This is the English transcript of an interview between David Deutsch (@DavidDeutschOxf) and Dennis Hackethal (@dchackethal) about the German translation of David's book "The Beginning of Infinity' which Dennis translated. The interview, originally in German, can be found here: https://www.youtube.com/watch?v=UHm88w_BnMU is 35 minutes but all in German. As a service to the community, Jannik Wiese (@jannikwiese) alongside Dennis, worked to translate the entire interview here for those unable to speak German.
Enjoy.
Enjoy.
In parentheses: (clarifying additions by the translators).
In brackets: [original German words used in the interview]
[00:10] Dennis Hackethal: Welcome to our listeners and viewers. Several years ago I read a fascinating book and as soon as I finished reading, I knew it was my new favourite. I’m talking about ‘The Beginning of Infinity’ by British physicist David Deutsch. It is now available in German for the first time, titled ‘Der Anfang der Unendlichkeit: Erklärungen, die die Welt verwandeln’ (the original title is ‘The Beginning of Infinity: Explanations that Transform the World’).
My name is Dennis Hackethal, I am the translator and today I speak with the author about his book. Hello David Deutsch!
[00:41] David Deutsch: Hello!
[00:42] Dennis: What is your book about? What is the overarching theme?
[00:47] David: You could say that the theme is optimism - or, perhaps better: progress. Progress is the first word in the book. But there are other topics, for example knowledge [Wissen] - as you say - or (alternatively) insight [Erkenntnis]. But we settled on knowledge [Wissen]. Plus there are further topics.
[01:22] Dennis: You write that the core idea of the Enlightenment is that progress is not only possible but desirable. And many people today say ‘yes, well, the progress we’ve made so far is nice but we shouldn't expect that this can just go on indefinitely or that it even should go on - that that would be a good thing.’
For example, many hope for a return to a ‘life in harmony with nature’, as they call it... as these people call it. And in general some people seem to have a cynical attitude regarding progress.
You give the example of people who say that antibiotics are great because they allow us to cure illnesses that we couldn’t cure before but they also lead to more and more antibiotic-resistant bacteria; accordingly, progress, as you write, gets reframed as ‘so-called progress’. How do you counter such objections?
[02:18] David: We know that all progress comes with problems. The endpoint of solving problems is not an unproblematic state but rather better problems. Hence we should expect that new problems will appear. With antibiotics it is just like that. We shouldn’t expect that the problem is solved conclusively. Rather we should always expect new problems to appear but that they are soluble. That is not to say that they have to get solved. We can always be mistaken and we can also always reach wrong solutions that go in the wrong direction but that is also a problem which is also solvable.
[03:44] Dennis: That’s one of the core theses in your book – that problems are inevitable but also soluble. Why are problems inevitable?
[03:57] David: Because perfect theories don’t exist. We are fallible creatures. That has to be the case. It’s not just humans, but all things that could solve problems are fallible. That is to say there is never a guarantee that we will manage to solve a particular problem before it kills us. But it is possible. And we know more than that. We also know what is required to overcome them - by solving problems.
[05:00] Dennis: Progress seems to depend on an understanding of the world. However, there are surely things that we can’t comprehend, say because evolution didn’t equip our brains with the capabilities that it would need to understand everything. If that’s the case then sooner or later we will hit some kind of barrier to progress, no?
[05:24] David: No, because optimism does not depend on properties of the brain but rather on the universality of the laws of physics. It is not a purely human thing, it is a universal thing - optimism as well as knowledge. That is to say there are two types of things in the universe, those that can create knowledge and those that can’t. There are no additional things, for example creatures that are much smarter than us the same way we are smarter than monkeys.
We can invent things like computers that speed everything up, that can store more information. However, that doesn’t make us smarter and there is nothing smarter than us because whatever these smarter creatures would have, we could have as well, using computers and so on.
[07:01] Dennis: You just pointed to a core difference between people, as you call them, and everything else. There exist humans but humans are just a subset of the set of all people. And if I understand you correctly, that means, for example, that aliens that are intelligent like us, if they exist, are people as well. But ants or monkeys are not people, did I get that right?
[07:28] David: Yes. Ants, monkeys, rocks, planets… those are not people. We and the aliens are people and equal in a deep sense.
[07:51] Dennis: That means we can imagine an alien civilisation that doesn’t necessarily resemble ours and yet the members of such a civilisation are intelligent like us. But we could also picture other planets where ‘dumb’ life exists, like ants for example or other insects. What would differentiate the alien ants from the intelligent alien creatures?
[08:19] David: That they (ants) can’t create knowledge. That is to say their species can create knowledge through evolution, but ants themselves cannot, and that is a big difference. Not just because we people can do it much faster but also because with ants many millions have to die to create the tiniest amount of knowledge. With us (people) nobody has to die, only our ideas, as Popper says.
[09:07] Dennis: Exactly. That’s the Austrian-British philosopher Karl Popper whom you often reference in the book and who had a strong influence on you.
[09:13] David: Yes indeed.
[09:15] Dennis: He says that we can let our theories die in our place. I want to return to the comprehensibility of the world and how that relates to the question of whether we can understand the world completely. You mentioned earlier that this might not be the case. I want to cite the physicist Harald Lesch who studies the same question. He also reaches the conclusion that we can’t understand the world once and for all, like you. (Lesch:) ‘The world is open.’ That sounds a lot like Popper by the way. He (Lesch) refers to the mathematician Kurt Gödel. Now, you can explain to me how that relates but Kurt Gödel has shown that there will always be, say, statements in arithmetic that are neither provable nor disprovable. Lesch refers to Gödel’s incompleteness theorem and says that it could also be called Gödel’s hope theorem because Lesch sees in it proof of the world’s openness and our potential for progress. On the other hand one could say that it’s cause for despair because it means there are things we can never know or prove. Isn’t that in itself an insurmountable barrier to progress?
[10:25] David: It isn’t. But first I have to say that the fact that mathematics is unbounded is not the same as the fact that physics or other forms of knowledge are.
I would say that physics is open because it consists of explanations and explanations always explain the seen by way of the unseen.
When we understand something we have explained it and the explanation speaks of unexplained things. That means that explanations can never be complete, and that's very hopeful in physics as well as mathematics because it means that we can always improve our understanding. It also means that our lives - life generally - can always improve because that depends on knowledge.
[12:02] Dennis: We explain the seen with the help of the unseen like for example explanations. Explanations can’t be seen.
[12:10] David: Yes, no but that’s not exactly what I’ve said. Explanations themselves explain the seen by way of the unseen and therefore even the best explanations we could ever have need unseen things that we have not explained yet.
[12:36] Dennis: Okay, that means it always continues.
[12:38] David: Yes and that’s very hopeful because if you consider how things would be if that were wrong then progress would come to an end. And not a good end because problems would continue to exist.
[12:57] Dennis: Problems would continue to be inevitable, and yet we would encounter a barrier to progress. So eventually one of these problems would obliterate us and we couldn’t do anything about it.
[13:11] David: Exactly.
[13:12] Dennis: Luckily that is not the case.
[13:14] David: Luckily not.
[13:16] Dennis: You start off the first chapter by talking about epistemology [Erkenntnistheorie], or, as you said earlier, we prefer the term knowledge [Wissen], so ‘philosophy of knowledge’ [Wissensphilosophie]. Over the centuries there have been many different answers to the question of what demarcates science from other fields. Perhaps the most important answer so far - at least up until the publication of your book - was Karl Popper’s, namely that what differentiates science from non-science or ‘metaphysics’ is testability. But is testability truly sufficient?
[14:00] David: I don’t think it’s as important as Popper thought. More important than testability is criticizability. Once stated this way the criterion of being criticizable applies not just to science but to everything.
[14:32] Dennis: In the first chapter you introduce a new criterion that is superior to testability for the reason you just described. You call it the criterion of good explanations. What characterises a good explanation and what differentiates a good explanation from a bad one?
[14:50] David: A good explanation is an explanation where every part of it plays a role in the explanation. That is to say, if you changed any part of the explanation it would no longer explain what it is supposed to explain.
[15:13] Dennis: That means every part of the explanation has to be constituted in such a way that it can’t be easily varied without making the explanation fall apart.
[15:22] David: Exactly.
[15:23] Dennis: Bad explanations are different. In the book you mention old myths for example, which aim to explain seasons by referring to certain gods and their feelings, in which case you could swap out a specific god for another one without impacting the quality of the explanation one way or another.
[15:42] David: Yes or (you could swap out) different feelings and so on.
[15:45] Dennis: The sciences are characterised by the search for good explanations and by the rejection of bad explanations in favour of good ones. That does not just apply, as you write, to the sciences, but to every field in which progress is to be made. The only way to make progress in any field is by striving for good explanations. That is the case for philosophy, for example, as well as aesthetics and so on.
[16:08] David: Yes, (it applies to) everything.
[16:10] Dennis: Now we can ask, okay, we’re searching for good explanations. That means once you find a good explanation... you find it out there by making observations somehow? How do you do that? Where does this knowledge come from?
[16:25] David: That doesn’t mean the explanation has to be true. We can always be mistaken even when we have very good explanations. But the process by which we find good explanations is a process of correcting or avoiding errors. That means even when the explanation is wrong we still have corrected errors which we previously thought to be true.
[17:02] Dennis: Back in the day - and many still believe this today - there was this conception of knowledge creation [Wissensschöpfung], and with this term I’m already hinting at something important, which claimed that knowledge creation worked differently from the way we know it has to work today.
It was believed - that’s empiricism - that knowledge can be derived from sense data. Basically you just open your eyes, take a look around and then ‘read’ from the ‘book of nature’. That’s an old empiricist metaphor. Today that is still common sense [in German: healthy sense] - maybe not so healthy - but still a widespread assumption.
[17:03] David: Not so healthy, since Popper already refuted it 80 years ago. Why (Popper’s refutation) isn’t common sense today I don’t know. It’s sometimes as if poor Popper had never lived. I think maybe at this point more people and even philosophers take him seriously. I hope that is the case or maybe that’s just the people who talk to me.
[18:26] Dennis: Yes, I hope so too. Popper solved another problem, the so-called problem of induction. That’s about inductivism. Inductivism is - I quote from your book - ‘The misconception that scientific theories are obtained by generalizing or extrapolating repeated experiences, and that the more often a theory is confirmed by observation the more likely it becomes’. And this was actually already known to be false centuries before through the work of David Hume.
[18:59] David: That it can’t be logically valid. That it could nevertheless be the case that we still do it like that (inductively) was believed by many, even to this day.
[19:14] Dennis: Why can’t inductivism be right? There is this old example: I see 10 white swans, therefore I should expect the eleventh swan also to be white or maybe even every swan to be white. What’s wrong with that? That sounds pretty reasonable, no?
[19:31] David: As Popper has shown, that isn’t the only reason why that can’t be right. I think a deeper reason is... how do we know beforehand when the same thing reoccurs? For example, when you see swans, how do you know that you should count swans and not rivers or swans with 5cm long beaks but not 4cm long beaks? If you know that already then you already know something about swans and that something is exactly what you believe you… induced (?).
[20:42] Dennis: ...from which one inductively inferred (which one induced)...yes
[20:46] David: ...inductively inferred.. yes..
[20:48] Dennis: So that means we already have a theory that it’s swans we are supposed to observe and count. That means we have a criterion determining what we are currently interested in, and this criterion itself cannot be derived from repeated observations.
[21:02] David: Yes.
[21:03] Dennis: That means inductivism doesn’t even tackle the question of where knowledge comes from.
[21:06] David: Yes, and these criteria can also be wrong and often are wrong.
[21:14] Dennis: And, as it turns out, not all swans are white. Instead there are black swans also.
[21:19] David: That’s exactly why (inductivism is wrong)
[21:22] Dennis: And... indeed if inductivism were right, it would always be the case - well it rules itself out this way - it would always be the case that you would be the most confident that your theory is correct the moment before it’s refuted. There is this old example of farm animals which are happy that they get lots of food every day. The day before their slaughter they are the most confident that they will get fed again the next day.
[21:55] David: The philosophers who induce could say that you could always be mistaken, just like I and Popper say that we could always be mistaken. The fact that one could be mistaken is no proof that one has actually done anything wrong. However, with inductivism it’s different. It’s not just that we can be mistaken. It’s that our ideas about swans and so on couldn’t stem from observations because we could induce the new theory from the very same observations. That and many other reasons make inductivism untenable.
[23:05] Dennis: You contrast this with something you mentioned just now, namely error correction as well as Popper’s philosophy of knowledge which, as you lay out, is about error-correction by guessing. Here new theories are not derived from observations, instead they are guessed. To learn more about that I would invite our listeners to read the book wherein you cover the topic in more detail in chapter one as well as in a wonderful fictional dialogue between Socrates and Hermes.
[23:33] David: You liked it, nice!
[23:35] Dennis: It’s one of my favourite chapters, definitely. The question about knowledge creation, how it works and how it can’t (possibly) work is deeply related to artificial intelligence. The original goal of AI researchers - which they unfortunately lost track of, it seems to me - once was to understand the human mind and to write it down as a program to create and run it on a computer.
[24:03] David: The first one who wanted that in modern times was Turing himself.
[24:08] Dennis: Yep. Alan Turing the great computing pioneer. In 1950 he wrote a paper called ‘Computing Machinery and Intelligence’ - in German ‘Rechenmaschinen und Intelligenz’ - wherein he asks this question: ‘Can machines think?’. And he says: yes. His predecessors disagreed, Ada Lovelace for example, but he says ‘yes they can think’, and he says that in the year 2000 it will be commonplace. However, as you write, the year 2000 has come and gone by now - indeed 10 more years have passed since you first wrote the book - and we still don’t have machines that think. Now, one could ask oneself why that is the case. Could it be that Turing was simply wrong, that machines cannot think in principle?
[24:53] David: I think that can’t be right because we and machines are made of atoms and the laws of nature dictate what we can do and what we can think. We and the computers are exactly the same in this respect. So the question why no progress has been made can’t be answered by referring to brains or laws of nature. The mistake is philosophical. AI has achieved enormous progress, but AGI has made none. Why? Because almost every researcher who attempted this thought that they were supposed to build an induction machine, and that doesn’t just lead in the wrong direction but towards a dead end. It leads away from the goal. You can tell by the fact that AI has made enormous progress while AGI has made none. That couldn’t be the case if the two goals were related.
[26:51] Dennis: Just to delineate that clearly… AI - that’s how the terminology has developed in the past few decades - AI is about things like computer programs that can play chess particularly well, or text-recognition programs, stuff like that. But AGI – artificial general intelligence – is about something different entirely. Naturally humans can play chess too and recognize text, but humans have an underlying ability - creativity as you say - that these programs lack. So that’s what AGI researchers would have to investigate if they want to achieve progress.
[27:39] David: Yes. And when humans play chess for example they don’t play in the same way as computers. But they play almost as well as computers. How can that be? We do something totally different. The brain is much slower, not as exact and so on and yet we play almost as well. Why? Again, it has to be the case that the brain does something that the best AI computers don't do at all. It’s not that they don’t do it as well, they don’t do it at all.
[28:39] Dennis: What AI research does at the moment reminds me of remarks from Karl Popper’s autobiography. Now, he writes about psychology, but I think there’s the same underlying mistake. He says what psychologists are looking for is a theory of successful thinking. And it seems to me that AI researchers also want something like that because they want, say, chess programs that play successfully. They want text-recognition programs that recognize text successfully. But when it comes to real creative thinking, there is no guarantee for success.
[29:09] David: Yes. If the computer really were creative it could say: ‘This doesn’t interest me’ or ‘This doesn’t interest me anymore’ or ‘I’ll invent a new game’. Today's computers can’t do this. It’s not that they do it rarely. They never do it and are never able to.
[29:36] Dennis: Okay, while we’re on the topic of artificial intelligence. There are many who are very concerned about so-called ‘superintelligence’. This is about programs which - they haven’t been built yet but many are concerned that they will be built soon; apparently they’re right around the corner - and as soon as they get built they improve themselves so quickly and become more and more intelligent at a speed we could never catch up to. If this so-called superintelligence should decide to not like us, if it were malicious, then... that’s it for us. Then we are doomed. You discussed this with the philosopher Sam Harris, who is one of the better-known proponents of this theory. He believes this as well, that superintelligence would basically be fatal. What do you think?
[30:30] David: You can only believe that if you don’t understand what general intelligence is. It’s not about the brain, it’s about programs. Every universal computer can execute the same programs as every other (universal computer). Now, maybe we wouldn’t have enough memory for that or maybe our brain isn’t fast enough. But that’s all. There can't be any other difference between us and these superintelligent computers. But we have computers too, and could use them to think better. We do that already! We’ve been doing this since the invention of paper and pencil and computers, like nowadays. That will continue. And progress in intelligence or general intelligence that will come about in this way will happen to us just as it will to computers. So if you believe that this development will bring about humanity’s demise, not only do you have to believe that the computers will be faster and have more memory but also that the first superintelligent programs will be able to build machines that are faster and have more memory without us knowing, and for long enough so that once we finally notice, it’s too late because they have it (next-level computers) and we don’t.
However, if they are capable of that, they won’t need superintelligence. They could simply produce poison gas. And once you understand that this is the crux you have to ask yourself: Why couldn't humans do the same? And the answer is: They could! And that’s an important problem. And the same problem for computers is less serious because we will program those computers. The first ones – maybe not the second ones but the first ones. With humans it’s the opposite. There are already people who want to poison us... but no computers, including the first ones.
[34:21] Dennis: So there is an underlying universality to the human mind, which any artificial intelligence would also have. And because such an AI is already universal it’s a mistake to believe that there could be something beyond that point.
[34:37] David: Exactly.
[34:37]Dennis: This so-called superintelligence would also be an AI just like us - well, we are no AI but an I…
[34:43] David: AGI, yes.
[34:44] Dennis: Right. So there is no qualitative difference. Of course, just as there are more and less intelligent humans, there would be more and less intelligent AGIs, but there couldn’t be superintelligent AGIs on a level that we could never reach.
[35:00] David: Yes, and if they are (more) intelligent because they have a better brain we could have the same.
[35:13] Dennis: Okay, let’s take a break here. I again want to recommend to our listeners the book ‘Der Anfang der Unendlichkeit’. It is now available for purchase and there will be a second interview in which we will continue this conversation in English.
[35:30] David: Great. Good-bye!
Buy the German translation here: http://anfa.ng/kaufen
In brackets: [original German words used in the interview]
[00:10] Dennis Hackethal: Welcome to our listeners and viewers. Several years ago I read a fascinating book and as soon as I finished reading, I knew it was my new favourite. I’m talking about ‘The Beginning of Infinity’ by British physicist David Deutsch. It is now available in German for the first time, titled ‘Der Anfang der Unendlichkeit: Erklärungen, die die Welt verwandeln’ (the original title is ‘The Beginning of Infinity: Explanations that Transform the World’).
My name is Dennis Hackethal, I am the translator and today I speak with the author about his book. Hello David Deutsch!
[00:41] David Deutsch: Hello!
[00:42] Dennis: What is your book about? What is the overarching theme?
[00:47] David: You could say that the theme is optimism - or, perhaps better: progress. Progress is the first word in the book. But there are other topics, for example knowledge [Wissen] - as you say - or (alternatively) insight [Erkenntnis]. But we settled on knowledge [Wissen]. Plus there are further topics.
[01:22] Dennis: You write that the core idea of the Enlightenment is that progress is not only possible but desirable. And many people today say ‘yes, well, the progress we’ve made so far is nice but we shouldn't expect that this can just go on indefinitely or that it even should go on - that that would be a good thing.’
For example, many hope for a return to a ‘life in harmony with nature’, as they call it... as these people call it. And in general some people seem to have a cynical attitude regarding progress.
You give the example of people who say that antibiotics are great because they allow us to cure illnesses that we couldn’t cure before but they also lead to more and more antibiotic-resistant bacteria; accordingly, progress, as you write, gets reframed as ‘so-called progress’. How do you counter such objections?
[02:18] David: We know that all progress comes with problems. The endpoint of solving problems is not an unproblematic state but rather better problems. Hence we should expect that new problems will appear. With antibiotics it is just like that. We shouldn’t expect that the problem is solved conclusively. Rather we should always expect new problems to appear but that they are soluble. That is not to say that they have to get solved. We can always be mistaken and we can also always reach wrong solutions that go in the wrong direction but that is also a problem which is also solvable.
[03:44] Dennis: That’s one of the core theses in your book – that problems are inevitable but also soluble. Why are problems inevitable?
[03:57] David: Because perfect theories don’t exist. We are fallible creatures. That has to be the case. It’s not just humans, but all things that could solve problems are fallible. That is to say there is never a guarantee that we will manage to solve a particular problem before it kills us. But it is possible. And we know more than that. We also know what is required to overcome them - by solving problems.
[05:00] Dennis: Progress seems to depend on an understanding of the world. However, there are surely things that we can’t comprehend, say because evolution didn’t equip our brains with the capabilities that it would need to understand everything. If that’s the case then sooner or later we will hit some kind of barrier to progress, no?
[05:24] David: No, because optimism does not depend on properties of the brain but rather on the universality of the laws of physics. It is not a purely human thing, it is a universal thing - optimism as well as knowledge. That is to say there are two types of things in the universe, those that can create knowledge and those that can’t. There are no additional things, for example creatures that are much smarter than us the same way we are smarter than monkeys.
We can invent things like computers that speed everything up, that can store more information. However, that doesn’t make us smarter and there is nothing smarter than us because whatever these smarter creatures would have, we could have as well, using computers and so on.
[07:01] Dennis: You just pointed to a core difference between people, as you call them, and everything else. There exist humans but humans are just a subset of the set of all people. And if I understand you correctly, that means, for example, that aliens that are intelligent like us, if they exist, are people as well. But ants or monkeys are not people, did I get that right?
[07:28] David: Yes. Ants, monkeys, rocks, planets… those are not people. We and the aliens are people and equal in a deep sense.
[07:51] Dennis: That means we can imagine an alien civilisation that doesn’t necessarily resemble ours and yet the members of such a civilisation are intelligent like us. But we could also picture other planets where ‘dumb’ life exists, like ants for example or other insects. What would differentiate the alien ants from the intelligent alien creatures?
[08:19] David: That they (ants) can’t create knowledge. That is to say their species can create knowledge through evolution, but ants themselves cannot, and that is a big difference. Not just because we people can do it much faster but also because with ants many millions have to die to create the tiniest amount of knowledge. With us (people) nobody has to die, only our ideas, as Popper says.
[09:07] Dennis: Exactly. That’s the Austrian-British philosopher Karl Popper whom you often reference in the book and who had a strong influence on you.
[09:13] David: Yes indeed.
[09:15] Dennis: He says that we can let our theories die in our place. I want to return to the comprehensibility of the world and how that relates to the question of whether we can understand the world completely. You mentioned earlier that this might not be the case. I want to cite the physicist Harald Lesch who studies the same question. He also reaches the conclusion that we can’t understand the world once and for all, like you. (Lesch:) ‘The world is open.’ That sounds a lot like Popper by the way. He (Lesch) refers to the mathematician Kurt Gödel. Now, you can explain to me how that relates but Kurt Gödel has shown that there will always be, say, statements in arithmetic that are neither provable nor disprovable. Lesch refers to Gödel’s incompleteness theorem and says that it could also be called Gödel’s hope theorem because Lesch sees in it proof of the world’s openness and our potential for progress. On the other hand one could say that it’s cause for despair because it means there are things we can never know or prove. Isn’t that in itself an insurmountable barrier to progress?
[10:25] David: It isn’t. But first I have to say that the fact that mathematics is unbounded is not the same as the fact that physics or other forms of knowledge are.
I would say that physics is open because it consists of explanations and explanations always explain the seen by way of the unseen.
When we understand something we have explained it and the explanation speaks of unexplained things. That means that explanations can never be complete, and that's very hopeful in physics as well as mathematics because it means that we can always improve our understanding. It also means that our lives - life generally - can always improve because that depends on knowledge.
[12:02] Dennis: We explain the seen with the help of the unseen like for example explanations. Explanations can’t be seen.
[12:10] David: Yes, no but that’s not exactly what I’ve said. Explanations themselves explain the seen by way of the unseen and therefore even the best explanations we could ever have need unseen things that we have not explained yet.
[12:36] Dennis: Okay, that means it always continues.
[12:38] David: Yes and that’s very hopeful because if you consider how things would be if that were wrong then progress would come to an end. And not a good end because problems would continue to exist.
[12:57] Dennis: Problems would continue to be inevitable, and yet we would encounter a barrier to progress. So eventually one of these problems would obliterate us and we couldn’t do anything about it.
[13:11] David: Exactly.
[13:12] Dennis: Luckily that is not the case.
[13:14] David: Luckily not.
[13:16] Dennis: You start off the first chapter by talking about epistemology [Erkenntnistheorie], or, as you said earlier, we prefer the term knowledge [Wissen], so ‘philosophy of knowledge’ [Wissensphilosophie]. Over the centuries there have been many different answers to the question of what demarcates science from other fields. Perhaps the most important answer so far - at least up until the publication of your book - was Karl Popper’s, namely that what differentiates science from non-science or ‘metaphysics’ is testability. But is testability truly sufficient?
[14:00] David: I don’t think it’s as important as Popper thought. More important than testability is criticizability. Once stated this way the criterion of being criticizable applies not just to science but to everything.
[14:32] Dennis: In the first chapter you introduce a new criterion that is superior to testability for the reason you just described. You call it the criterion of good explanations. What characterises a good explanation and what differentiates a good explanation from a bad one?
[14:50] David: A good explanation is an explanation where every part of it plays a role in the explanation. That is to say, if you changed any part of the explanation it would no longer explain what it is supposed to explain.
[15:13] Dennis: That means every part of the explanation has to be constituted in such a way that it can’t be easily varied without making the explanation fall apart.
[15:22] David: Exactly.
[15:23] Dennis: Bad explanations are different. In the book you mention old myths for example, which aim to explain seasons by referring to certain gods and their feelings, in which case you could swap out a specific god for another one without impacting the quality of the explanation one way or another.
[15:42] David: Yes or (you could swap out) different feelings and so on.
[15:45] Dennis: The sciences are characterised by the search for good explanations and by the rejection of bad explanations in favour of good ones. That does not just apply, as you write, to the sciences, but to every field in which progress is to be made. The only way to make progress in any field is by striving for good explanations. That is the case for philosophy, for example, as well as aesthetics and so on.
[16:08] David: Yes, (it applies to) everything.
[16:10] Dennis: Now we can ask, okay, we’re searching for good explanations. That means once you find a good explanation... you find it out there by making observations somehow? How do you do that? Where does this knowledge come from?
[16:25] David: That doesn’t mean the explanation has to be true. We can always be mistaken even when we have very good explanations. But the process by which we find good explanations is a process of correcting or avoiding errors. That means even when the explanation is wrong we still have corrected errors which we previously thought to be true.
[17:02] Dennis: Back in the day - and many still believe this today - there was this conception of knowledge creation [Wissensschöpfung], and with this term I’m already hinting at something important, which claimed that knowledge creation worked differently from the way we know it has to work today.
It was believed - that’s empiricism - that knowledge can be derived from sense data. Basically you just open your eyes, take a look around and then ‘read’ from the ‘book of nature’. That’s an old empiricist metaphor. Today that is still common sense [in German: healthy sense] - maybe not so healthy - but still a widespread assumption.
[17:03] David: Not so healthy, since Popper already refuted it 80 years ago. Why (Popper’s refutation) isn’t common sense today I don’t know. It’s sometimes as if poor Popper had never lived. I think maybe at this point more people and even philosophers take him seriously. I hope that is the case or maybe that’s just the people who talk to me.
[18:26] Dennis: Yes, I hope so too. Popper solved another problem, the so-called problem of induction. That’s about inductivism. Inductivism is - I quote from your book - ‘The misconception that scientific theories are obtained by generalizing or extrapolating repeated experiences, and that the more often a theory is confirmed by observation the more likely it becomes’. And this was actually already known to be false centuries before through the work of David Hume.
[18:59] David: That it can’t be logically valid. That it could nevertheless be the case that we still do it like that (inductively) was believed by many, even to this day.
[19:14] Dennis: Why can’t inductivism be right? There is this old example: I see 10 white swans, therefore I should expect the eleventh swan also to be white or maybe even every swan to be white. What’s wrong with that? That sounds pretty reasonable, no?
[19:31] David: As Popper has shown, that isn’t the only reason why that can’t be right. I think a deeper reason is... how do we know beforehand when the same thing reoccurs? For example, when you see swans, how do you know that you should count swans and not rivers or swans with 5cm long beaks but not 4cm long beaks? If you know that already then you already know something about swans and that something is exactly what you believe you… induced (?).
[20:42] Dennis: ...from which one inductively inferred (which one induced)...yes
[20:46] David: ...inductively inferred.. yes..
[20:48] Dennis: So that means we already have a theory that it’s swans we are supposed to observe and count. That means we have a criterion determining what we are currently interested in, and this criterion itself cannot be derived from repeated observations.
[21:02] David: Yes.
[21:03] Dennis: That means inductivism doesn’t even tackle the question of where knowledge comes from.
[21:06] David: Yes, and these criteria can also be wrong and often are wrong.
[21:14] Dennis: And, as it turns out, not all swans are white. Instead there are black swans also.
[21:19] David: That’s exactly why (inductivism is wrong)
[21:22] Dennis: And... indeed if inductivism were right, it would always be the case - well it rules itself out this way - it would always be the case that you would be the most confident that your theory is correct the moment before it’s refuted. There is this old example of farm animals which are happy that they get lots of food every day. The day before their slaughter they are the most confident that they will get fed again the next day.
[21:55] David: The philosophers who induce could say that you could always be mistaken, just like I and Popper say that we could always be mistaken. The fact that one could be mistaken is no proof that one has actually done anything wrong. However, with inductivism it’s different. It’s not just that we can be mistaken. It’s that our ideas about swans and so on couldn’t stem from observations because we could induce the new theory from the very same observations. That and many other reasons make inductivism untenable.
[23:05] Dennis: You contrast this with something you mentioned just now, namely error correction as well as Popper’s philosophy of knowledge which, as you lay out, is about error-correction by guessing. Here new theories are not derived from observations, instead they are guessed. To learn more about that I would invite our listeners to read the book wherein you cover the topic in more detail in chapter one as well as in a wonderful fictional dialogue between Socrates and Hermes.
[23:33] David: You liked it, nice!
[23:35] Dennis: It’s one of my favourite chapters, definitely. The question about knowledge creation, how it works and how it can’t (possibly) work is deeply related to artificial intelligence. The original goal of AI researchers - which they unfortunately lost track of, it seems to me - once was to understand the human mind and to write it down as a program to create and run it on a computer.
[24:03] David: The first one who wanted that in modern times was Turing himself.
[24:08] Dennis: Yep. Alan Turing the great computing pioneer. In 1950 he wrote a paper called ‘Computing Machinery and Intelligence’ - in German ‘Rechenmaschinen und Intelligenz’ - wherein he asks this question: ‘Can machines think?’. And he says: yes. His predecessors disagreed, Ada Lovelace for example, but he says ‘yes they can think’, and he says that in the year 2000 it will be commonplace. However, as you write, the year 2000 has come and gone by now - indeed 10 more years have passed since you first wrote the book - and we still don’t have machines that think. Now, one could ask oneself why that is the case. Could it be that Turing was simply wrong, that machines cannot think in principle?
[24:53] David: I think that can’t be right because we and machines are made of atoms and the laws of nature dictate what we can do and what we can think. We and the computers are exactly the same in this respect. So the question why no progress has been made can’t be answered by referring to brains or laws of nature. The mistake is philosophical. AI has achieved enormous progress, but AGI has made none. Why? Because almost every researcher who attempted this thought that they were supposed to build an induction machine, and that doesn’t just lead in the wrong direction but towards a dead end. It leads away from the goal. You can tell by the fact that AI has made enormous progress while AGI has made none. That couldn’t be the case if the two goals were related.
[26:51] Dennis: Just to delineate that clearly… AI - that’s how the terminology has developed in the past few decades - AI is about things like computer programs that can play chess particularly well, or text-recognition programs, stuff like that. But AGI – artificial general intelligence – is about something different entirely. Naturally humans can play chess too and recognize text, but humans have an underlying ability - creativity as you say - that these programs lack. So that’s what AGI researchers would have to investigate if they want to achieve progress.
[27:39] David: Yes. And when humans play chess for example they don’t play in the same way as computers. But they play almost as well as computers. How can that be? We do something totally different. The brain is much slower, not as exact and so on and yet we play almost as well. Why? Again, it has to be the case that the brain does something that the best AI computers don't do at all. It’s not that they don’t do it as well, they don’t do it at all.
[28:39] Dennis: What AI research does at the moment reminds me of remarks from Karl Popper’s autobiography. Now, he writes about psychology, but I think there’s the same underlying mistake. He says what psychologists are looking for is a theory of successful thinking. And it seems to me that AI researchers also want something like that because they want, say, chess programs that play successfully. They want text-recognition programs that recognize text successfully. But when it comes to real creative thinking, there is no guarantee for success.
[29:09] David: Yes. If the computer really were creative it could say: ‘This doesn’t interest me’ or ‘This doesn’t interest me anymore’ or ‘I’ll invent a new game’. Today's computers can’t do this. It’s not that they do it rarely. They never do it and are never able to.
[29:36] Dennis: Okay, while we’re on the topic of artificial intelligence. There are many who are very concerned about so-called ‘superintelligence’. This is about programs which - they haven’t been built yet but many are concerned that they will be built soon; apparently they’re right around the corner - and as soon as they get built they improve themselves so quickly and become more and more intelligent at a speed we could never catch up to. If this so-called superintelligence should decide to not like us, if it were malicious, then... that’s it for us. Then we are doomed. You discussed this with the philosopher Sam Harris, who is one of the better-known proponents of this theory. He believes this as well, that superintelligence would basically be fatal. What do you think?
[30:30] David: You can only believe that if you don’t understand what general intelligence is. It’s not about the brain, it’s about programs. Every universal computer can execute the same programs as every other (universal computer). Now, maybe we wouldn’t have enough memory for that or maybe our brain isn’t fast enough. But that’s all. There can't be any other difference between us and these superintelligent computers. But we have computers too, and could use them to think better. We do that already! We’ve been doing this since the invention of paper and pencil and computers, like nowadays. That will continue. And progress in intelligence or general intelligence that will come about in this way will happen to us just as it will to computers. So if you believe that this development will bring about humanity’s demise, not only do you have to believe that the computers will be faster and have more memory but also that the first superintelligent programs will be able to build machines that are faster and have more memory without us knowing, and for long enough so that once we finally notice, it’s too late because they have it (next-level computers) and we don’t.
However, if they are capable of that, they won’t need superintelligence. They could simply produce poison gas. And once you understand that this is the crux you have to ask yourself: Why couldn't humans do the same? And the answer is: They could! And that’s an important problem. And the same problem for computers is less serious because we will program those computers. The first ones – maybe not the second ones but the first ones. With humans it’s the opposite. There are already people who want to poison us... but no computers, including the first ones.
[34:21] Dennis: So there is an underlying universality to the human mind, which any artificial intelligence would also have. And because such an AI is already universal it’s a mistake to believe that there could be something beyond that point.
[34:37] David: Exactly.
[34:37]Dennis: This so-called superintelligence would also be an AI just like us - well, we are no AI but an I…
[34:43] David: AGI, yes.
[34:44] Dennis: Right. So there is no qualitative difference. Of course, just as there are more and less intelligent humans, there would be more and less intelligent AGIs, but there couldn’t be superintelligent AGIs on a level that we could never reach.
[35:00] David: Yes, and if they are (more) intelligent because they have a better brain we could have the same.
[35:13] Dennis: Okay, let’s take a break here. I again want to recommend to our listeners the book ‘Der Anfang der Unendlichkeit’. It is now available for purchase and there will be a second interview in which we will continue this conversation in English.
[35:30] David: Great. Good-bye!
Buy the German translation here: http://anfa.ng/kaufen