
  is going to receive a very different snapshot that will be tailored to them by AI on the basis of
• 42:04
  what they what the AI knows engages them and makes them likely to click and swipe and like and
• 42:11
  comment and so on. What does the world look like when we're all walking around with our own
• 42:19
  individually tailored worldview? Again, I think that undermines our society. Our society,
• 42:25
  we don't we lose those those common touch points that people, even today, I mean the internet
• 42:30
  you see is unique to you. I see a completely different internet, the adverts, I see a completely
• 42:35
  different, even when I read an online newspaper, I'm seeing different stories because it's already
• 42:41
  making choices for me as the ones I tend to click on are the ones that's going to put up first.
• 42:48
  All the prices I see on the internet are unique to me. You see different prices. People
• 42:53
  aren't really aware that the internet is just tailored to your preferences and tailors. Did you see
• 42:59
  that scandal where there was a I think it was an airline or a travel website that was providing
• 43:07
  different fares for people on Apple products than on PCs because Apple people have a higher income.
• 43:14
  But more for hotel rooms than people on PCs. So search things on PCs or in an incognito
• 43:21
  high behind the VPNs. Yeah. Extraordinary. Okay. And that turns the prices. That does, you know,
• 43:29
  that does we undermine our sense of justice, right? We're used to walking down this
• 43:35
  high street and seeing a product for sale and thinking, well, anyone is allowed to walk in and buy
• 43:39
  it for that price to realize that actually now most of the prices you see on the internet are ones
• 43:44
  that were chosen for you because they're the ones they think you're willing to pay that amount
• 43:48
  for it. Extraordinary. So we all walk around with our own perception of what is, you know,
• 43:54
  dramatic and important in the world. You only need to look at what has happened lately to the
• 43:58
  conversation around hot button issues, whether that's drag queens reading to children in libraries
• 44:05
  or pediatric transgender treatment or the climate crisis or, you know, who is the real threat to
• 44:13
  democracy, woke people or the far right. That's wonderful. I haven't heard about the drag queens.
• 44:19
  You're living in a different bubble. This is a huge thing. I mean, the bubble that tells you
• 44:23
  about all the tech challenges that I get. All these tech stories thrown out to me. That's the first
• 44:28
  time I can imagine. I love that. It's a huge controversy about whether or not drag queens should
• 44:33
  be allowed to read. They're called it drag queen story hour. And it's been going on for years where
• 44:38
  drag queens who are essentially clowns, right, and reading to kids and getting kids involved in
• 44:44
  reading. But you'll get the occasional viral video of some hypersexualized drag queen in some
• 44:53
  counties in San Francisco where all the rainbow families have gathered together and she's
• 44:57
  twerking or something or she's got, she's got comically oversized breasts that she's rubbing in
• 45:02
  the children's faces. This goes viral and gets pushed by the right wing media and everyone in
• 45:07
  Missouri is horrified and it becomes two weeks of talking point and becomes a really important thing
• 45:14
  that people think they need to be focusing on because the children are being corrupted,
• 45:18
  so save the children. So we'll walk around with our own priority list, our own hierarchy of what
• 45:25
  ails the world. And if none of those lists intersect, how do we get common, how do we get on
• 45:32
  the same page to find common solutions to the problems that actually do ails us like climate chaos
• 45:38
  and nuclear weapons and AI? Yeah, I think you put your finger on a really important thing which is
• 45:45
  we will lose that collective society if we're all thinking about different things or we're not
• 45:51
  sharing those same cultural touch points, we're not sharing those same news touch points,
• 45:56
  that it will undermine the cohesion of our society which requires us to pretend that we're still
• 46:05
  back in the village and the town square that we used to be in. Let's escalate to the next level
• 46:11
  of concern here which is a word that I've heard you use which I think is a wonderful word,
• 46:18
  serendipity. Let's set aside the challenge of misinformation and the challenge of receiving all
• 46:27
  of our news tailored to us and just talk about what it will be like, we're already seeing glimpses
• 46:34
  to live in a world where all of our behavior is essentially nudged into place by algorithms
• 46:41
  that know us better than we know ourselves, tailoring everything to our own whims. What's your,
• 46:51
  the current incarnations of this are the fact that the music you hear on Spotify,
• 46:56
  the television shows you watch on Netflix, the articles that are offered to you to read
• 47:03
  as you surf the web are all, and obviously the posts that you see on social media are all
• 47:07
  tailored to what they think is going to engage specifically you, this individualisation of everything
• 47:15
  indeed even dating now where Tinder, you called it, you've called Tinder as sort of a vast genomic
• 47:21
  experiment in some ways because we don't really know what, well explain that because I won't explain
• 47:28
  it for you. What, we're handing over some of the most important decisions of our civilisation,
• 47:34
  which is like who we procreate with, to algorithms that the majority people now meet their spouse
• 47:42
  through one of these dating apps and out of all the thousands, tens of thousands of profiles that
• 47:48
  you can be displayed, the algorithms are making the choices to avoid which ones you're going to see
• 47:53
  and therefore which ones you're going to swipe on. And that is introducing a different mixing
• 48:01
  into our gene pool. It used to be that we met the people who were friends of ours, who were in our
• 48:07
  immediate social circle, who were geographically close to us, or that we might have bumped into
• 48:12
  in a bar. But now we're mixing the gene pool much more widely and on different criteria.
• 48:19
  The criteria being perhaps quite significantly more appearance driven because they're photographs
• 48:24
  on apps. And that's an interesting long-term experiment on the human gene pool. What are those,
• 48:31
  you know, the next generation are decided by those matches and the genes that we then mix
• 48:36
  according to those matches? I don't think anyone thought, well, well, is this going to be a good
• 48:42
  impact? Or are we even allowed to talk about this? I mean, we're in dangerous territory now
• 48:48
  that's, you know, verging on eugenics where we're talking about, well, what is going to be the
• 48:51
  impact of allowing people to find their spouse and mate through algorithms, supposed to through
• 49:01
  the old-fashioned way, which was, you know, did you talk to someone drunk in a bar? I mean,
• 49:04
  I'm not sure which was better, but it's going to be different, whatever it is.
• 49:07
  It will be a different selection of chromosomes in any of the...
• 49:11
  And that's the future of the human race being decided by those algorithms.
• 49:15
  Yeah. And more and more so, and the algorithms will, I mean, you know, those babies that get
• 49:21
  created through those Tinder hookups are going to, going to procreate in a world in which,
• 49:27
  who knows what the Tinder algorithm is going to be like or the successor of Tinder is going to be
• 49:33
  20 or 30 years. Right. So serendipity, I've heard you talk about, you know, that the sparkle in the eye
• 49:42
  of someone at the bar who may not have been someone you would necessarily select by swiping them on
• 49:48
  on Tinder, they wouldn't have put together the perfect photo for you, but there's just something
• 49:53
  ineffable about them. It might be their pheromones, it might be the twinkle in their eye, whatever it is.
• 49:58
  The loss of that and the rise of us sitting alone in apartments, swiping desperately and chatting
• 50:06
  with people on in text instead of on sort of face to face. What does that world look like? Is that a...
• 50:17
  Forget set aside Tinder specifically, just the growth of all of my experiences being selected,
• 50:24
  being pre-selected for me by algorithms. What does that look like?
• 50:32
  It's... I certainly have different world. I mean, it's not more clear to me whether it's a better
• 50:37
  world or a worse world in the sense that we're going to be again. I come back to the phrase I
• 50:41
  used to earlier on, which is we're going to be well entertained and we're going to be
• 50:46
  satiated. Our desires are going to be met. But whether we're going to be challenged in the way
• 50:51
  that we are now, whether we're going to grow in the way that we are now, because life doesn't
• 50:56
  always have to be exactly the taste that we like. Sometimes it's good to be pushed in directions
• 51:02
  that are the first and not very comfortable. And what do you say to people who say, I mean,
• 51:06
  this has always been the sort of evolution of humankind. Do you look back at photos of men
• 51:12
  commuting to work in the 1950s and they were all sitting there on their trolleys and trams in
• 51:16
  suits reading newspapers. They were reading what they wanted to read. They weren't chatting to
• 51:21
  each other amnacably. They weren't sparkling and twinkling in each other's eyes. They were
• 51:25
  heads down just like we are now with our phones and they were associating with their own cohort
• 51:31
  and reading their favourite newspaper and socialising with their favourite people and
• 51:36
  in their particular office. Yeah, it's hard not to sound like an old phoenix. I remember I had a
• 51:45
  moment 10 years ago. I remember I went to Harvard and got the train from the airport. And I was
• 51:53
  sitting on the train and I looked around and it was full of young students at MIT and Harvard.
• 52:00
  So the cream of American intelligence. And I was the only person who had a book
• 52:08
  everyone else was on their phone scrolling. And I thought, what have we become? I mean,
• 52:13
  we're being entertained. You do sound like an old phoenix. I know. This was 10 years ago.
• 52:19
  This was the intellectual population of the United States already just being entertained on their
• 52:25
  phones. What's wrong with that? Well, you know, there are various different parts of
• 52:35
  either the reptilian part of our brain that just wants to be amused and
• 52:39
  associated and then there's the cognitive part of our brains that want to be stimulated and
• 52:45
  take, and if we want to grow as a society, I think we have to have to stretch ourselves in those
• 52:50
  directions. Is boredom essential? Yes. That's one of the greatest gifts I think you can give
• 52:58
  your children to allow them to be bored. To give them time to be bored. I remember being
• 53:02
  as a child being bored. Lots of the time. And that encourages you to, well, first of all,
• 53:08
  be happy in yourself, but equally, to go off and be more creative. You don't have to be
• 53:15
  internally stimulated by the latest Hollywood blockbuster, but the simple things.
• 53:23
  Actually, ultimately, I think give us greater pleasure in the longer term.
• 53:26
  It's interesting that you mentioned the Harvard train because, as you know, I spent most of my
• 53:30
  professional life in New York City, and I remember vividly when the New York subway brought in
• 53:37
  mobile phone reception on the subway because my commute had previously been underground
• 53:43
  and phones didn't work. And so it was wonderful. You'd sit there and you'd stare at other
• 53:49
  people's faces and pretend not to be staring at you. And people would, you know, pretend to
• 53:54
  look at their shoes or people would read books or, you know, someone would do a dance in the
• 53:58
  subway car. And it's all human psychology on the New York subway. It's a zoo. And then overnight,
• 54:05
  you know, click of the fingers, they introduce mobile phone reception, and it all dies. Everyone's
• 54:12
  staring down into their own universe of entertainment. Wouldn't be able to tell you what the
• 54:18
  person next to them looks like was the day before they would have. It's doing something.
• 54:25
  Yes, yes, I'm not sure whether it's contributing to the polarization that we see with
• 54:32
  an our society, this function that we start to see that we're not understanding each other
• 54:38
  anymore because we're losing these shared values. We're losing these cultural touch points
• 54:45
  off in our own little worlds. Will, is there an optimistic take here akin to the optimistic
• 54:52
  spin that you put on this, you know, army of bullshit in that maybe we're just at a point
• 54:59
  where the machines aren't really good enough to give us fulfillment and they're selecting for the
• 55:03
  wrong kinds of novelty and they're essentially feeding us intellectual fast food. But could we
• 55:09
  imagine there were hints of that. So for example, Amazon have opened a real world store
• 55:16
  because, you know, they are the world's store online, but they were discovering that they weren't
• 55:24
  learning people's real preferences because people would just go type in what they want to buy it.
• 55:28
  Right. And they weren't making all those serendipitous other purchases you would in a real store
• 55:34
  way as you go to the cash point you see an object you think, oh, I'll buy that.
• 55:40
  And so to get the data on what the other things that people might buy that they weren't finding
• 55:45
  online, Amazon opened some stores to get that data. They lose money on the stores.
• 55:52
  Right. Interesting. And they're running the stores just to collect data so that they can offer
• 55:56
  you better choices as to, well, people who buy this product also buy this. Right. You might
• 56:01
  want to buy this to get that data. They're willing to lose money on physical stores.
• 56:05
  Interesting. I mean, one thing that I think is good is we're all going to be living with pretty
• 56:11
  soon that we'll look back on today and think it's incredible that we didn't have is an interaction
• 56:17
  with machines that are indistinguishable from humans, right? So that instead of having to type
• 56:24
  things in and having to talk to Siri and it's kind of clunky way or having to wait on hold to speak
• 56:30
  to an associate at a bank or an airline, all of those conversations will be automated and we'll
• 56:35
  have machines doing at least as good a job if not better than the Indian call centres that we
• 56:40
  mostly communicate with with no wait times. And I assume that I'll be able to talk to my phone
• 56:46
  and talk to systems without having the clutter of physical devices that I'll just be sort of
• 56:51
  constantly interacting with intelligences that feel like they're, they're conscious even if they're
• 56:58
  not. Well, come to a point where we walk in the room, we just say lights on. Turn the
• 57:02
  table. Well, we're sort of already there, aren't we? Lights on. And, you know, who won the food
• 57:06
  last night? We won't even need to say lights on. Lights have just come on. In fact, that
• 57:10
  technology has been around for 40 years. But, you know, who won the foodie last night? We just say
• 57:15
  it into the end. Expect something in the room is listening. We'll recognise our voice, we'll look
• 57:20
  up, you know, which is our point. But I mean, more interesting would be, can you tell me,
• 57:23
  you know, what can I cook for? What are a few things I could cook for you? And I give them
• 57:26
  what's in the fridge? And, you know, I don't know what's in the fridge. It'll look at what's
• 57:30
  in the fridge. We'll just say to the room, those sorts of things and it will say you've got some
• 57:34
  eggs for an omelette, a few vegetables. You can make yourself a veggie omelette. Yeah,
• 57:40
  and I'll talk you through the recipe if you don't know how to create a veggie omelette. Yeah,
• 57:45
  I'll turn the stove on. Yeah, I mean, it could be, you know, you could walk in and it might
• 57:50
  notice that your vital signs are a bit stressed and that might say, you know, if you have any
• 57:55
  interesting going away this weekend, there's an Airbnb in the blue mountains. Yeah, in the blue
• 58:01
  mountains, which has a 40% discount on it, if you want it, if you want me to jump on it.
• 58:05
  Which is, which is, you know, in some sense, our lives are going to be much more convenient and
• 58:10
  this can be less friction in our lives. But equally, if you were someone who was employed in one of
• 58:13
  those core sensors, then you're going to have to say, well, wait a second, what happened to me?
• 58:17
  Right. What happened to my job? You know, I've been replaced and what am I doing now?
• 58:22
  Before we get to that dark, you know, question about employment, can we just linger on the light one
• 58:28
  of, on the positive side of, does that level of sophistication mean that the zombified subway car
• 58:37
  of people gazing at their phones maybe goes away and maybe people are still in some kind of
• 58:45
  AI world, but maybe they're not looking at a device, maybe there, maybe some people are listening
• 58:50
  to things, maybe other people are looking at augmented reality version of the subway car, maybe
• 58:55
  other people are, I don't know, being told a poem by their AI, I don't know if that's better or
• 59:02
  worse than the status quo. Well, some people were sitting there with their, in their 3D world,
• 59:06
  that's projected onto the front of their glasses, that they're in this metaverse virtual world
• 59:15
  that doesn't exist and then their glasses. And what do you think about the consequences then for
• 59:21
  employment? I mean, where, you know, I hear one half of people say, this is going to decimate
• 59:30
  people's lives and there are going to be so many people out of work and then I hear other people say,
• 59:34
  well, there have always been revolutions in the workforce and people will do what the people who
• 59:40
  happened whales did when that was out of work, they'll find another line of employment.
• 59:44
  I think the honest answer is we have no idea, absolutely no idea what the balance is going to be,
• 59:48
  whether more jobs get created than destroyed. And we absolutely certainly can be disruptive,
• 59:54
  I mean, there's no doubt at all, it's going to, some jobs are going to disappear completely.
• 59:58
  And probably more disruptive than any previous era, at least since perhaps the industrial revolution.
• 1:00:05
  I think it's going to be more disruptive in terms of the speed, it's going to happen quicker.
• 1:00:10
  So if you think about, we've had other disruptions to the way the employment worked, industrial
• 1:00:15
  revolution was one of those where we stopped farming and most of us were employed in agriculture,
• 1:00:20
  now 3% of us are employed in agriculture. But that happened slowly, it took 50, 50 odd years
• 1:00:27
  for the industrial revolution to percolate around the world. Equally, the internet, you know,
• 1:00:33
  the internet, again, happened slowly, we had to plumb the world with the internet before it
• 1:00:37
  actually started to be useful. But in this case, this revolution put the plumbing in already.
• 1:00:44
  So it's going to happen overnight. And that's why, you know, for example, Chatchy PT was the
• 1:00:49
  fastest growing app ever. It's in less than six months, it's in the hands of over a billion people.
• 1:00:56
  So I think if anything, you know, the thing that's going to challenge us is the speed,
• 1:01:01
  it's going to happen so quickly. It's going to be also interesting, you know, the white collar
• 1:01:07
  classes have been perhaps shamefully sanguine about truck drivers and check out Chatchy as losing
• 1:01:14
  their jobs. But once it comes for the rest of us, and as you mentioned earlier radiologists,
• 1:01:20
  you know, doing a poorer job than AI is at spotting cancers on x-rays of lungs and so many other
• 1:01:26
  professions get swept aside. Then it'll be interesting what the dynamic is there in terms of
• 1:01:32
  the political and cultural and social upheaval? Yeah, you're right. And I think, you know,
• 1:01:36
  people got the wrong idea. They thought that blue collar was the most at risk and white collar
• 1:01:41
  was at least at risk. And in fact, actually, probably the opposite. There are many places,
• 1:01:45
  many blue collar jobs which are either too lowly paid to be worth replacing you with some
• 1:01:51
  automation or too fiddly to, you know, there's still robots are still expensive, difficult
• 1:01:59
  objects to build, whereas knowledge work is easy to replace. And so become a plumber or a nurse?
• 1:02:08
  Plumbers are going to exist. I mean, one of the safest professions that's going to exist for
• 1:02:13
  hundreds of years into the future is being a plumber, right? So, you know, pay someone extra to
• 1:02:16
  come in and fix those, there's the legal price. So robots, I'm going to do that very easily,
• 1:02:24
  very well or certainly cheaply enough. Whereas, as you say, being a radiologist, well, we're
• 1:02:30
  already trained algorithms to read x-rays more reliably, more quickly than humans. And this
• 1:02:37
  is not a holy, bad thing. I mean, I've done stories on my radio show about regional centers in
• 1:02:42
  Australia, which has a unique challenge with healthcare because it's the same size as the
• 1:02:47
  contiguous United States, but has the population of Florida huddled on largely just the southeast
• 1:02:53
  coastline and say what these enormous distances with small populations scattered around. And this
• 1:03:00
  was a story about how AI radiology is being used and, you know, telehealth so that you can spot
• 1:03:07
  things that can then be verified by a human being in one of the capital cities, but they're bringing
• 1:03:12
  services to people that would never be possible if you were alive on a human being because AI can do it.
• 1:03:18
  So that's good at saving people's lives. It's putting radiologists out of work, but it's
• 1:03:22
  potentially saving people's lives. I don't think any radiologists have been put out of work yet.
• 1:03:26
  It will. Possibly. Why wouldn't it? Well, at the moment, you know, there's a shortage of radiologists.
• 1:03:34
  Why aren't we using AI? We're starting to use AI. So for example, you know, the trials on
• 1:03:39
  place now, for example, if you go and have a breast mammogram, it's read by two doctors.
• 1:03:48
  Well, if you can replace one of those doctors by an AI, right? And the idea is if the two doctors
• 1:03:53
  agree, you've not got breast cancer or two doctors degrees, you've got breast cancer, then that's
• 1:03:58
  the conclusion. If they disagree, then of course, then they go and look more carefully at what's
• 1:04:02
  going on here, right? So if you can, if you can replace one of those doctors by an AI, which is
• 1:04:07
  just checking, in other words, what the human radiologists just said, you can double the throughput.
• 1:04:13
  Right. So you can now have twice the number of women have breast cancer. And so spot more breast
• 1:04:20
  cancer. Interesting. All right, let's escalate to the worst case scenario, which is a lot of
• 1:04:25
  people's minds, which is like an existential threat that AI could go rogue. What do people
• 1:04:34
  mean when they talk about an alignment problem here is probably a place to start?
• 1:04:38
  So this is the problem that if we're going to build really intelligent, much more intelligent,
• 1:04:44
  maybe the nice machines, then we have to align their values with human values. We have to make
• 1:04:49
  sure that they go off and do things that are in humanity's interest. And how do we do that?
• 1:04:55
  That's actually quite an interesting challenge. Why is it not as simple as saying, don't hurt people,
• 1:05:00
  do what we say? Well, it's very hard to specify exactly what it means to don't hurt people,
• 1:05:05
  right? So for example, you might say to the AI, can you, can you eliminate cancer from the world?
• 1:05:13
  That would be a great thing to do, right? So go off and eliminate cancer from the world.
• 1:05:19
  And the AI says, okay, human humans, I've worked out how to eliminate cancer from the world.
• 1:05:24
  I'm going to kill every one of you. Or at least everyone with cancer.
• 1:05:29
  There will be no cancer in the world. But that assumes that you haven't put in line one,
• 1:05:32
  which is don't hurt people. But even don't hurt people, right? There's lots of places where
• 1:05:40
  there is no solution that doesn't hurt people. It's all about trade-offs. Right.
• 1:05:46
  So it would be paralyzed from doing anything if it had to, if it had to slavishly a base such a
• 1:05:50
  command. It's very hard to specify, you know, what are the actual goals that can't be misinterpreted
• 1:05:57
  and that don't have unexpected consequences. And in some sense, we already have this problem.
• 1:06:03
  We already see the value alignment problem, which is with the superintelligence. So the superintelligence
• 1:06:09
  that already exists is the superintelligence you find in a modern day corporation. So no one on the
• 1:06:15
  planet knows how to build an iPhone. No one on the planet knows how to build a nuclear power station.
• 1:06:21
  But you know, the employees of Apple and the employees of General Electric, collectively,
• 1:06:26
  between them have the collective intelligence, superintelligence, to be able to do those things.
• 1:06:30
  Well, the employees of Apple probably don't, but the employees of their subsidiary in Shenzhen,
• 1:06:33
  which actually puts the fans together every problem they know is more than the Apple.
• 1:06:37
  They include all their supply chain, right? Between the collectively, there is the intelligence
• 1:06:41
  to be able to do those things, but it doesn't reside in one person.
• 1:06:45
  And then, so we can, you know, perhaps entertain the idea that there's, you know, a greater
• 1:06:51
  than human intelligence that resides within those corporations, within the collective
• 1:06:55
  expertise of those corporations. But we already have the problem that those corporations are not
• 1:07:00
  aligned to humanity's survival. Just look at what big oil has done for the last 50 years.
• 1:07:06
  And the climate emergency is a direct consequence of the fact that, you know, for 50 years now,
• 1:07:11
  we've known this is happening. And it's been very well known to the scientists of these corporations
• 1:07:16
  that this was going to happen. And they've chosen, you know, nothing was done about it.
• 1:07:22
  Whether they, their values were to maximize shareholder value, which they have done very successfully.
• 1:07:28
  And so we now have to ask the question, well, how do we align the values of those corporations?
• 1:07:34
  Or how do we have aligned the values of superintelligence to be more closely aligned with those of
• 1:07:40
  the flourishing of humanity? Right. What's a scenario? Can you give us a specific, because I think
• 1:07:46
  people sometimes lose the thread, including myself here, about, so what is the disaster scenario,
• 1:07:52
  in which, you know, an AI that can do as much thinking in a second as I can do in 10,000 years,
• 1:08:00
  is tasked with solving the problems of climate chaos and pandemics.
• 1:08:06
  Seize some vision that is so sophisticated, it's impossible to articulate to our puny human minds
• 1:08:11
  and starts undertaking that. Like a lot of people say, well, turn it off, just stop doing that,
• 1:08:18
  embed an override in it. It doesn't have arms. What does it do? What is this existential end game
• 1:08:27
  that people are worried about? Well, yeah, it doesn't have arms, but we've already got to a point
• 1:08:33
  where we can't turn the systems off. We're totally dependent upon computer systems in our lives now.
• 1:08:40
  Because switching off the internet would switch off our hospitals and water supply.
• 1:08:46
  The water supply electricity grid, air traffic control, you name it, you know, it's not surprising,
• 1:08:52
  you know, when the internet goes down, most businesses tell people to go home. There's nothing
• 1:08:57
  we can do anymore. So don't go into our lives, right? So we're not in a position where
• 1:09:03
  really we can turn. And we're not even testing AI in secure situations where it's not connected
• 1:09:09
  to the internet. I mean, we're literally testing it on the internet and giving it to everybody now,
• 1:09:12
  and it's learning on the web. So if there was ever a chance to quarantine it and not let it out of
• 1:09:18
  the box, I assume that's passed. Yeah, GPT is already on the web. It's already on the web.
• 1:09:23
  It's already doing stuff. It's already people already giving it agency to do stuff.
• 1:09:27
  And we've seen, you know, we do have examples already of how we give
• 1:09:32
  some agency to the algorithms and things go terribly wrong. So in particular, I'm thinking of
• 1:09:39
  the stock market. So we've got stock market now. Many of chair transactions are not decided by
• 1:09:44
  humans, but decided by fast trading algorithms. And every now and again, it goes wrong. We're
• 1:09:49
  now again, there's a flash crash. And we've, we've realized that's going to continue to happen. We
• 1:09:55
  don't know why it happens. We don't know how to fix it. So they put circuit breakers in and they
• 1:09:59
  say, okay, stop, stop everything. We have a record of fortunate evil with the transactions.
• 1:10:07
  Let's just unwind everything, give everyone back their money and say, none of what happened
• 1:10:11
  to the last hour happened. Let's begin again. But if those weren't algorithms that were sitting
• 1:10:18
  on the stock market, but algorithms that were sitting on the DMZ between North and South Korea,
• 1:10:24
  maybe you've created a flash war. And the nuclear missiles are on their way. And you can't
• 1:10:31
  unkill the people you've just killed. Right. Right. That's an example of a glitch. Is there an
• 1:10:40
  example of a more conniving AI strategy? I mean, is there an example? Because I listened to
• 1:10:53
  LA as a Yudkowski or people who are the real chicken littles here who feel like
• 1:11:01
  it is foolish to bring into existence to give birth to a system that is
• 1:11:07
  as much more intelligent than we are as we are than a dog and expect that it isn't going to
• 1:11:13
  develop priorities of its own that do not treat our interests as sacrosanct.
• 1:11:21
  So we're opening a portal and something's coming out of it and we don't
• 1:11:25
  understand how to control that thing. Really, we come back to the conversation we had at the start,
• 1:11:30
  which is whether these machines will have any sentience consciousness.
• 1:11:34
  Do they need to in order to can live in order to manipulate?
• 1:11:38
  Then you could certainly not. It's certainly plausible then to consider that maybe they're going
• 1:11:43
  to have goals of their own. They're going to have survival instincts of their own, just like
• 1:11:47
  you and I have survival instincts to keep our consciousness alive.
• 1:11:52
  Right. But a humanoid robot could also have survival instincts.
• 1:11:56
  Like you don't need our lights to be on. You don't. And it would make sense to actually
• 1:12:03
  program those into machines. Right. So they don't they're going to to help up those
• 1:12:08
  machines avoid hurting themselves. You want to program them with something like a survival instinct,
• 1:12:15
  which is, you know, don't allow your hand to be crushed in the door or don't allow your
• 1:12:22
  don't allow accidents to happen to you and that will allow you to the machine to survive.
• 1:12:27
  They're sorts of instincts that we would want to program even if they don't occur.
• 1:12:31
  So setting aside the question of whether or not the AI becomes intelligent,
• 1:12:36
  do you give any credence to the fear that a system
• 1:12:42
  sufficiently advanced, sufficiently intelligent, sufficiently strategic could manipulate
• 1:12:50
  humans, manipulate civilization, manipulate our information flow in order to achieve aims that
• 1:12:55
  it thinks that we're too stupid to, I say, thinks in quotation marks, right? It's not doing any
• 1:13:00
  thinking, but you can imagine a, I can imagine myself in relation to a dog in doing things that I
• 1:13:10
  think are in the dog's interest or in the interest of all dogs that my dog might not think that
• 1:13:16
  it wants me to do. Get it to take medicine, for example, and it lands for the dog as being abuse,
• 1:13:21
  but I know better. Do you envision a scenario where dogs have done quite well out of their
• 1:13:27
  relationships with more intelligent human? This is true. They've actually played this very well.
• 1:13:34
  I do sometimes wonder if you ask a bunch of really intelligent people, like my colleagues,
• 1:13:39
  about, you know, what's the most important value you have, they say, intelligence.
• 1:13:44
  They're worried about an excess of it. You know, I've hung around a few politicians,
• 1:13:49
  think probably more intelligence around the politicians should be a good thing. It's not
• 1:13:53
  intelligence, you know, that I think is the worrying thing is power, who is given the power to do
• 1:13:57
  things. And maybe there's a little bit of hope, a little bit of me, the hopes that
• 1:14:03
  with intelligence might come from wisdom and that we would have to be less worried about.
• 1:14:08
  You know, the more intelligent my car is, myself driving car, the happier I'm going to be.
• 1:14:14
  It's going to be smarter, avoiding accidents and thinking ahead. So I don't think, you know,
• 1:14:20
  intelligence in my car seems to be a particularly bad thing. When I start, when I give the car,
• 1:14:26
  the permission, the power to be able to impact other parts of my life, I start to worry.
• 1:14:31
  But intelligence per se, I don't see as being that problematic.
• 1:14:36
  Right. Interesting. So, well, let's end on that then. When my children, who are currently five,
• 1:14:43
  are your age, let's assume that the world is full of vastly more intelligence than it currently
• 1:14:49
  has right now. We have artificial systems that are capable of piecing together strategies
• 1:14:54
  that nobody can comprehend currently. Are you an optimist or a pessimist or a little of both
• 1:15:01
  about what that world looks like? Oh, I'm definitely a glass half full person at that point.
• 1:15:07
  I mean, I do think the next 20 or 30 years are going to be pretty painful. I think we've,
• 1:15:12
  we've put ourselves in a pretty unique position in the history of humanity,
• 1:15:17
  which is that we have gifted our children a worse planet, guaranteed. And that's not,
• 1:15:22
  that hasn't ever been true before now. Every generation has inherited a slightly better planet.
• 1:15:28
  And we've managed to reverse that. Well, in 1910, were people, people who were born in 1900?
• 1:15:36
  Well, technology was coming along. Life expectancy. Well, well, one, well, two.
• 1:15:41
  The Holocaust. Of course, the Passiondale. The Great Depression. History has,
• 1:15:48
  there have been some events on top, which have changed our lives. But if you look at the
• 1:15:53
  grand course of history, you look at life expectancy, you look at median income.
• 1:16:00
  Since the Dutch Revolution, even in Australia, life expectancy is doubled. We used to
• 1:16:04
  die enough, what is now we die in our eighties. We live like kings and queens. We have machines
• 1:16:11
  that wash our clothes, wash our dishes. Those are things that we used to have servants for.
• 1:16:17
  So we have overall, I mean, you're right, you know, there have been some blips of first world
• 1:16:22
  war or the Great Depression. It hasn't been strictly one way. But overall, the general trend has
• 1:16:32
  been for life to get better. And technology has driven that. It has been technology has driven
• 1:16:37
  it. And in the last 50 years, it's been computing. The productivity gains, the growth in
• 1:16:43
  our economy has been driven by the digital revolution. So I'm optimistic in the sense that
• 1:16:52
  we will continue that progress. But just like, as you say, the first world war made things worse.
• 1:16:58
  We have guaranteed ourselves 10 or 20 years of pain as we have to work our way through
• 1:17:03
  the climate emergency, increasing inequality, the polarization of our political discourse,
• 1:17:09
  who would imagine, you know, we're fighting wars again in Europe. The global insecurity,
• 1:17:16
  the tensions that we see, those are challenging problems. I'm pretty optimistic, though,
• 1:17:22
  in sense of if we get through the next 20 or 30 years, then it will be about a planet and technology
• 1:17:28
  AI in particular will be helping us to live better, longer, healthier, hopefully happier life.
• 1:17:35
  All right. Hang in there, humankind. You heard it here from Toby Walsh. There's
• 1:17:39
  there's Sunshine and Lollipops and Milkenhoney on the other side of the, of the Abyss. Thank you,
• 1:17:44
  Toby. Thank you.