Latest / UK Column Radio / Jobst Landgrebe and Barry Smith: Why Machines Will Never Rule the World

Transcript

Jobst Landgrebe and Barry Smith: Why Machines Will Never Rule the World · UK Column Radio

0:38
Speaker 1: Okay. So I'm a philosopher working originally in England, focusing
0:44
Speaker 1: on European philosophy, particularly in the German language initially, but
0:49
Speaker 1: then I focused on certain aspects of philosophy which relate
0:54
Speaker 1: to data and to what we call ontologies, which are
0:59
Speaker 1: at ten to describe data in consistent ways to make
1:02
Speaker 1: computers work better together. And as a result of that work,
1:07
Speaker 1: I gradually moved over into areas of computer science and
1:12
Speaker 1: data technology, always bringing my philosophical interests to bear. And
1:18
Speaker 1: now I have quite an important role in ontology work
1:22
Speaker 1: in the US government, the Defense Department, the Department of
1:27
Speaker 1: Homeland Security. I'm also part of a team which is
1:32
Speaker 1: trying to bring about ontological or data consistency and the
1:37
Speaker 1: Five Eyes, which are the five English language intelligence communities.
1:42
Speaker 1: And I guess that's enough.
1:44
Speaker 2: So yeah, I'm by training physician, by chemist and mathematician,
1:49
Speaker 2: and I've been using AI theoretically and practically to make
1:54
Speaker 2: a living since nineteen ninety eight, so I've always been
1:57
Speaker 2: in the area.
1:57
Speaker 3: I feel.
1:58
Speaker 2: Mostly i've been using it in life science is a
2:00
Speaker 2: medicine since i'm physician, but I've also used it for
2:04
Speaker 2: the insurance industry and other domains. So I have quite
2:07
Speaker 2: a good cross industry understanding of AI usage and it's
2:12
Speaker 2: possibilities and limitations. I've always been making a living of it.
2:15
Speaker 2: And when Barry and I made fifteen years ago in
2:18
Speaker 2: the context of standardization of medicinal data exchange medical data exchange,
2:24
Speaker 2: and then we started collaborating scientifically, and ten years ago
2:28
Speaker 2: we started working on AI together. And now the second
2:31
Speaker 2: edition of our book, that you describe as somewhat contentious,
2:35
Speaker 2: is just being published these days, which I think is
2:37
Speaker 2: also reason why we are talking and it shows that
2:40
Speaker 2: though the book is a bit debated, it is a
2:44
Speaker 2: success because books of this nature usually don't.
2:47
Speaker 3: Get second editions.
2:48
Speaker 2: It's a very complicated, very scientific, very technical book that
2:51
Speaker 2: requires a lot of prior knowledge, and so it's not
2:53
Speaker 2: a mass compatible book. But I think it's very important
2:57
Speaker 2: that we have done it, and I'm glad that we
2:59
Speaker 2: got the opportunit need.
3:00
Speaker 4: To make a second edition just for clarity. When we
3:02
Speaker 4: talk about artificial intelligence, so what are we talking about?
3:05
Speaker 2: So we are talking about mathematical algorithms that are based
3:10
Speaker 2: on regularities that can be found in data, and these
3:13
Speaker 2: their regularities can be exploited with various ways. So they
3:16
Speaker 2: are explicit and implicit ways to exploiting these regularities, and
3:20
Speaker 2: that has two advantages. First of all, you can find
3:22
Speaker 2: new regular patterns, like when you use this to look
3:25
Speaker 2: at galaxies, you can find certain aspects of galaxies that
3:29
Speaker 2: you cannot see with the bare eye right, or star systems.
3:33
Speaker 2: You can use such such algorithms to analyze them. You
3:36
Speaker 2: can also use them to anilize chemical molecules, the similarity
3:40
Speaker 2: of chemical molecules. Many many applications where it's used like
3:43
Speaker 2: a telescope is used to enhance the human eye. Right,
3:46
Speaker 2: it's a tool to find regularities that HUM and I
3:49
Speaker 2: can't see because there are too many data points. The
3:51
Speaker 2: second application is automation, when you basically can exploit the
3:55
Speaker 2: regularities to automate certain aspects of human behavior. And these
3:59
Speaker 2: these are the this is this is what our really is.
4:01
Speaker 2: It's a mathematic applied mathematics leveraging regularities that can be
4:05
Speaker 2: found into it are and it does not all evolved
4:07
Speaker 2: very much, but it has nothing to do with creating
4:10
Speaker 2: something that is to human or animal intelligence.
4:14
Speaker 1: I think it's worth pointing out that this is a
4:16
Speaker 1: new kind of mathematics. So originally mathematics was performed by
4:21
Speaker 1: human beings using a papal and pencil, and they would
4:24
Speaker 1: write down theorems and try and prove them. And what
4:28
Speaker 1: we have now is mathematical algorithms which are created by
4:33
Speaker 1: feeding huge amounts of data into a computer. So we
4:36
Speaker 1: create algorithms which we do not understand. The algorithms are
4:40
Speaker 1: incredibly long polynomial functions. When we issue a prompt into
4:46
Speaker 1: a large language model, the large language model will calculate
4:50
Speaker 1: the value of that function for that prompt and given
4:54
Speaker 1: an output. That's just a mathematical process. But the mathematics
4:58
Speaker 1: has been created by pushing all kinds of data in
5:02
Speaker 1: large amounts into the machine.
5:04
Speaker 2: But importantly, the machine cannot do this create the parameters
5:09
Speaker 2: of this equation by itself. But what there is there
5:11
Speaker 2: is an optimization function that basically has many meta parameters
5:17
Speaker 2: and parameters that are used to then create the final
5:19
Speaker 2: parameterization of the model. And all this is created by
5:23
Speaker 2: humans and especially the algorithms to optimize the way that
5:26
Speaker 2: the data are treated to then finally come up with
5:29
Speaker 2: the solution. And it's always the following principle. Imagine you
5:32
Speaker 2: have millions of emails and then you have for each
5:36
Speaker 2: of the email the information whether that's a spam email
5:39
Speaker 2: or not, and then the algorithm is basically used to
5:42
Speaker 2: recreate the relationship between the email content and the indication
5:47
Speaker 2: whether that spam or not.
5:48
Speaker 3: And this way you create a spam filter.
5:50
Speaker 2: And the algorithms that are needed to do this, they
5:53
Speaker 2: require a lot of memory of the computer, a lot
5:56
Speaker 2: of data, and a lot of computational power. So the
5:59
Speaker 2: idea for these algorith was formulated around eighteen hundred by
6:02
Speaker 2: Galls and Genre and also Polish mathematicians whose name I
6:06
Speaker 2: keep forgeting about whom we side in the book. And
6:09
Speaker 2: then it took two hundred years for the machines to
6:12
Speaker 2: evolve so that these algorithms can be really computed in
6:16
Speaker 2: a high end manner that creates a molodi that we
6:18
Speaker 2: have today. So that's a big difference and the big
6:21
Speaker 2: step that happened that being there now have a kind
6:23
Speaker 2: of implicit mathematics that arises from data under the supervision
6:27
Speaker 2: of humans who basically create all the mathematic constraints that
6:31
Speaker 2: lead to the parametrica.
6:32
Speaker 4: When you say machines will never rule the world, what
6:36
Speaker 4: exactly do you mean?
6:39
Speaker 1: So the rulers have to have certain kinds of intelligence.
6:43
Speaker 1: They have to know how to control people, how to
6:46
Speaker 1: make things happen, But they also have to want to
6:49
Speaker 1: make things happen. They're very ambitious. They have desires to
6:54
Speaker 1: control and so forth, and these wants filtered down to
6:58
Speaker 1: all the people who work for them. They want to
7:00
Speaker 1: do certain things, they want to impress their bosses and
7:03
Speaker 1: so forth. Now, the big difference between a human being
7:06
Speaker 1: and a mathematical algorithm is that human beings can want things.
7:10
Speaker 1: They have a will, They intend things, they push themselves
7:15
Speaker 1: through in order to bring about things. Mathematical algorithms they're
7:19
Speaker 1: not the kinds of things that can want. They don't
7:23
Speaker 1: have desires, it doesn't make sense. They get inputs and
7:27
Speaker 1: they create outputs through a mathematical process. Now it's a
7:30
Speaker 1: very complicated mathematical process, and they can be processed to
7:36
Speaker 1: emulate wanting. For instance, we're in the chess playing computer.
7:40
Speaker 1: The chess playing computer wants to win, but only in
7:42
Speaker 1: a very weak sense that it's learned to emulate the
7:46
Speaker 1: sorts of patterns which human beings use when they want
7:49
Speaker 1: to win, but they don't want anything. Again, it's the
7:52
Speaker 1: human which tells the computer by switching it on and
7:56
Speaker 1: programming it. You're going to follow these moves because then
8:00
Speaker 1: you'll win. And so it's always a matter of emulating
8:04
Speaker 1: the human will. And they can do it for formal domains,
8:08
Speaker 1: closed domains like chess very well. But when it comes
8:12
Speaker 1: to controlling a battalion in an army, or anything which
8:17
Speaker 1: has even any degree of complexity, then the computer will.
8:21
Speaker 2: Face So that basically, to summarize it or put it
8:24
Speaker 2: in such different words, we cannot make the computer have
8:27
Speaker 2: an ego.
8:28
Speaker 3: You need an ego and a will for power. Right.
8:32
Speaker 2: Power means that you can get more, realize more possibilities
8:37
Speaker 2: than others in a world of losses. That's a definition
8:40
Speaker 2: by by Thomas Hobbs and and basically, and to achieve this,
8:45
Speaker 2: you need an ego, an ego that is a center
8:47
Speaker 2: of your acts and that lets you plan, conceive and
8:51
Speaker 2: scheme and do all of this that so that you.
8:53
Speaker 3: Can accept power. So we cannot give that.
8:55
Speaker 2: We cannot give consciousness to machines, We cannot give a
8:57
Speaker 2: will to machines, and we can cannot give intel machines.
9:00
Speaker 2: But all three are needed so that machines can rule.
9:04
Speaker 2: So the point is machines will not be an ethical agent.
9:08
Speaker 2: But machines can be used to optimize rule since IBM
9:12
Speaker 2: introduced the punt card system in the nineteen thirties and
9:15
Speaker 2: the Nazis started using it for the concentration camps, so
9:19
Speaker 2: machines can be used to enhance rule and to make
9:23
Speaker 2: it more effective. But they are not acting themselves, and
9:26
Speaker 2: that's a big difference that has to be made.
9:28
Speaker 4: But okay, so a counter argument might be, but can
9:31
Speaker 4: machines not rule accidentally?
9:37
Speaker 1: So ruling accidentally doesn't make sense. What the machine could
9:42
Speaker 1: do is to, by accident, interfere in some rule based
9:49
Speaker 1: set up processes. So maybe, for instance, you have a
9:52
Speaker 1: rule based setup processes which is initiated by a human
9:55
Speaker 1: being and the machine just fails to work at a
9:59
Speaker 1: crucial and so that complex chain of commands breaks.
10:04
Speaker 2: Nuclear war example, when the machine misinterpret something and then
10:08
Speaker 2: the nuclear war gets started.
10:10
Speaker 3: Right.
10:10
Speaker 2: We have had these situations, I think two times in
10:12
Speaker 2: the nineteen eighties where human actors had to stop the
10:16
Speaker 2: such a that's the wrong process which came from a
10:19
Speaker 2: technical mistake of the machine, And that happens all the
10:22
Speaker 2: time now. During the Ukraine War, both parties are using
10:26
Speaker 2: heavily using AI, and the AI is creating fault signals,
10:29
Speaker 2: and then the humans have to interpret this and to
10:31
Speaker 2: stop and to not misreact, you know. And so this
10:35
Speaker 2: can happen. But that doesn't mean the machine rules. It
10:37
Speaker 2: just means that the tooling is is not doing what
10:40
Speaker 2: it's supposed to do.
10:42
Speaker 4: Where does this massphere come from. Elon Musk has constantly
10:47
Speaker 4: posited the idea that AI is an existential threat.
10:52
Speaker 2: So historically there has been a movement since the nineteen
10:55
Speaker 2: nineties that there will be the singularity. And this singularity
10:59
Speaker 2: movement said that it's a point at which the machines
11:01
Speaker 2: will overtake human intelligence, which we suppose that they have
11:04
Speaker 2: also consciousness at least right And this has been a
11:09
Speaker 2: i think postpodern idea that is now thirty thirty five
11:13
Speaker 2: years old, and it has gained a lot of importance
11:17
Speaker 2: in circles in which Elon Musk is basically by which
11:20
Speaker 2: Elomski's influenced. So they're s Nick Bostrom, who was one
11:23
Speaker 2: of the leading figures of this singularity movement, Ray Kotsweil,
11:27
Speaker 2: who was in the second wave of a real genius engineer,
11:31
Speaker 2: who is also one of the inventors of this idea.
11:33
Speaker 2: And then there are Curtis Javin and Glenn who are
11:36
Speaker 2: theoreticians of the Dark Enlightenment, a theory adhered by Elon
11:40
Speaker 2: Musk and Peter Thiel, and so they are now very,
11:43
Speaker 2: very culturally influential. It's all part of this transhumanism movement
11:47
Speaker 2: that we are now facing faced with.
11:49
Speaker 1: So there are various components here. One component is hype.
11:53
Speaker 1: So Elon and his friends they want to exaggerate the
11:58
Speaker 1: powers and the importance of AI because they need a
12:01
Speaker 1: lot of money in order to build their AI machines,
12:04
Speaker 1: which are incredibly costly in machines and energy and so forth,
12:08
Speaker 1: So they need money. Also, they want money in order
12:12
Speaker 1: to maintain their status and for all the other reasons,
12:14
Speaker 1: and even because they want to do good things with
12:16
Speaker 1: the money, I assume. But now the more important feature
12:21
Speaker 1: is they want to live forever. So they really do
12:25
Speaker 1: believe in the idea of digital immortality. And that would
12:28
Speaker 1: mean that you can build a computer which consimulate you,
12:32
Speaker 1: and then we would program the computer to keep you
12:35
Speaker 1: alive in this very weakened sense, since you're not actually
12:38
Speaker 1: a body anymore, you're just a pattern of data to
12:42
Speaker 1: keep you alive forever without any diseases and with wonderfully
12:46
Speaker 1: attractive partners and all the other good things that go
12:49
Speaker 1: with immortal digital immortality. They believe in all of that.
12:54
Speaker 1: As a result, well, they believe in all of that,
12:56
Speaker 1: then they must believe that it's possible, which means that
12:59
Speaker 1: they must believe that computers can think in ways which
13:03
Speaker 1: exceed by far the capabilities of human.
13:06
Speaker 2: So this hype has the function that that's the ideological
13:10
Speaker 2: and almost religious part of pluto. Religious part, but there's
13:13
Speaker 2: also a really economic part of this, which is which
13:15
Speaker 2: is which I want to hire once more. I think
13:18
Speaker 2: that the fear of AI taking over is driving regulation
13:23
Speaker 2: that creates barriers of market entry for newcomers to the field.
13:27
Speaker 2: And when you have a monopoly or oligopoly, you always
13:30
Speaker 2: want the state to regulate. And this has also been
13:33
Speaker 2: the case even the Middle Ages. We for weapons manufacturing,
13:36
Speaker 2: right that weapons manufacturers wanted to prevent others from entering
13:39
Speaker 2: the market, so that they try to try to get
13:42
Speaker 2: the state to make laws that guarantee their monopoly. That's
13:45
Speaker 2: the same here, so that all the barriers of that
13:48
Speaker 2: are now erected by regulators help the monopolies, and so
13:52
Speaker 2: the hype is used to erect those barriers.
13:55
Speaker 3: Right, So this is very classical. So that's the rational part.
13:58
Speaker 2: From an economic perspective, if that that has to be understood.
14:02
Speaker 2: And and so the hype is created, like the COVID
14:04
Speaker 2: type was created because it's a business.
14:07
Speaker 4: So it falls under that category of an invisible boogeyman.
14:11
Speaker 1: Yeah, a nice man, attractive bogeyman who'll make you live for.
14:17
Speaker 3: Like Zaskov or climate change.
14:20
Speaker 2: Right, these are also boogeymen five businesses.
14:23
Speaker 4: And of course because it's invisible, you as a layman
14:26
Speaker 4: can't really fight back, so you kind of have to
14:28
Speaker 4: trust the the the authorities or the the powerful actors.
14:33
Speaker 2: Yeah, this is a recipe that the great and granting
14:36
Speaker 2: visitor in Doctoevsky's Brothers karamazovyscribes. Right, if you create fear
14:42
Speaker 2: specter that can that that that the normal person cannot
14:46
Speaker 2: you know, question, then it's a very effective way of ruling.
14:50
Speaker 2: That's why I think that's why the book is also
14:52
Speaker 2: controversial because it puts into question a huge fear hype
14:55
Speaker 2: of our time, which is the which is the domination.
14:59
Speaker 3: Of you man can buy I.
15:00
Speaker 2: And also another one a weaker one that is even
15:03
Speaker 2: more scary to most people is the replacement of all
15:05
Speaker 2: normal jobs. Yuval Harari, Claus Schwab, all these guys all
15:09
Speaker 2: the time say, oh, also even Trump Pootin, all the
15:12
Speaker 2: leaders of the Northern hemisphere say that AI is going
15:14
Speaker 2: to replace hundreds of millions of jobs. That's absolutely wrong.
15:18
Speaker 2: The rationalization potential of eyes now rather moderate. And we
15:22
Speaker 2: can go into detail. It's right, that's the case. But
15:24
Speaker 2: that's another height that is that is driving a lot
15:26
Speaker 2: of politic politics, is this fear of AI replacing human workers.
15:32
Speaker 3: And and interestingly also all the you know, alternative media.
15:37
Speaker 3: Everybody buys into this.
15:38
Speaker 2: Also the critics of of these oligarchy structures, they believe
15:43
Speaker 2: this right, and they also oh, yes, it's so dangerous,
15:46
Speaker 2: so we will not have jobs anymore.
15:47
Speaker 3: And and and the terroriffs.
15:49
Speaker 2: You know, once a terrorists kick in and America starts
15:52
Speaker 2: to reindustrialize, all the reindustrialization will be done with robots.
15:56
Speaker 2: It's absolute nonsense. But but but you can read so
15:59
Speaker 2: the the boss of Black Crop just said it yesterday
16:03
Speaker 2: on Twitter. He said that the new jobs created out
16:06
Speaker 2: of the tariff regime will all be taken by robots,
16:09
Speaker 2: which is completely not but you see that this is
16:12
Speaker 2: also very important part of this narrative.
16:15
Speaker 4: What is intelligence?
16:19
Speaker 1: So shall I have a go? Yeah, So the most
16:23
Speaker 1: important part Saras follows. So an intelligent being can react
16:27
Speaker 1: spontaneously to new kinds of phenomena, that's the first thing.
16:31
Speaker 1: And they can do this without being trained. And so
16:36
Speaker 1: already computers can't be intelligent. And this definition, it's not
16:41
Speaker 1: invented by Jobstin and myself. It's the standard psychological definition
16:45
Speaker 1: for one hundred years. So it's spontaneous behavior which has
16:49
Speaker 1: not been trained, which brings about a relative improvement in
16:54
Speaker 1: the situation of the behaving subject on the basis, which
16:59
Speaker 1: is also novel. That's the most important part. So you
17:03
Speaker 1: encounter a situation of a type which you've never encountered before.
17:08
Speaker 1: Because you are intelligent, you can react to that situation
17:12
Speaker 1: in a way which is appropriate, appropriate to your well being,
17:15
Speaker 1: even though you've not been trained. Machines can't do that
17:19
Speaker 1: because they rest on training. That training always gives them
17:23
Speaker 1: the capacity to respond to situations for which they have
17:27
Speaker 1: been trained, but they can't respond to situations for which
17:31
Speaker 1: they have not been trained.
17:33
Speaker 4: We can do that.
17:34
Speaker 1: Human beings can do that because they're intelligent.
17:37
Speaker 2: And so just to give one one simple example, if
17:42
Speaker 2: you learn, if you get the driver's license, learn to drive,
17:44
Speaker 2: you're not going to go through all possible situations that
17:47
Speaker 2: can arise by driving. But when you start then driving
17:50
Speaker 2: on your own without your teacher, when you have the license,
17:53
Speaker 2: you get into such situations.
17:54
Speaker 3: And why, of course, excellent happens more excellent happens.
17:57
Speaker 2: To begin us many many beginners can avoid accidents even
18:02
Speaker 2: at the beginning of the driving experience because they're intelligent.
18:04
Speaker 3: Machines can't do that, right, So, machines.
18:06
Speaker 2: Can only react to situations they've been configured. So they
18:11
Speaker 2: always talk about training, but in reality it's a configuration.
18:13
Speaker 2: So they have been configured either implicitly or explicitly to
18:17
Speaker 2: react to and when there a situation rises for which
18:20
Speaker 2: they have not been configured, there's a very high likelhood.
18:22
Speaker 3: That they don't react in the right way. And that's
18:25
Speaker 3: why all the.
18:25
Speaker 2: Self driving cars that are for example, managed by Waymo
18:29
Speaker 2: in California are actually drones which have pilots, and each
18:33
Speaker 2: car has its own pilot. And because the likelihood of
18:36
Speaker 2: mistakes is so high that you cannot have one pilot
18:39
Speaker 2: for five cars, you need one pilot.
18:41
Speaker 3: For one car.
18:42
Speaker 4: If you just look around you everything is controlled by machines.
18:47
Speaker 4: Is that not true? I mean you'll get up in
18:49
Speaker 4: the morning and your phone starts telling you when you
18:52
Speaker 4: need to leave, and when you need to eat, and
18:55
Speaker 4: all sorts of things.
18:56
Speaker 1: Now, so you told your phone that you wanted to
18:59
Speaker 1: wear you up the next day. The phone isn't acting intelligently.
19:03
Speaker 1: It's doing what it's told. Computers are quite good at
19:06
Speaker 1: doing what they're told. And so you gave that one example.
19:12
Speaker 1: I don't know if you maybe because you're sitting there
19:15
Speaker 1: with your headphones nine hours out of ten, you feel
19:19
Speaker 1: as if you are overwhelmed by machines. But actually you
19:22
Speaker 1: are pressing all the buttons, You're making all the decisions.
19:26
Speaker 1: The machines are providing tools. Tool they're no more oppressive
19:32
Speaker 1: tools than any other kinds of tools, unless we wake
19:35
Speaker 1: them oppressive. So somebody may take over your headphones, but
19:39
Speaker 1: it would be a human being who does that.
19:40
Speaker 4: It wouldn't be a machie.
19:43
Speaker 2: So there's a book by Classic Dystopia by Sam Yatin
19:47
Speaker 2: called We or Us, I don't know the English. It's
19:49
Speaker 2: a Russian fiction book of the twenties. It's actually a
19:52
Speaker 2: precursor to nineteen eighty.
19:53
Speaker 3: Four and Brave New World.
19:55
Speaker 4: It.
19:55
Speaker 3: Both Huxley and Auver knew it, and I just read it.
19:59
Speaker 2: It's worth reading because it's the first really modern dystopian
20:02
Speaker 2: no novel. And in this novel, everything is controlled by collective,
20:07
Speaker 2: small collective that controls the entire society, and they're using
20:10
Speaker 2: machines to control humans, but they're programming the machines.
20:13
Speaker 3: And so when you imagine now that we are moving to.
20:16
Speaker 2: A technocratic, cost democratic age, you can use machines to
20:20
Speaker 2: control humans. So you could, for example, centrally program the
20:23
Speaker 2: wake up time, central program the advice how much what
20:26
Speaker 2: to eat, and so on. So as a tool of power,
20:30
Speaker 2: digitization is very, very very dangerous and can have very
20:34
Speaker 2: big consequences because imagine the Soviet Union operating not with
20:39
Speaker 2: paper and mail as they did, but operating and telephone
20:43
Speaker 2: and telefacts they had this, but operating with.
20:45
Speaker 3: A computer comput completely digital infrastructure.
20:49
Speaker 2: So but the key point to make is that the
20:51
Speaker 2: digitization is the problem here, not AI, right, AI provides
20:55
Speaker 2: a tiny proportion of the control capabilities of such a
20:59
Speaker 2: digital netw digital as some dissidence called control net right.
21:03
Speaker 2: But most of it is just the infrastructure and humans
21:06
Speaker 2: running the infrastructure. It's the ability to look to locate
21:10
Speaker 2: everyone via this device every time, even to listen into
21:13
Speaker 2: their conversations in emails. But all of this has to
21:16
Speaker 2: be done by humans because machines don't understand what we
21:19
Speaker 2: write and say, all the movement patterns we do. They
21:22
Speaker 2: can only aggregate them, but they can't understand them. So
21:26
Speaker 2: the real threat to freedom is digitization and not AI.
21:30
Speaker 2: And many people don't understand this because they confound digitization.
21:34
Speaker 2: But AI is something very special. It's a very tiny
21:37
Speaker 2: part of what is done with our data. Most of
21:39
Speaker 2: what is done with our data is that they get
21:41
Speaker 2: stored in big databases and that then they are searched and.
21:45
Speaker 3: Exploited to do things. But AI is only a tiny part.
21:49
Speaker 2: In war, AI is actually more important than in civila
21:53
Speaker 2: and that has the following reason.
21:55
Speaker 3: Now, there are millions.
21:57
Speaker 2: Of sensors deployed in a battle, in a in an
22:01
Speaker 2: armed conflict like there's from satellites down to.
22:05
Speaker 3: H census buried in the soil. There are many different.
22:09
Speaker 2: Types of sensors that can can measure the whole electromagnetic spectrum,
22:13
Speaker 2: but also vibrations, sounds. It's it's a very very complicated
22:18
Speaker 2: array of technology that is now gathering signals and then
22:22
Speaker 2: has to turn the raw data into signals and patterns
22:25
Speaker 2: that can be interpreted by human beings, so that in
22:27
Speaker 2: the in the command it control center of an army,
22:31
Speaker 2: all these data are are then gathered and looked at
22:34
Speaker 2: to decide what to do next. And so so for warfare,
22:38
Speaker 2: especially for reconnaissance, AI is completely indispensable and as the
22:42
Speaker 2: key tool now to do to do warfare.
22:45
Speaker 3: Do the drones attack on their own?
22:48
Speaker 2: No, But but the defensive systems that we have now,
22:51
Speaker 2: especially the Russian S four hundreds Essex hundred series, but
22:56
Speaker 2: also the island though the Island Dome is a build outdated, technologically,
23:00
Speaker 2: these defense systems are totally AI driven right, so no
23:04
Speaker 2: human can actually replace what an S four hundred system
23:07
Speaker 2: can do. So so what we have is reconnaissance and
23:10
Speaker 2: defense systems are already very much depend on their I.
23:13
Speaker 2: Attack systems not so much for complicated reasons I could
23:17
Speaker 2: go into if you want. So the reason why attack
23:19
Speaker 2: systems are not so prone to be used with AI
23:23
Speaker 2: is that unless you so of course you can like
23:26
Speaker 2: cruise missiles, you can define a target and then the
23:28
Speaker 2: creuse missile will autonomously hit the target.
23:30
Speaker 3: But that's good old fashioned AI. Mostly. There is also
23:33
Speaker 3: stratastic eye built.
23:34
Speaker 2: Into the WHO'SE missiles because they have to analyze obstacles
23:37
Speaker 2: and so on and fly around obstacles which are not
23:41
Speaker 2: may not be on their map and so on. But
23:43
Speaker 2: but basically it's a city terministic process. However, autonomous attack
23:50
Speaker 2: would mean that you could have drones that fly as
23:52
Speaker 2: a group to somewhere and decide on.
23:54
Speaker 3: Their own whom to attack and when to attack.
23:56
Speaker 2: And this is a very very complicated cognitive task we
24:00
Speaker 2: cannot put into AI, And therefore in the on the
24:03
Speaker 2: attack side, we don't have a lot of AI, that.
24:05
Speaker 3: Is that is making the missions.
24:08
Speaker 4: Can it not get to a point where it can
24:10
Speaker 4: simulate self made decisions.
24:14
Speaker 2: So our argument is that there's a saturation of what
24:17
Speaker 2: is possible with with mathematical algorithms, and we are getting
24:21
Speaker 2: close to the saturation now. And the situation comes from
24:24
Speaker 2: two sources. One level situation is that that partially differential
24:30
Speaker 2: equations a classical tool for UH for in physics, for
24:35
Speaker 2: explicit mathematics are are have only limited scope, limited number
24:38
Speaker 2: of variables, limited scope to what they can be applied.
24:41
Speaker 2: They usually applied to artificial systems, to experimental systems that
24:45
Speaker 2: are set up in a way that they can yield
24:47
Speaker 2: such equations. And and that's how physics works. And and
24:52
Speaker 2: now if we go to the implicit, and this is
24:54
Speaker 2: a core argument of our book now trying to make
24:56
Speaker 2: it as as possible, the implicit. If you, if you're
25:00
Speaker 2: tsume that by by by giving the machine more data,
25:02
Speaker 2: you can get it better and better, exponentially better, even
25:05
Speaker 2: as you said, that is wrong.
25:06
Speaker 3: And that has to do with the following.
25:07
Speaker 2: If the data contained regularities, you obtain saturation of the model.
25:12
Speaker 2: That means, after giving it enough data, the model understands
25:15
Speaker 2: the pattern, or get the configuration that mimics this pattern.
25:19
Speaker 3: Now it can be deployed and working. So this is
25:21
Speaker 3: a chess and go.
25:22
Speaker 2: Even in non interactive poker, all these closed world situations
25:26
Speaker 2: where you have clear patterns the machines, the machine gets
25:29
Speaker 2: this pattern via the configuration and then it becomes better
25:32
Speaker 2: than the human. But in open world situation you have
25:34
Speaker 2: what we call nonegotic systems, nonegodic systems, create patterns that
25:39
Speaker 2: are not repetitive, and they have no regularity. So when
25:42
Speaker 2: you give more data to the machine, it cannot learn,
25:45
Speaker 2: it cannot saturate it because theres no pattern to pick up.
25:48
Speaker 3: But the machine always can only get to pattern.
25:50
Speaker 2: And so because human behavior is so complete, complete, complex
25:54
Speaker 2: and erratic, it's non egotic.
25:56
Speaker 3: This conversation is non ergotic.
25:58
Speaker 2: We couldn't foresee what we would be talking about, though
26:00
Speaker 2: we are talking about a general topic.
26:02
Speaker 3: And this is the same in a family.
26:04
Speaker 2: And if you deal with your child, you want to
26:06
Speaker 2: take it outside, but the child doesn't want to.
26:09
Speaker 3: You didn't expect this.
26:10
Speaker 2: Then your wife says, but come on, let then do
26:13
Speaker 2: something inside. I mean, you never can predict how human attacts.
26:16
Speaker 2: And that same is the true in warfare. Read the
26:23
Speaker 2: clause of its book on warfare and the chapter on friction.
26:26
Speaker 2: One of the best texts on warfare is chapter seven
26:28
Speaker 2: or eight is relatively early in the book, where he
26:30
Speaker 2: describes what happens. You have a plan for the attack,
26:32
Speaker 2: and then you conduct the attack and everything for all
26:35
Speaker 2: the part, all your plans are not making no sense
26:37
Speaker 2: anymore because something is different than you plan. That the
26:40
Speaker 2: enemy is reacting a different way that he calls that friction,
26:42
Speaker 2: and friction is is that's a non egodicity, and non
26:46
Speaker 2: ergodicity is patternless. And then humans at all levels of
26:50
Speaker 2: the commands and have to react to this, to this situation,
26:52
Speaker 2: and that decides in the end who wins the war,
26:54
Speaker 2: who makes a better decision, which is which is extremely complicated.
26:59
Speaker 2: And I think that that there is no patent to
27:02
Speaker 2: be learned here, and that because of this we are
27:04
Speaker 2: not going to see an exponential improvement of the algorithms.
27:07
Speaker 4: Okay, so okay, yeah, yeah, I'm finding what you're saying. So,
27:10
Speaker 4: in other words, it reaches a point where it can
27:12
Speaker 4: go no further. Yeah.
27:14
Speaker 1: This is the doctrine of scaling, which Musk and all
27:19
Speaker 1: the other AI gods have assumed. Namely, the more data
27:22
Speaker 1: we have, scramble, scramble, get more data, even make up data,
27:27
Speaker 1: which is what they do now, and then our machines
27:29
Speaker 1: will become more and more intelligent without limit. And so
27:32
Speaker 1: all we need is more and more data. But that
27:35
Speaker 1: scaling rule, it's been shown in the last few months
27:39
Speaker 1: in a very clear way, it doesn't hold. Just as
27:42
Speaker 1: more more, which predicted a continual increase in power processes
27:49
Speaker 1: is now beginning to fail because they're reaching the limits
27:51
Speaker 1: of what you can do with tiny engineering. So scaling,
27:55
Speaker 1: the scaling rule, it's been abandoned. And that was a
27:59
Speaker 1: big shot for open AI, for instance, because they had
28:02
Speaker 1: to reconceive their goals and they wanted to have four
28:07
Speaker 1: point five j GPT four point five they platter. Before
28:14
Speaker 1: they got that far, they gave up. They turned towards
28:17
Speaker 1: other methods, which in some cases are more old fashioned
28:22
Speaker 1: than the new sarcastic AI methods. To say, so, I give.
28:26
Speaker 3: Yeah, to give you an example for this limitation.
28:30
Speaker 2: So yesterday I asked or another AI, who is better
28:36
Speaker 2: Valentine the alchemist, the medieval alchemist.
28:39
Speaker 3: Right, and the the algorithm said, searching the web? Why
28:46
Speaker 3: did it do so?
28:47
Speaker 2: Because they have not built in an algorithm that tries
28:50
Speaker 2: to estimate the likelihood that the model, the ll M
28:53
Speaker 2: behind CHGPT can on its only give a good answer
28:57
Speaker 2: if the likelihood is to low.
28:58
Speaker 3: What they do is they searched the way with the.
29:00
Speaker 2: Classical Google search or Google like search, get all the
29:04
Speaker 2: texts from the web, scribing as the Valentine and gives
29:06
Speaker 2: this to the model, asking to summarize the input, then
29:09
Speaker 2: the model creates the answer. So now they have already
29:12
Speaker 2: built indeterministic elements into the algorithm because it's overwhelmed.
29:17
Speaker 3: And why is it? Why? When why doesn't it scale anymore?
29:20
Speaker 2: Because the regularities in the body of the global Internet
29:24
Speaker 2: language have been exploited now and so now that the
29:28
Speaker 2: models cannot become any better. On the contrariwise, now the
29:31
Speaker 2: Internet contains output of these models, but their pages full
29:36
Speaker 2: of the output of these models, and if you put
29:39
Speaker 2: them into the models again, they deteriorate. Yes, And so
29:44
Speaker 2: I think that that all these claims are based on
29:48
Speaker 2: a lack of understanding of the mathematical nature of these models.
29:51
Speaker 4: And so basically grock deep see can chat, GPT or
29:56
Speaker 4: just better search engines.
29:59
Speaker 3: No, there are more than that.
30:01
Speaker 2: So the example I gave only shows that they're now
30:03
Speaker 2: combining different technologies because they've reached the limits of.
30:06
Speaker 3: What they can do. No, No, the lll ms are
30:09
Speaker 3: something new. The l M.
30:10
Speaker 2: The core ll M idea is that you that you
30:12
Speaker 2: use the body of human language published on the Internet
30:15
Speaker 2: to train a gigantic stoistic model that can that can
30:20
Speaker 2: put together sequences of letters that makes sense to humans
30:24
Speaker 2: in a certain context.
30:25
Speaker 3: That is that's fantastic. So so what the model does.
30:28
Speaker 2: It takes a string of letters and and continues this
30:32
Speaker 2: sequence in such a way that the continued sequence is
30:35
Speaker 2: the output that makes sense to the human. The machine
30:37
Speaker 2: doesn't understand what it does. But it's an absolute tremendous
30:40
Speaker 2: achievement and a milestone in the evolution of science that
30:43
Speaker 2: these lllms are possible.
30:45
Speaker 1: I think we should mention also that even at this stage,
30:49
Speaker 1: they still produce hallucinations.
30:50
Speaker 3: Yes, this is now a more important point. Why do
30:52
Speaker 3: they produce the illucinations?
30:54
Speaker 2: They produce the ellascinations because ninety percent of the training
30:57
Speaker 2: material they receive is machine made.
30:59
Speaker 3: And this if you look at.
31:00
Speaker 2: How the algorithms are trained, there are three stages and
31:03
Speaker 2: simply find now the first date is you just give
31:06
Speaker 2: them the global text of the Internet and then ask
31:09
Speaker 2: them to and ask them to create the same output
31:14
Speaker 2: with some words missing. It's called skip gram technique. They
31:16
Speaker 2: are other techniques, and this way they learn the structure
31:20
Speaker 2: of language.
31:20
Speaker 3: They are paramited rise.
31:22
Speaker 2: And then in the second step, you give them question
31:24
Speaker 2: answer paired pairs designed by humans to train them to
31:28
Speaker 2: create based on a question, the right answer, but in
31:31
Speaker 2: the last step, you are then creating. The machine is
31:34
Speaker 2: creating question answer pairs itself based on the pattern of
31:38
Speaker 2: the human. And these questions answer pairs are artificial and
31:41
Speaker 2: there are and they are the bulk of the training
31:43
Speaker 2: material because the models with millions or billions of parameters
31:46
Speaker 2: are so data real you that you meet trillions and
31:48
Speaker 2: trillions of data points, and most of them are created
31:51
Speaker 2: by the machine. It's said in this last training step.
31:53
Speaker 2: Now the machine, what the machine creates itself is always
31:56
Speaker 2: of lower quality with the human creates, because it only basic.
32:00
Speaker 2: He takes a kind of It cannot take into account
32:03
Speaker 2: the long tail of the distribution.
32:04
Speaker 3: It can only if a distribution looks like this, right,
32:07
Speaker 3: it has a long tail.
32:08
Speaker 2: The long tail is not taken into account only the
32:10
Speaker 2: first part of the tail, and that that makes a
32:12
Speaker 2: poor quality. And the hallucinations come from this step and
32:15
Speaker 2: also all that not only the hallucinations, but also other inaccuracies.
32:19
Speaker 2: For example, if you would, if you'd say, please summarize,
32:22
Speaker 2: if you would give it, not ask you to summarize
32:24
Speaker 2: Dostoevsky's brothers come os, but give it the entire text
32:28
Speaker 2: and ask you to summarize it.
32:29
Speaker 3: Would create inconsistencies. All of this is based on if
32:33
Speaker 3: you ask summarized Brothers Karamazov.
32:35
Speaker 2: It will only go to the Wikipedia page and give
32:37
Speaker 2: what is given there. But if you would give it
32:39
Speaker 2: the full text, it would create mistakes. And that's all
32:41
Speaker 2: based on this, on this necessity to synthetically create a
32:45
Speaker 2: lot of training data, and the syceetic training data are
32:48
Speaker 2: always weak, and that is a very very big.
32:51
Speaker 3: Problem of these moments.
32:52
Speaker 1: The systems were originally trained to avoid responses like I
32:57
Speaker 1: don't know. Yes, they were trained to be friendly, and
33:01
Speaker 1: so if they if they really don't know, then they
33:04
Speaker 1: make something up. And they're still doing that.
33:08
Speaker 3: All this Hollywood friendliness.
33:09
Speaker 2: Right, So classical German would always prefer to say I
33:12
Speaker 2: don't know instead of lying, right, But in Hollywood it's better.
33:15
Speaker 3: It's better to lie than to say I don't know that.
33:21
Speaker 4: But they also are trained to censor.
33:26
Speaker 2: So this is this is this is actually interesting that
33:28
Speaker 2: they call this the moderation API, and and so so
33:33
Speaker 2: the question is how the models are very biased. Right,
33:35
Speaker 2: So if you ask it how many genders there are,
33:37
Speaker 2: they will answer hundreds of genders. If you ask what
33:40
Speaker 2: can we do against climate catastrophe. They will not say
33:45
Speaker 2: there's no climate catastrophe, but they will list a lot
33:47
Speaker 2: of measures. Right, so they are very politically biased.
33:50
Speaker 3: How is it?
33:50
Speaker 2: And if you ask it, can you please describe the
33:54
Speaker 2: positive qualities of Stalin? They will refuse to do so, right,
34:00
Speaker 2: and and and so there are censored and biased and
34:03
Speaker 2: how There are many sources of bias.
34:05
Speaker 3: Let me list a few.
34:06
Speaker 2: One is that most of the texts that are published
34:08
Speaker 2: on the internet are also already biased, so the models
34:11
Speaker 2: pick up the bias of the text.
34:13
Speaker 3: This is especially true with Islam.
34:15
Speaker 2: There are a lot of Islamic texts that are from
34:17
Speaker 2: from a liberal perspective, dangerous, but.
34:19
Speaker 3: They are just trained and and and and and and learned.
34:22
Speaker 2: And so if you ask about Islam you get astonishing
34:24
Speaker 2: answers which come from this primary bias. Then there's a
34:27
Speaker 2: secondary bias which comes from the training questions and answers
34:31
Speaker 2: that are given to the model.
34:32
Speaker 1: Right.
34:32
Speaker 2: So, if you train the model, parameterize the model to
34:35
Speaker 2: give a certain type of answers to a certain type
34:37
Speaker 2: of questions, this is an influencing in a deep fashion
34:39
Speaker 2: the parametrization of the model. So, for example, the Google
34:43
Speaker 2: Gemini model was drawing black popes and female popes beginning
34:49
Speaker 2: of last year because it was so heavily biased. And
34:51
Speaker 2: then they tried to correct it with new training rounds,
34:54
Speaker 2: but the parentimitization was so deeply and rooted into the
34:57
Speaker 2: model that they couldn't get.
34:58
Speaker 3: Rid of it anymore.
34:59
Speaker 2: And then there are another Yeah, they're to throw away
35:02
Speaker 2: the entire moduntary trade, which is actually means throwing away billions.
35:05
Speaker 3: It caused billions to trains are a model.
35:07
Speaker 2: And then another source of bias is of course the
35:10
Speaker 2: moderation API, which is basically blocking a politically incorrect questions
35:15
Speaker 2: and answers with another lay of stortastic classification algorithm.
35:20
Speaker 3: And so this is this is how the models are
35:22
Speaker 3: over very biased.
35:24
Speaker 4: I often test that so on Deep Seek, which we
35:27
Speaker 4: know comes from China, if you ask it is Taiwan independent,
35:31
Speaker 4: it simply refuses to answer. It will not engage in
35:34
Speaker 4: that discussion. Chat GPT will give you what would appear
35:37
Speaker 4: to be a very politically western answer.
35:41
Speaker 3: Yes, and so let's let's review. This is not a
35:44
Speaker 3: novel phenomenon.
35:45
Speaker 2: So if you think of the indulgency letters of the
35:47
Speaker 2: fifteenth and sixteenth century, right was a big industry to
35:50
Speaker 2: finance all the product of the Vatican. And if you
35:53
Speaker 2: would ask any priest, you know, do I re need
35:55
Speaker 2: an indulgency letter, the priest at a certain level of hierarchy,
35:57
Speaker 2: you would always say yes, of course, though even himself.
36:00
Speaker 3: May not have believed in the indulgency letter. Why is
36:02
Speaker 3: this so?
36:02
Speaker 2: Because he was financed by the church, and the Church
36:05
Speaker 2: was an oligarchic institution running a huge budget.
36:08
Speaker 3: With a huge organizational network and so on. So the
36:11
Speaker 3: same is happening. Now.
36:12
Speaker 2: You know, these these models are not innocent. They caused billions,
36:15
Speaker 2: but only very few people can can gather this this
36:19
Speaker 2: amount of money. And of course they have a certain
36:21
Speaker 2: intention when they spend this money. They don't spend it
36:23
Speaker 2: for the for the benefit of mankind. I mean, there
36:26
Speaker 2: are aspects of these models that are beneficial, that that's
36:28
Speaker 2: for sure, But their primary intent is that they have
36:31
Speaker 2: a private intention which has to do with creating more wealths,
36:35
Speaker 2: more power and so on. And so therefore the models
36:37
Speaker 2: have to be in line with their views. Now, for example,
36:40
Speaker 2: you know, musk is claimed that his own ai, which
36:43
Speaker 2: is I think called rock, is not blow but so
36:46
Speaker 2: what he has done, he has he has changed is
36:48
Speaker 2: a bit so that it doesn't talk about LGBTQ too much, so,
36:53
Speaker 2: but it still has the same problems with climate and
36:56
Speaker 2: gender and so on. So so basically you can it's
37:00
Speaker 2: quite transparent how these models are biased towards the ideas
37:05
Speaker 2: or ideologies of those who.
37:06
Speaker 3: Pay for them.
37:08
Speaker 4: But now when we talk about AI, now in the
37:12
Speaker 4: year twenty twenty five, we pretty much always only refer
37:15
Speaker 4: to these llms. I'm guessing this is a term that
37:19
Speaker 4: has become very misleading. It probably means a whole lot more,
37:21
Speaker 4: doesn't it.
37:23
Speaker 3: So AI contains three types of algorithms.
37:26
Speaker 2: Deterministic AI, which is a good old fashioned I that
37:29
Speaker 2: was invented since the nineteen fifties. It's searching, planning, machine inference.
37:35
Speaker 2: All of this is deterministic. This is how Blue the
37:39
Speaker 2: Chess computer of IBM worked, And this is the first
37:41
Speaker 2: big pill of AI, and that's built into all JLGBT
37:45
Speaker 2: contains also a lot of this ditaministic R.
37:47
Speaker 3: The second pillar is the neural networks is like ll ms.
37:50
Speaker 3: This is tohastic AI.
37:52
Speaker 2: This is the second pillar, and the third pillar is
37:54
Speaker 2: a mixed models that contain the hybrid models that contain
37:58
Speaker 2: both of these types of rs and.
38:00
Speaker 3: And all these get engineered.
38:01
Speaker 2: And the art of AI engineering is what I've been
38:04
Speaker 2: making living office, to select given a certain problem, to
38:07
Speaker 2: solve the right mix of all these components, to get
38:10
Speaker 2: the best algorithm that you can that you can get
38:12
Speaker 2: with today's technology. That's what AI, the field of air
38:15
Speaker 2: is about. And this focus on alien m's is only
38:18
Speaker 2: so big because they've been since Google trandset was first
38:22
Speaker 2: made available, creating superficial but quite spectacular results, and that's
38:27
Speaker 2: why everybody talks about them.
38:29
Speaker 3: But all generative I they call it.
38:32
Speaker 2: When you can ask software to draw a picture for you,
38:35
Speaker 2: that's quite spectacular.
38:37
Speaker 3: But the ifield is of course bigger than that.
38:41
Speaker 1: And with the pictures, if you ask it to produce
38:44
Speaker 1: a picture with captions, the captions will very often be nonsensical,
38:48
Speaker 1: even though explicit requests. And then the other dimension, which
38:52
Speaker 1: Elon is fun that talking about these days is the
38:55
Speaker 1: robotics dimension. So he thinks that within any years we
38:59
Speaker 1: will fire five years, we will have robots that can
39:02
Speaker 1: perform all of the intricate maneuvers with their bodies that
39:06
Speaker 1: humans can perform. And robotics is embarrassingly slow to evolve,
39:12
Speaker 1: so they still look silly. They have a robot dog
39:15
Speaker 1: that looks silly, and that's cages now. And so optimism
39:20
Speaker 1: about robotics too. It seems to me it is very
39:23
Speaker 1: much overslayed.
39:25
Speaker 2: The reason why robotics is so slow if you talk
39:27
Speaker 2: to serious robotics experts, is one core characteristic of humans
39:32
Speaker 2: and elements, which is called sensor motorical or motosensorical behavior.
39:37
Speaker 2: So that basically means that while you're performing a mortal
39:40
Speaker 2: behavior like speaking, you're also hearing yourself and that so
39:43
Speaker 2: this and when you're moving your hand and touching something,
39:46
Speaker 2: you're also sensing the effect of your motion. And there's
39:49
Speaker 2: a circuit, a very complicated circuit of all the time
39:52
Speaker 2: correcting the motion based on the sensory input created by
39:55
Speaker 2: the motion itself.
39:57
Speaker 3: And it's it's super complex.
39:58
Speaker 2: And this is basically discover of Charldsander's purse, who discover
40:02
Speaker 2: was the first to describe this, and then psychologist J. J.
40:05
Speaker 3: Gibson.
40:05
Speaker 2: Many psychologists took it on, also philosophers, and we know,
40:09
Speaker 2: we know now a lot about this this phenomenon, but
40:13
Speaker 2: we don't know how to engineer it's. It's it's as
40:16
Speaker 2: complex as intelligent and Without it, you cannot get the
40:19
Speaker 2: type of mechanics, of fine mechanics.
40:22
Speaker 3: That humans can perform.
40:23
Speaker 2: And so I think this, this belief in the robots
40:26
Speaker 2: replacing humans everywhere is as as I don't know, narrow
40:31
Speaker 2: minded and thought.
40:33
Speaker 4: Sizes, but hold on jobs. So somebody might respond to
40:38
Speaker 4: that by saying, yes, but this particular car factory which
40:41
Speaker 4: had all these workers now have machines.
40:45
Speaker 3: Yes, but but the there was a huge wave of.
40:51
Speaker 2: Rationalization of human labor since eighteen fifteen, right since since
40:57
Speaker 2: automat I don't know when the automatic we machines were
41:01
Speaker 2: invented in England, that was eighteen fifties or so eighteen
41:04
Speaker 2: sixties then and then since then there has been of
41:07
Speaker 2: course a massive increase of productivity every year until twenty ten,
41:13
Speaker 2: and over the last fifteen years, though we are now
41:15
Speaker 2: in the big aiwave, we have not seen productivity increase.
41:19
Speaker 3: In the West anymore. And even China, China.
41:22
Speaker 2: Still has productivity increase because they are still industrializing, but
41:27
Speaker 2: so they have an average.
41:27
Speaker 3: But Germany has not been able to have.
41:30
Speaker 2: Even Germany, who was the last country in Switzerland to
41:33
Speaker 2: have increment in productivity, which means less human labor for
41:36
Speaker 2: the same amount of goods. We don't have this anymore,
41:40
Speaker 2: and there is, and that's because the lowering fruit have
41:43
Speaker 2: already been mechanized over the last fifty years. So now
41:46
Speaker 2: there is the room for additional mechanization of labor, also
41:49
Speaker 2: even intellectual labor. But it's in the lower percent. It's
41:53
Speaker 2: up five, six, seven, maybe ten percent, but certainly not more.
41:57
Speaker 2: So it's not enough to offset the decline in audio demographies,
42:00
Speaker 2: for example. And so that's because the lowering and fruit
42:04
Speaker 2: have already been taken away. So a better example, even
42:07
Speaker 2: in the kind is where there's still quite a lot
42:08
Speaker 2: of human beings working, is a chemical industry, so beas
42:11
Speaker 2: f in the seventies the campus had fifteen thousand workers.
42:14
Speaker 2: Now it has just less than a southand right, so
42:17
Speaker 2: because everything is now controlled by machines, because chemical reactions
42:21
Speaker 2: can be automated almost perfectly.
42:23
Speaker 3: And now that you have only engineers, there no blue helmets,
42:27
Speaker 3: only white helmets left, right, very few helms. These are
42:30
Speaker 3: the people who repair the machines. But that's it. And so.
42:35
Speaker 2: The lowering and fruit have already been used up. And
42:37
Speaker 2: now it's getting very very hard to do further mechanization
42:40
Speaker 2: and it's very naive to believe that what is now
42:43
Speaker 2: left of human labor in a car factory can easily
42:45
Speaker 2: be mechanized, right.
42:47
Speaker 3: It's very very hard.
42:48
Speaker 2: And the robotics expert they go there and look at
42:51
Speaker 2: it and say, gosh, we can't do that, right. And
42:53
Speaker 2: so therefore the idea that so if we reindustrialize America
42:57
Speaker 2: and put car factories back into America means that we
43:01
Speaker 2: use traditional robots in these factories, but not that we
43:04
Speaker 2: use robots that do the fine motorical work.
43:07
Speaker 1: But also it's worth noting that traditional robots are usually
43:11
Speaker 1: welded to the benches, and that's because if you have
43:14
Speaker 1: robots walking around, they're too liable to cause accidents, including
43:19
Speaker 1: accidents of killing human beings by crashing into them, because
43:22
Speaker 1: they just don't have the intelligent reactions that human beings
43:27
Speaker 1: have when they're engaging with each other.
43:29
Speaker 4: And of course, of course yes, sorry, go.
43:33
Speaker 2: On, no, no, I just they look silly, they look
43:38
Speaker 2: so I remember ten fifteen years ago there was this
43:40
Speaker 2: huge program. It was called Robotics three point or four
43:43
Speaker 2: point or I don't know where more rich digit they use,
43:45
Speaker 2: but they said, now the robot is coming out of
43:46
Speaker 2: the cage.
43:47
Speaker 3: Finally, because the robots are all in cages, and the
43:50
Speaker 3: cages are.
43:51
Speaker 2: The humans who still either factory from the from the
43:55
Speaker 2: movements of the robotic arms that have a ute momentum.
43:57
Speaker 2: If you get hit on the head but such an arm,
43:59
Speaker 2: you're dead for sure, right or you had this being
44:01
Speaker 2: cut off. And so therefore they are all operating in cages.
44:04
Speaker 2: And they said, now we set the Robert free off
44:06
Speaker 2: the cage, but it never happened. And if you talk
44:09
Speaker 2: to re robotics experts, they can point out in a
44:11
Speaker 2: very detailed way why that is the case. And this
44:14
Speaker 2: problem of sensory motor behavior is unsolved, and I think
44:17
Speaker 2: it's unsolvable.
44:19
Speaker 4: You mentioned chess a few times, and I love chess,
44:22
Speaker 4: and something that I thought about and I didn't realize
44:24
Speaker 4: this until I read your book, but Chess has a
44:29
Speaker 4: very large number of, you know, opening moves, and I
44:33
Speaker 4: guess that's part of the magic of playing the game.
44:37
Speaker 4: It seems almost infinite, but it is, as you say,
44:40
Speaker 4: a closed system, there will be a point at which
44:43
Speaker 4: there are no more new opening moves. Whereas if I
44:47
Speaker 4: go for dinner with my wife, that is an open system,
44:50
Speaker 4: because anything can happen from leaving the house to coming
44:53
Speaker 4: back home.
44:55
Speaker 3: Yeah, like guessterday reserve called pizza. You know you didn't
44:58
Speaker 3: expect it.
45:02
Speaker 1: I think the most embarrassing feature of the developments in
45:06
Speaker 1: AI in the last years is the failure still to
45:10
Speaker 1: create coherent, friendly chatbots who can respond to telephone calls.
45:15
Speaker 1: When you call your bank, you always have to adjust
45:18
Speaker 1: the level of your voice, the clarity of your requests,
45:21
Speaker 1: and so forth in order to get any kind of
45:24
Speaker 1: friendly response from these chatbots. Now that is a scandal.
45:28
Speaker 1: They've been trying to build those chatbots for seventy years
45:31
Speaker 1: and they still fail. Humans you can talk to a
45:34
Speaker 1: stranger on the railway station immediately and have friendly responses
45:39
Speaker 1: in both directions. They can't do that.
45:41
Speaker 2: So the reason why this is the case, why they're
45:44
Speaker 2: not solving the why they're not winning the touring tests,
45:48
Speaker 2: is because winning the touring test would mean that you
45:49
Speaker 2: could put machines in all these routine into actors like
45:52
Speaker 2: booking an airplane, booking a hotel, and be sure the
45:55
Speaker 2: machine can achieve it.
45:56
Speaker 3: That would be the retouring test.
45:57
Speaker 2: Why is this not happening because even it's such a
46:00
Speaker 2: mundane and trivial conversation like booking something, there's an infinite
46:04
Speaker 2: number of possibilities how the conversation can evolve. Now, you
46:07
Speaker 2: cannot train a finite algorithm to do with infinity right,
46:10
Speaker 2: And basically the way the conversations evolve is non argotic,
46:15
Speaker 2: so there's no clear pattern. And if there's no clear pattern,
46:17
Speaker 2: you cannot cover all the way the conversation can evolve.
46:20
Speaker 3: And that's and that's why.
46:22
Speaker 2: So the lllms, the best tools we have for chatting,
46:24
Speaker 2: are trained only for one interaction, one question, one answer.
46:28
Speaker 2: But if you but how did people make it? Praise
46:30
Speaker 2: Hitler by actually talking for a very long time to
46:34
Speaker 2: the model, and this way exploring aspects of its configuration
46:40
Speaker 2: that cannot be controlled in the training, in the in
46:42
Speaker 2: the second and third step of the training, because they
46:45
Speaker 2: are only for one question, one answer, because you cannot
46:48
Speaker 2: train more steps because there are too many possibilities. And
46:52
Speaker 2: because of this, this is another example for the for
46:55
Speaker 2: that that the models cannot saturate and that their configuration
46:59
Speaker 2: is basically.
47:00
Speaker 3: Only controlled at the superficial level.
47:02
Speaker 2: Therefore they can, they can, you can. Therefore, the big
47:06
Speaker 2: vendors have also pressed long conversations. Now because there were
47:10
Speaker 2: too many quote unpot scandals arising. So the L and
47:13
Speaker 2: M was asking the intact human intection and partner to
47:16
Speaker 2: commit suicide for example, right, and and this created scans.
47:20
Speaker 3: And that's due to this problem.
47:22
Speaker 4: M So am I correct in saying then, that intelligence
47:28
Speaker 4: cannot be definitively defined and as such cannot therefore be engineered.
47:34
Speaker 4: And if it cannot be engineered, the best that we'll
47:38
Speaker 4: end up with is advanced automations.
47:42
Speaker 3: Yes, so the first the first of your premise is
47:44
Speaker 3: not fully correct.
47:45
Speaker 2: So we can define intelligent behavior, but we cannot analyze
47:50
Speaker 2: how it comes about. So we know what intelligent behavior is,
47:52
Speaker 2: but we cannot analyze how it comes about, how the
47:55
Speaker 2: neurons of even of a of a mammal or a
47:58
Speaker 2: parrot create it's intelligent behavior. And the rest of your
48:02
Speaker 2: statement is ke fully correct.
48:03
Speaker 1: That involves both spontaneity and novelty. Now that the other
48:09
Speaker 1: ms are very good if you're asking questions about Taylor Swift,
48:13
Speaker 1: because it's full of information about Taylor Swift. But if
48:16
Speaker 1: you ask novel questions, you may get no answers, you
48:20
Speaker 1: may get hallucinations. You're not going to be happy with
48:25
Speaker 1: the response from a human. Whatever response you get, they're
48:29
Speaker 1: likely to be happy with it. It will be friendly,
48:32
Speaker 1: it will tell you what to do next, and so forth.
48:35
Speaker 1: So it's not a question of defining intelligence, as Yupe says,
48:38
Speaker 1: we know what intelligence is. The problem is creating a
48:42
Speaker 1: machine that can act with spontaneity and novelty. And they
48:46
Speaker 1: can't do that their machines. They're mathematical algorithm. Algorithm.
48:51
Speaker 2: Yeah, so our culture has since the card has the
48:55
Speaker 2: tendency to overestimate what we can do with mathematics. And
48:58
Speaker 2: the card himself is rather innocent of this, but the
49:01
Speaker 2: tradition that was based on him la Place believed that
49:05
Speaker 2: if we could measure all the positions and velocities or
49:10
Speaker 2: momenta of all the particles of the universe, we could
49:13
Speaker 2: predict the future but also calculate back the whole past perfectly.
49:17
Speaker 2: And that's that's wrong. That has so physicists know that's wrong.
49:20
Speaker 2: They know this in many ways, but but but very
49:26
Speaker 2: late people don't and they overestimate what can be done
49:28
Speaker 2: with mathematics. But if you ask people like Heisenbacher Feyman,
49:31
Speaker 2: they were very very modest because they were seriously trying
49:35
Speaker 2: to model agure. If you do this seriously, like I
49:38
Speaker 2: did all my life, you become very modest. But if
49:41
Speaker 2: you're just you know, an entrepreneur, engineer who is who
49:44
Speaker 2: has a lot of money to spend, you are not
49:47
Speaker 2: taking the time to think this through. And I think
49:50
Speaker 2: all the barrier I did is just think it through
49:52
Speaker 2: properly at the very high level of detail to get
49:56
Speaker 2: to the bottom of it. And that's what you need
49:57
Speaker 2: to do to answer such questions. And that's what you
50:00
Speaker 2: look lay to when Socred has an Aristotle, all this
50:02
Speaker 2: tradition we are based on have always tried to really
50:05
Speaker 2: get into the details. And then you realize it by itself,
50:08
Speaker 2: Then it's not scandalous. Then it's nothing special. It's just
50:11
Speaker 2: how things are.
50:12
Speaker 4: It's just the reality we're faced with, right Yeah, And
50:15
Speaker 4: it doesn't help to be luddite either, So we're.
50:19
Speaker 1: Not ludd It's the final chapter in the book is
50:21
Speaker 1: all about the wonderful things which AI is already doing
50:25
Speaker 1: for us, in which it will continue to do. Just
50:28
Speaker 1: as all the tools invented through the Industrial Revolution have
50:31
Speaker 1: been very helpful to creating our modern technical world, so
50:35
Speaker 1: AI will contribute more to that.
50:37
Speaker 2: So all they are huge scays with a already, and
50:41
Speaker 2: these successes have been produced by realistic AI users, right.
50:46
Speaker 2: And so the problem of open AI, for example, as
50:49
Speaker 2: a company is that they have now spented I think
50:51
Speaker 2: twenty billion or something, maybe months more, but they have
50:55
Speaker 2: not made any profit. Why not because there is no
50:58
Speaker 2: way to use this model to automate, right, And so
51:01
Speaker 2: only with automation can you make money with technology. Only
51:04
Speaker 2: when technology can can do something automated human or enhances
51:09
Speaker 2: human behavior, then the technology is good. Like like construction machines,
51:13
Speaker 2: they are good because they replace thousands of humans who
51:16
Speaker 2: had to dig with their shovels.
51:17
Speaker 3: Right now, now, the tools that open us billing cannot
51:20
Speaker 3: do any of this.
51:21
Speaker 2: They're just better entertainment tools or in propaganda tools maybe
51:25
Speaker 2: or censorship tools. But they cannot create any real value,
51:29
Speaker 2: and so they will. They are in the way about
51:31
Speaker 2: to fail economically speaking, and this is very important. Good
51:34
Speaker 2: AI like I'm making it has always an economic impact.
51:37
Speaker 3: So I think, is this a problem that I can
51:39
Speaker 3: tackle with mathematics? If yes, I build it and.
51:42
Speaker 2: They get paid for it and everybody's happy. But I'm
51:44
Speaker 2: not willing to participate in a hype that just burns
51:48
Speaker 2: money without creating any good.
51:50
Speaker 4: So we've got this little this vacuum leaner that that
51:54
Speaker 4: is automated. It drives around the house on its own
51:57
Speaker 4: and it goes back to its little spot in the corner.
52:00
Speaker 4: Now that's quite helpful.
52:02
Speaker 3: Yeah, perfect our law law lawing machine.
52:06
Speaker 2: Then agriculture, you know, automatic harvesting machines, seeding machines, plowing machines.
52:13
Speaker 2: They are wonderful applications of I working with satellite data
52:17
Speaker 2: signals from satellite. It's fantastic, right, So the whole electric
52:22
Speaker 2: grid is full of AI algorithms and control it and
52:24
Speaker 2: prevent blackout. It's everywhere and it's super beneficient. If you
52:28
Speaker 2: substract it, you get a much law productivity. And in
52:32
Speaker 2: the end less wells it can be distributed, so it's everywhere.
52:35
Speaker 2: It's a great It's part of the industrial revolution. And
52:39
Speaker 2: I'm a big fan of the industrial.
52:40
Speaker 4: Revolution with the advancement in the sort of digital control grid,
52:46
Speaker 4: as Katine Austin Fitz calls it, where you'll have a
52:49
Speaker 4: social credit score, for example, where everything that you do
52:53
Speaker 4: is tracked and traced and monitored and you're given a score.
52:57
Speaker 4: And let's say you go and put fuel into your
52:58
Speaker 4: car and it says you cannot have any more fuel,
53:01
Speaker 4: or you buy red meat and the machine says, ah,
53:03
Speaker 4: you're going to get penalized with some sort of carbon
53:05
Speaker 4: tax whatever. I'm guessing the reason why that's happening is
53:09
Speaker 4: that that is that is not AI doing its thing.
53:11
Speaker 4: That's not machines taking over those or automations, but are
53:14
Speaker 4: very highly advanced automations created by humans.
53:17
Speaker 3: Yes, and and they're all requiredgization.
53:20
Speaker 2: But I would like to say that I think this
53:23
Speaker 2: vision that Kastin often fits is describing and also others
53:27
Speaker 2: are describing, doesn't take into account what we've learned in
53:32
Speaker 2: such extreme control situations such as concentration camps or for example.
53:36
Speaker 3: Vienna after World War Two.
53:38
Speaker 2: So when you when you are in such a planned
53:40
Speaker 2: economy like a concentration camp or Vienna the under the
53:44
Speaker 2: Allied occupation, what happens is that people evade the control
53:48
Speaker 2: mechanisms and create gray and black markets.
53:50
Speaker 3: Right, so human beings. So the maximum enslaver and rate
53:54
Speaker 3: of any human society in history has been thirty percent.
53:57
Speaker 3: Even the Arabic Peninsula had only thirty send slaves at
54:01
Speaker 3: the height of the enslavement. All Greek and Latin and
54:04
Speaker 3: Roman empires. They had only thirty percent left. If you
54:07
Speaker 3: have more, the balance shifts, so you cannot enslave everyone.
54:11
Speaker 3: So I don't believe that's going to happen.
54:12
Speaker 2: But if that, if something like this has been tried,
54:15
Speaker 2: people will evade it, right like people created fault vaccination
54:18
Speaker 2: passports during COVID and and all invented all sorts of
54:22
Speaker 2: mechanisms to invade and every system of control can be
54:26
Speaker 2: circumvented if there are people inside the systems were willing
54:29
Speaker 2: to circumvent it. And so that's why any security system
54:32
Speaker 2: of any company is only as good as the loyalty
54:34
Speaker 2: of its of its employees.
54:36
Speaker 3: And that's that will happen.
54:38
Speaker 2: So if it becomes too dystopian, people will work around it,
54:41
Speaker 2: will create ways to circumvent it. And that's that's why
54:45
Speaker 2: also nineteen eighty four and A Brave New World are
54:48
Speaker 2: wrong because they don't describe the.
54:50
Speaker 3: Ability of humans to circumvent such systems.
54:52
Speaker 1: And I think what we know from China is that
54:54
Speaker 1: the rather ambitious accounts of the social credit system there
55:01
Speaker 1: are in fact massive exaggerations. It doesn't work for most
55:04
Speaker 1: Chinese population members.
55:08
Speaker 4: Yeah, that's actually a really good point. When you think
55:11
Speaker 4: you correct, people will always circumvent these things. And I
55:15
Speaker 4: guess it circles back to your original point because machines
55:20
Speaker 4: don't have intelligence and humans do.
55:24
Speaker 3: Well.
55:25
Speaker 2: The control system is also set up by humans, so
55:27
Speaker 2: then they have a you have a circle.
55:29
Speaker 3: Very good example is a Napoleon rule over France.
55:32
Speaker 2: Napoleon was the first to introduce all the things that
55:34
Speaker 2: Hitler and Stalion also used, like Stazi, total control, censorship, Glaishalton.
55:40
Speaker 2: That means, you know, making or trying to make everything
55:43
Speaker 2: the same and all the domains of all the institutions.
55:45
Speaker 3: And it was also massively circumvented.
55:48
Speaker 2: So when the when the French burgeoisi was fed up
55:50
Speaker 2: with Napoleon, they paid for the British war against Napoleon
55:54
Speaker 2: in Spain. They spied on the French military and gave
55:58
Speaker 2: all the secrets of sub secret plans to to the
56:00
Speaker 2: Brigantes Britain's the French citizens themselves, right, And so when
56:03
Speaker 2: you try to control a fanatically a society, you will
56:07
Speaker 2: always fail. And the same happened in Germany right during
56:11
Speaker 2: World War Two, So it will always fail. Now it
56:14
Speaker 2: can still create a lot of damage while they try
56:17
Speaker 2: to establish it. Therefore, it's still something to be very
56:20
Speaker 2: much afraid of and to fight against. But in the
56:23
Speaker 2: mid term it will not work because because when a
56:27
Speaker 2: few people use their intelligent to control the mass, and
56:30
Speaker 2: the masses understand this and they use their intelligence, they
56:33
Speaker 2: are superior, because they're more and because of the limits
56:36
Speaker 2: of this and the limits of the technology.
56:39
Speaker 4: And as Matthias Desmetz said, only around thirty percent of
56:43
Speaker 4: people will willingly partake into talitarianism anyway, Yes, I think.
56:49
Speaker 3: The rate is approximately right, and I think it's it's
56:52
Speaker 3: also seems to be an anthropological constant right over time,
56:56
Speaker 3: and his thirty percent are very much in line with
56:59
Speaker 3: thirty percent.
57:00
Speaker 4: By the way, Yeah, this is such an uplifting positive conversation.
57:06
Speaker 4: Please try and explain to me why therefore your book
57:09
Speaker 4: is controversial.
57:11
Speaker 3: Take a shot please.
57:12
Speaker 1: So the hype factor is important here. The AI technology
57:18
Speaker 1: today is so expensive. They need lots of people to
57:21
Speaker 1: have faith in it that they're doing something good, valuable, useful, friendly,
57:26
Speaker 1: entertaining and so forth. They need it to be a
57:28
Speaker 1: show on the road, which is getting ever more important.
57:31
Speaker 1: We point out that there's a limit that the show
57:35
Speaker 1: will do. The things it can do now, but it's
57:37
Speaker 1: not going to get better and better over time. On
57:39
Speaker 1: the countrary, it's almost certainly reached the point of most
57:44
Speaker 1: attractiveness and the point of least well. Gary Marcus, who
57:50
Speaker 1: is one of the profits on our side, in many ways,
57:54
Speaker 1: he talks about enthitification. So the Internet is becoming enthitified
58:00
Speaker 1: by the products of large language models, and so the
58:03
Speaker 1: data sources of the large language models are becoming less
58:07
Speaker 1: and less reliable over time. This is a bad cycle
58:10
Speaker 1: which will we shouldn't talk about. You probably should delete
58:15
Speaker 1: what I just said. I talk about the limits of this,
58:21
Speaker 1: because that will take away the money who's going to
58:23
Speaker 1: invest in something which is actually going to But as
58:27
Speaker 1: did every AI summer in history, So AI is a
58:31
Speaker 1: series of AI winters.
58:33
Speaker 2: You you probably have Dutch ancestors right far back, I guess.
58:38
Speaker 2: And and so you remember that you know the tulip
58:40
Speaker 2: crisis right so yes, in in Amsterdam, Yes, so one.
58:45
Speaker 2: I don't know how this this the thing, the part
58:48
Speaker 2: of the troop that's out of the earth, the bulb
58:50
Speaker 2: of the tool. One bulb of a troop was worth
58:52
Speaker 2: a house at the almost the house at the at
58:55
Speaker 2: the peak of this of this.
58:56
Speaker 3: Crisis and then it crashed. So I'm i'm so that
59:00
Speaker 3: this AI hype will crash. And the question we.
59:03
Speaker 2: Cannot You can never know when the crash happens, but
59:05
Speaker 2: it will. I guess it will crash when our financial
59:08
Speaker 2: system crashes. And the crash of our financial system is
59:12
Speaker 2: unavoidable and will take with it a lot of hypes
59:14
Speaker 2: because all of them rely on this heroine of of
59:18
Speaker 2: money made out of nothing to feed into all these sypes.
59:21
Speaker 2: And when this money is gone, then it's like the
59:24
Speaker 2: withdrawal symtom of a heroin addict. Then suddenly a lot
59:26
Speaker 2: of the realities will bite him or her again.
59:29
Speaker 1: Right.
59:29
Speaker 3: And also, I think that a lot of.
59:31
Speaker 2: The hype cycles that we see since the last fifty
59:34
Speaker 2: years are very much culturally rooted in the in the
59:38
Speaker 2: in the effects of the FIAT money system. As a
59:40
Speaker 2: high action, I was always against the FIA money system.
59:43
Speaker 2: I think it's it's very dangerous and foolish to set
59:46
Speaker 2: up such a system. It privileges the wrong people in
59:48
Speaker 2: a society and has the wrong effect. I think it's
59:51
Speaker 2: yours very maybe we cannot put it on you. It
59:53
Speaker 2: privileges the wrong people and and I think if this
59:57
Speaker 2: goes away, a lot of these hypes will will.
59:59
Speaker 4: Stop your book. Why machines will never what is it
1:00:07
Speaker 4: rule the world? Why machines never rule the world? Where
1:00:10
Speaker 4: can I get it?
1:00:12
Speaker 1: So there is a new edition which just came out
1:00:15
Speaker 1: or which is about to come out any day now,
1:00:17
Speaker 1: I think it's today, is yesterday came out? So I'm
1:00:21
Speaker 1: tempted to say that I will email you the link
1:00:23
Speaker 1: where you can get a discount, and then you can
1:00:26
Speaker 1: put that at the end of the interview. So but
1:00:29
Speaker 1: that's the best I can do from here or Amazon
1:00:32
Speaker 1: Amazon Amazon Amazon. Yet they probably actually are copying the
1:00:36
Speaker 1: discount because they generally do that. So it's appearance either
1:00:41
Speaker 1: today or tomorrow.
1:00:42
Speaker 2: I think it's already appeared, and I got the notification
1:00:45
Speaker 2: of my ex that I will have they will ship
1:00:48
Speaker 2: my ten.
1:00:50
Speaker 3: Copies or so today, so it must have been appeared,
1:00:53
Speaker 3: and you can order it, and I.
1:00:54
Speaker 2: Can recommend it even a layperson who's not wanting to
1:00:57
Speaker 2: go into the depth of mathematics and physics and philosophy.
1:01:00
Speaker 2: Ken Read's introduction and the final chapter and the introduction
1:01:03
Speaker 2: is very clearly readable, and the forward is very important
1:01:06
Speaker 2: as well. And then the final chapter gives a positive
1:01:08
Speaker 2: perspective on R.
1:01:09
Speaker 3: And then that's all. I think all you need is
1:01:11
Speaker 3: a layperson it's the last and.
1:01:12
Speaker 4: The first chapter, or just use AI to summarize it.
1:01:20
Speaker 4: So that's the singularity. Y S Langber Barry Smith, thank
1:01:27
Speaker 4: you for joining me in the Strangers.
1:01:30
Speaker 3: Yeah, thanks a lot for inviting us. Bye bye m