2021-12-13 Robin Hanson To What Extent Can We Predict the Future? With Robin Hanson and David Deutsch
Duration: 00:58:04
Transcript
Daniel Martin
This is Daniel Martin. I’m here with Robin Hanson and David Deutsch. Yesterday I was asked if I’d like to help moderate a conversation. So here we are. Have fun. And Robin, do you have a question for David?
Robin Hanson
David, it’s great to talk to you. I like you. I’m a big fan. I’ve spent a lot of my career doing various kinds of future forecasting. I hear you’re not entirely a fan of all of that. So I wanted to ask more specifically which kinds of forecasting you do or don’t like. And just to start with very concrete examples, so for example, solar power costs have been falling steadily and there’s a standard power law relationship that predicts that fall of cost. And in fact, surveys of technology show that more than half of technologies do follow that power law of cost as a function of the quantity produced. I presume you’re OK with that kind of forecasting.
David Deutsch
Not entirely OK, actually, because you seem to be when you say half of all whatever you said, innovations or whatever technologies that they forecast. You’re kind of implicitly assuming that’s the frequency, half of them. And you’re kind of assuming that that’s equal to a probability, namely, you know, where it’s going to be. There’s a there’s a probability of one half that in the particular innovation we’re considering, like solar power, solar power or whatever will follow that law. And that’s that’s the kind of inference I think isn’t valid. However, I think part of the reason I think it may not be as bad as it sounds, because part of the reason why people, for example, confuse frequency and probability in their their language and why they express forecasts in terms of probability and then use frequency to judge that is that there’s a there’s an unspoken assumption there, which is also about the technology. Like, like, you know, that that, for example, the maximum distance that a human will travel from Earth as a function of time may have followed a certain law until we reach the edge of the atmosphere and then followed a completely different law between the Earth and the moon. And then we’ll follow a completely different law again from the moon to Mars. And when you talk about solar power, you are implicitly referring not only to the statistics, but to a mental model that you have or theory or explanatory theory that you have about that specific technology. And although the answer you get is entirely derived from that explanatory theory and not from the statistics, it doesn’t matter if they match well. And for example, in the in the in the case of solar power, I think it’s correct, although that power law will eventually stop. And it may stop before the laws of physics say so, if there are contingent factors that stop it.
Robin Hanson
I mean, I think you are right that we’d like to talk in terms of frequencies because it’s a safer thing to say. And we are implicitly trying to get someone to think in terms of probabilities as the thing we’re trying to get in their head. But because it’s harder to justify a claim about a probability than a frequency, we speak in terms of frequencies. And of course, whenever we see a trend like the solar power trend, we can’t be overwhelmingly confident that will continue forever. And we should have a substantial expectation that it might deviate soon. But nevertheless, we might say, well, look, if the standard here is just better than chance, if we’re just trying to understand the future, surely we have to do things like this, like find the kind of trends that technologies follow and pick the one that’s most common. And if it seems to fit the trajectory so far, surely we’d want to do something like projecting that forward. We want to have some degree of uncertainty about that projection. But isn’t it a reasonable thing to do to take our best fit to past data and then try to use it to project future data?
David Deutsch
I mean, I think it’s a reasonable thing to do if our best explanations imply that. So I think the futurologists of the 1960s had a phrase, a surprise free forecast and a surprise free forecast is them saying, don’t blame us if we turned out if this turns out to be false, because we will always be able to look back and say, ah, this thing happened and it wasn’t in our in our predictions, but we were surprised by it. Well, a surprise, again, they’re referring to explanations there. A surprise free, a surprising outcome is one that your best explanation didn’t foretell. So a surprise free one is is what will happen if your current explanation is true. But then, you know, everything hangs on your current explanation.
Robin Hanson
But nevertheless, surely, I mean, we do look at, say, solar power and it does look similar to many other kinds of technologies. And we are implicitly using our sense of what the nature of solar power is compared to other kinds of, say, electronic technologies to compare it to, say, electronic technologies or materials technologies. And then when we and then we are looking at sort of rates of patterns that we’ve seen in those. And we are using many things we know to make this judgment that, say, solar power would probably follow this power at least for a while. Nevertheless, isn’t it a reasonable thing to do all of that, taking everything we know and using that to make predictions about the future of solar power? I mean, I don’t know that much about the details of solar power, but I’m assuming that that is a good assumption.
David Deutsch
But you’re making an assumption there about a very complex and detailed thing. You know, when you say this is like other technologies, you mean not only that the graphs look the same, you mean that you expect solar power technology to be dependent on making factories which use materials of a certain kind, which aren’t going to be throttled by a hostile foreign power and so on, and that it’s not going to be something that you haven’t foreseen. Like, for example, if you were if you were predicting the future of power stations in 1900, you wouldn’t be expecting nuclear power stations. So your forecast, based on the best existing knowledge, would not be just out by a factor. It would be out by having the wrong explanation, which would make the real things behave in a different way, not just be quantitatively different. And then, as I’ve written somewhere, was it in The Beginning of Infinity? I can’t remember. But after radioactivity was discovered, you could have then made the prediction that, OK, technology will harness that. And then you wouldn’t have been able to predict the environmental movement, which will sabotage that. So you would have been completely wrong again about.
Robin Hanson
But yeah, I mean, if the standard here is, can you ever predict things perfectly with certainty? Of course, we’re going to agree. No. But if the opposite standard is you can’t ever predict anything whatsoever compared to some absolute chance, that’s wrong, too. So isn’t obviously the question, what can you predict? How well not can you predict the future? And then we want to go through particular topics and just point out the various degrees to which we might be more or less confident and accurate in any one prediction. So isn’t it come down to those details?
David Deutsch
It comes down to the explanatory details. So like, you know, I like to say when people ask me about, say, Bitcoin, I talk in terms of whether someone’s going to win the Fields Medal or not. For cracking it. And that makes it clear that part of the answer to this question is unknowable. I can’t tell whether someone’s going to tomorrow discover a piece of mathematics that wins the Fields Medal and invalidates Bitcoin.
Robin Hanson
So if we think about predicting a longer time scales, of course, all else equal, that should be harder. But then if we go to the most solid things we know, the things we’re the most confident in, it would seem like that’s where we have the best chance of making the longer term predictions. And so since I tend to make longer term predictions and you might be especially, you know, wary of those, then it would seem like to focus on these most solid predictions. So one of the most solid predictions I would base one on is, say, the speed of light as a limitation on causal influence. I might say we’re limited to our future light cone. And that has consequences, substantial consequences for the future. Now, I’m going to admit that, of course, yes, that might be wrong, but surely it’s reasonable to tentatively guess that maybe that is a fundamental limit and explore those limits, making them explicitly conditional saying, yes, this could be wrong and here’s how, but we’re pretty sure of the speed of light limits. So let’s go with that and analyze that. So, for example, you know, I’ve done some work on predicting where aliens are in space and time, and that’s based on this assumption of the light cone limitation. If the light cone limitation weren’t there, then basically any alien anywhere in the universe could be here right now because there would be no limits on how fast they could get places. And therefore, if there’s no aliens now, we would have to predict there’s no aliens anywhere because or anyone who wants to get here because anyone could get here. But with the light cone limit, then we could say, aha, only the aliens and our backward light cone could be here. And therefore, then we can make many more concrete predictions about, you know, the nature of the advance development of advanced life based on that. And that’s something that I have done recently. And it seems like, again, a reasonable tentative thing to do is to assume that light cone is limited. It’s also another basic implication is that, you know, it seems like exponential growth in things we care about is hard to continue in a light cone limited world because the light cone only expands the volume at a cube rate and exponential growth requires much faster growth. So isn’t it reasonable to say, well, it looks like exponential growth can’t go on forever if we’re limited to the future light cone. So I think that those kind of speculations or predictions are very legitimate. In fact, we ought to be making them because it is a way of exploring the consequences, the cast iron consequences of our existing best knowledge. And there, I think it makes very little difference to say, whether you make the mistake of saying we’re doing this because we’re relatively very sure of the speed of light limitation and so on, or whether you say this prediction is subject to no Nobel Prize being won
David Deutsch
Invalidating it. Those are pretty much the same thing, although I think it’s just wrong to say that we’re sure of something. We should not believe things. But for some purposes, like in this case, it doesn’t matter. But although it is very legitimate to make speculations based on the finiteness of the speed of light, I think going on to say that will then have consequences for what we say about our future, that’s a bit more a bit more rickety because we don’t know that the limiting factor in our future will be accessibility of stars, basically, or galaxies in the longer run, because we might we might expand downwards. Like Feynman said, there’s plenty of room at the bottom and there’s not an infinite amount of room at the bottom. So then I would invoke thermodynamics and the fact that entropy seems to be finite and to say that’s going to set a finite limit on the amount of room at the bottom. And that also sets a limit on our future growth in terms of entropy, because entropy is bounded by a finite spatial density of entropy and a finite volume by the speed of light. Now, now you’re talking about things that are not as secure as the finiteness of the speed of light, for example, that there isn’t an infinite amount of room at the bottom is something that crazy theories of quantum gravity have contradicted. They may not be our best theories, but they’re not they’re not completely to be ruled out. They don’t predict infinite amounts of negentropy at the bottom. They predict infinite amounts of potential structure if you add a negentropy, that is, you …
Robin Hanson
Can add enough free energy to a system, then you could create lots of small scale structure. But, you know, we do have even within quantum gravity, standard theories of the finiteness of negentropy, say within the volume or within a black hole, etc. And then I could appeal to those. But I’m happy to take your point that any one analysis could have more ways it could be wrong. We could sort of count the ways things could be wrong. But we’d kind of like a way to weigh all the different ways something could be wrong in some sort of overall measure. And that’s what people usually use probability for. So you don’t want to use probability, then you need some other way to count. But we do roughly want to see for any one thing how much it depends on things we might be wrong about and make a judgment about focusing on the things we are going to be the most confident of. And it may not affect the bottom line that much in many cases. But I think
David Deutsch
It is simply wrong to use probability in that way. And it would be better to make the assumptions explicit rather than say, you know, well, well, you know, it’s not going to be we’re not going to supersede the second law of thermodynamics or something. By the way, thermodynamics, at least in one version of it, depends on information theory. Information theory has been overturned. I mean, Shannon information theory has been overturned by quantum information theory. So, for example, supplanted, perhaps is a better word. I mean, you know, still most of the most of the results of information theory are still true in quantum information theory. Most of the usual claims about information still continue there. So the entropy remains finite in quantum information theory.
Robin Hanson
It does. But the fact that entropy is governed by slightly different laws and that these allow exponentially more information to apparently exist in some situations and others means that we have to change our conception of information and then our conception of information may or may not feed into the practical conclusions that we want this information theory to inform. But the amount of negentropy in this class in this class is not substantially changed by quantum information theory. We can still say, you know, if that’s all you have, then and you can only grow it at a certain rate, then that means our future economy descendants can only grow at a certain rate. I mean, that’s still a robust conclusion, even if we switch to quantum information. It happens to be so for quantum information theory. But quantum information theory allows information channels to transmit twice as much information as classical channels. So it would have been, you know, if we’d been having this conversation 40 years ago or whatever it is, whenever channel capacity doubling was invented, we would have said, you know, we know for sure that if an object can hold to this, it can be in two distinguishable states and used for information transfer, then the rate at which it transfers information is also one bit per unit time or whatever it is. But it’s actually two. So we couldn’t have predicted that it won’t be 10 or 50 or infinite. Say you were with me on a project to develop a nuclear reactor for thorium, you know, you might have some qualifications for that, right. And say I’m a part of this project and I’m trying to work out a tentative design for our thorium reactor and some tentative ways to test this. And I make some suggested claims to you that this particular design could have this rate and this density and this heat conductivity, etc. And then for any tentative design I offered you, you could play the same game you’re playing now and say, well, yeah, but you’re using standard information theory and what about potential future information theories? And you’re assuming standard gravity and what about potential future gravities? And I might say, well, you’re not helping very much to design this reactor if you’re going to point out all the logical ways it could be wrong. Don’t you want to, like, make some tentative assumptions and get in the game?
David Deutsch
Right. Let’s help design something.
Robin Hanson
And that’s the way I want to ask about the future. I want to say, don’t we want to think about the future? Yes, of course, we could be wrong. Shouldn’t we, like, take our best understanding and make some tentative guesses here?
David Deutsch
We could be wrong is a is a bad objection because you could always make it. And so you’ve got to you have to reject on principle all objections to a project that are of the form. We could be wrong. Our best explanations could be wrong. So I’m not objecting to that part, and I wouldn’t be making those objections to the thorium reactor. And I don’t make those objections to your analysis of future aliens occupying galactic clusters and so on. It’s all legitimate stuff.
Robin Hanson
Well, then let’s talk more specifically about some other ways we could guess about the future. And then we could say specifically, like what assumptions we’re worried about being wrong that such might derail the analysis. But let’s talk about particular things we could guess about the future. Right. So, for example, we might say future stock prices will follow a random walk, just like current stock prices do, because our theory of why stock prices follow a random walk is very solid theory about information. Basically, it’s an information based theory about the random walk. And you say, as long as the information theory holds, then future prices should follow a random walk. That sounds like a reasonable long term claim. I mean, I go out a million years and say a million years in the future, stock prices will follow a random walk, because unless information theory is dramatically changed, that’ll just be true.
David Deutsch
Yes. So I would. So I agree with the conclusion. But I would be explicit about the explanatory assumption that to me, the explanatory assumption there is basically that if that were not so, then a speculator could make money out of it not being so. And eventually it would become so. And therefore it will be so. So I think that’s an epistemological argument fundamentally. And therefore, I would say it’s we can expect that to be true until somebody wins a Nobel Prize in economics for showing that the epistemology isn’t what we think it is. But for most purposes in the world, it’s perfectly reasonable to take what we think we know now and apply it, even if a Nobel Prize might be won someday for showing that it’s wrong.
Robin Hanson
Right. Yes. And so. So, for example, there’s a whole literature on evolutionary investment, which talks about if there’s an evolutionary competition between different investment strategies, what would be the investment strategies that win in the long run? And that’s based on something called the Kelly rule. And there’s a theorem basically that the maximum investment would have to invest in proportion to the future final value of different investment categories. And that seems pretty solid to me, at least given the assumptions that there’d be some evolutionary competition among investors, such that the one who would invest more successfully, reproduce and become more common.
David Deutsch
Yes. And underneath the assumption that there might be investors who have certain objectives, there are, again, epistemological assumptions about how success in achieving things can come about when people cooperate. And I think those theories are true. And therefore, as I keep saying, I think in many cases, the bottom line is the same, whether we are explicit about the explanatory basis of the prediction or not. But in some cases, you can wander off. I think you just did it just now. When the speed of light being constant is a different assumption, different species of assumption from the assumption that there aren’t the amount of information carrying capacity at small scales is not infinite. Those are two different kinds of thing, because one of them is kind of saying, well, we don’t know about it. Therefore, it’s probably finite because most things are. That is a different kind of assumption from one that says here is a physical limit and there is no rival theory on the horizon.
Robin Hanson
So, I mean, as you know, say, Bekenstein’s bound does give a finite entropy within a volume. And of course, you know, when we actually look at physical systems, we do tend to measure their entropy via temperature, for example, and we measure sort of total entropy by the integral over what would happen if temperature went to its limits. And so we do have physical estimates of the total amount of entropy in physical systems we actually deal with based on the physical properties. And those quantities are finite. Right. That’s a very standard thing to find in actual physical systems we interact with. And so it’s perfectly reasonable to at least tentatively say these look like finite entropy systems. The glass of water here is finite.
David Deutsch
The Bekenstein bound is a good example because actually Bekenstein had several bounds and I think one of them is solid, the one you mentioned, but he had other bounds. One of them turned out to be wrong. And I wrote a paper showing it was wrong, which purported to limit the maximum speed of a computer. And that was just wrong. It was now it looked right because the arguments were very similar to the arguments about the entropy being equal to the surface area and that kind of thing. But it wasn’t right. And I have actually authored a paper recently where I and co-authors corrected other people on mistaken judgments about quantum computation quantum quantum just entropy of computation bounds. So I think I think there’s a difference. And we could have known there was even before we discovered those errors, we could have known there was a difference between a Bekenstein bound and the speed of light. And that is sure I can find out is is is rests on ultimately on assumptions that aren’t the direct result of well tested, well explained physical theories, whereas the speed of light is.
Robin Hanson
But you and I are right now in the context of a larger intellectual world, which accepts most of the same basic theories that you and I are now accepting, but which mostly puts a mental block on analyzing the future. They basically go, that’s too speculative. We’re just not going to do it. Right. And so I mean, right. And so you and I can at least share this approval of let’s take our best theories and do the best we can to project them forward. And so, again, I’m interested in going with you there to sort of think about. So I just gave you sort of the evolutionary analysis or I pointed to the evolution analysis of investment returns. But even a more basic one is the idea that we have a discount rate today. So today people discount the future in the market and in their private actions. And that is a big issue with respect to say global warming or something, because the market value of distant future consequences is very low. And according to that usual discount rates, we shouldn’t bother to do much to prevent future harm. And people often think that’s wrong. And we have an evolutionary analysis that says that gives a plausible explanation for why we evolve such a high discount rate, which is to say that the analysis suggests that we often have a choice between investing in ourselves or investing in our future generations. And so that gives because our descendants have typically a factor of two difference in their genes related to us, that we then have a factor of two per generation discount rate, which does roughly match the typical market discount rates. And so that’s an explanation. And it says, well, we’re failing sort of to collectively care about the future because we each care about the future generation half as much as ourselves and asexually reproducing organisms wouldn’t have this discount rate, according to standard theory. And so these theories of sort of self-reproducing investment funds, they would be asexually reproducing. And so they wouldn’t have a discount rate. And so the prediction is that in the long run, something will arise that doesn’t have this long term discount rate and it will spread because it will care more about the future and so we’ll have more future consequences. And so my prediction based on this analysis is long views are coming, even though today we don’t care much about the future because of this inherited biological tendency to discount the future factor of two per generation future, more asexually reproducing organizations and units would care more about the future and therefore they will eventually care about the future. So eventually people will care about the future, even if not soon. That’s a prediction I will make about the future. Our descendants will care about their future. So I see you can point out the potential errors and the potential ways the analysis could go wrong and we should do that. But still, it’s worth having projections like that, at least to tentatively put them up and say, does that make sense?
David Deutsch
Well, the idea that decision theory should be analyzed in terms of how much we care about our future selves or about our descendants, I’m very skeptical of that theory because it’s a subjective psychological theory of really an objective thing, which is the growth of knowledge. So I haven’t actually heard before this theory of whether offsprings matter.
Robin Hanson
I mean, there’s also the I don’t know if you’ve studied the case as well. There’s sooner or later we’re going to go immortal. And then we won’t have offspring. And how will we value our future then? Well, I think this can’t be deduced from biology. It can’t be deduced from psychology. There will be objective factors. So evolution hasn’t encoded preferences in us for what happens when we get immortal, because that never happened. So at the best, it would project the preferences we have across our lifetime into the future. But that does seem to be roughly this factor of two per generation that we at the moment are discounting our own personal immortality with. That is, if we ask how much is it worth to us to prevent ourselves from dying in a century, we don’t actually willing to put very many resources into it because of this factor of two per generation. But over time, that will change. And then I might use an evolutionary analysis to talk about how that would change. I might say, well, when there are immortal creatures and there’s selection, variation and selection, then what kind of preferences will be selected for? That seems to me the way I want to do the analysis. But it seems sounds like you’re somewhat skeptical of analyzing preferences for time using evolution.
David Deutsch
Well, you’re using really biological evolution because it could be that this number, the discount rate, is dominated by, say, meme evolution. Suppose that some idea spreads, which say that it ought to be some or imply that it ought to be some value, then you will say, ah, but then there’s group selection because some other some subset of the people will not have that meme or religion or whatever it will be. And if they can produce weapons faster, then they will in the long run win. So that’s group selection. And as you know, group selection is an extremely problematic concept. And I’m inclined to go with Dawkins and Co. To say that group selection basically never happens, even in the biological case. And when you then take it to memes and societies and civilizations, it can go wrong in all sorts of other ways as well. There just aren’t enough subgroups of society because there will be subgroups within those subgroups who think that what they’re doing is wrong. Our whole society at the moment, the entire world, is dominated by ideas which say that we shouldn’t make progress on principle, either in particular cases or we shouldn’t go too fast or whatever. The idea that we shouldn’t make progress as fast as possible, such that our point of view will become dominant in the long run, is almost held by nobody. And you’re assuming that in the long run, that view will be superseded by some other view that you think is more viable. But it need not be so. It’s very rickety, it seems to me.
Daniel Martin
So, David, I think you disagree with Robin and just in the nature of what he’s proposing as well. So, for example, Robin has also said that, proposed that we humans have explicit goals in the future, we will evolve such that we may have explicit goal to have more descendants. That being an idea in our mind, it’s like a freely floating propensity, which our future generations will have more than we do. And the idea of this discount rate, my model of view, is that you see thoughts as highly iterated and things like culture and memes now dominate. So how would this freely floating number about, say, one half or two get in there?
Robin Hanson
So I agree, Daniel, that the key issue here is to think about what are our best robust models of evolution that includes different potential units like groups and includes culture as well as physical genes. That is, if we thought, say, a simple model of physical gene evolution was, you know, old hat, not really applicable, and that the conclusions I was talking about were dependent on that particular set of assumptions, then we might say, well, that’s just the wrong set of assumptions. It’s not relevant. I think the conclusions that I’ve talked about are relatively robust to the different size units of selection and to culture being part of the co-evolutionary package. And so I’m certainly happy to grant and am interested to think about that at the moment our genetically induced and culturally induced packages of behavior are far from equilibrium. That is, they’re far from the thing that would be selected for in the long run. And something has gone very wrong in the past to induce that. And I’ve thought a lot about what that could be. But nevertheless, when I make the long term predictions, I set aside current behavior and just ask what in theory would be selected for not just genes, but gene culture coalition packages. And most of the analysis I know is robust to which kind of packages those are. It doesn’t matter that much whether behaviors encoded in textbooks that people teach or in genes that influence individual construction, whatever it is, there’s evolutionary analysis of what sort of behavior would win out in the long run.
David Deutsch
You think things will equilibrate rather than staying out of equilibrium where culture will run away from you?
Robin Hanson
Yeah, that is the robust prediction that evolution would say. Yes, if some people have behavior that produces much fewer descendants or some people, culture packages do, then those packages will lose out. They won’t reproduce as much. And in the future, there’ll be more of the packages that do reproduce more.
David Deutsch
If the universe were infinite and there were infinitely many cultures and civilizations and so on, and if they were uncorrelated so that there was no particular reason why a particular type of culture might take over, even though it’s disadvantageous, then you could make a robust evolutionary theory of it. But I don’t see how it can be robust in predicting what errors people will make, for example. Like I said, in 1900, they couldn’t have predicted the environmental movement or nuclear power and all the ideas that have informed the use of those things. Now, yes, if there were a million Earths, well, actually, I think no, it’s not yes. Even if there were a million Earths or a trillion or whatever, there is no way of using evolutionary theory to predict what kind of mistakes of this kind will be made by most of them. There might be a theory of cultures that says that certain types of mistakes will be made. Some people say all cultures destroy themselves because they become weak and so on. I don’t believe that theory, but I’m just using it to illustrate the fact that there is a substantive assumption there that underlies the assumption that evolution theory alone can determine the outcome.
Robin Hanson
So as you know, our standard theory of thermodynamics both has equilibrium versions. It has fluctuation versions. The fluctuation versions predict how fluctuations change over time and decay to equilibrium. Of course, we have similar alternatives in evolution. That is, evolution isn’t just a theory of a final stable state. It’s also a theory of how initial deviations will decay or move toward the equilibrium. I think those parameters suggest that Earth is plenty big and the future is plenty long for us to move from where we are now to closer to an equilibrium. Do you have time estimates that say it would take a trillion years for us to equilibrate now?
David Deutsch
I think there’s no reason to think that we’re going to equilibrate because it depends on the creation of knowledge, which is not a relaxation process. It’s an error correcting process. Error correcting processes can go wrong. They’re not infallible. In fact, they go wrong all the time. And what you’re proposing is that there’s an overarching reason why things will come to equilibrium, in which case, you know, if you’d been around four billion years ago, you would have predicted that humans will never exist and that nothing from Earth will ever get to the moon.
Robin Hanson
Let’s just take concrete example. Today, religious people are having higher fertility than non-religious people. So the straightforward prediction is that over the next generation or two, there will be more religious people. At least people who descended from religious people. Now you might ask, will the children of religious people be as religious as their parents? Of course, that could be a thing that then means that religion doesn’t actually increase. But we can certainly expect a few generations now more people will have descended from religious people than will have descended from non-religious people because that’s the selection argument. You could say, well, some we might discover a new thing that suddenly helps the non-religious people, you know, have a big advantage. And yes, we might. But just like with the stock market, is there any particular reason to expect the correlation will be that way?
David Deutsch
If we don’t have a particular reason, then again, our expectation is that on average. That’s an explanatory assumption. And that explanatory assumption is a very kind of arbitrary assumption compared with the things that it’s kind of purports to be about, which is very rigid connections between the population at one time and the population a generation later. Some things about that relationship are rigid or, you know, they’re rigid in the sense that the explanatory theory underneath it doesn’t have any rivals, but any viable rivals. But there are other things which aren’t rigid. And, you know, people’s attitude towards the number of children they have changes on a time scale of a generation at the moment. It’s, you know, it’s that fast. So, you know, it might settle down and there might be an asymptotic form of it, but things that contain Turing-complete subsystems don’t have asymptotic forms. That’s, you know, you can almost define a Turing-complete thing as a thing that doesn’t have an asymptotic state if it runs forever. So, therefore, I think it’s a mistake to make biology type assumptions about human growth of knowledge, just as it’s a mistake to make chemistry type assumptions about the growth of life.
Robin Hanson
So, as an analogy that perhaps, you know, on a faster time scale, we have somewhat of an evolutionary theory of growth of the economy in the sense that we say that there are many firms in an industry and some of them are more productive than others and the more productive ones consistently grow relative to the others. And so we predict in some ways that each industry will be more productive in the future based on an evolutionary selective argument among the firms in the industry. It sounds like you’d say, no, we can’t predict that industries will get more productive in the future because that’s just making arbitrary assumptions about correlations between culture and productivity of firms, which we have no reason to believe. And then do you not believe that industries will get more productive in the future basically by the selection of the more productive firms winning?
David Deutsch
As I said, I resist believing things, but I think it’s a good theory. And, you know, if I want to plan my pension or whatever, then I would assume that there will be economic growth in the future.
Robin Hanson
But the reason is… And it’s a selection argument. It’s an evolutionary selection argument, which is one of the primary reasons we believe there is economic growth.
David Deutsch
Well, I don’t know about primary. That argument depends on things like civilization continuing, you know. Well, yeah. If again, in 19… No asteroid destroys everything. Yeah, of course. But, yeah, well, but, you know, no world war destroys everything. No crazy economic theory adopted by the majority destroys everything.
Robin Hanson
And again, if you’re choosing your pensions, you’re going to make those assumptions, aren’t you?
David Deutsch
Well, I think I’m going to hedge against those events. To some degree, but you’ll still mostly bet that that’s the most likely outcome. Well, I have no choice. So it’s more like… You know, they’ve got this Bitcoin thing.
Robin Hanson
If you really think everything’s going to hell, you could just buy a lot of Bitcoin, right?
David Deutsch
I’m sure you can find some people willing to sell it to you. People say that. And the reason is that they are not being explicit about their assumptions again. And I would rather say nothing is secure. Therefore, I’m going to make my investments on the basis that if this investment goes wrong, I will have bigger worries than that. For example, it may be that democracy has broken down or that there’s a world war or something like that. So then I will have been wrong about my investment. I should have put all my money into rubidium because that’s the basis of the future weapons. But I was not to know that.
Robin Hanson
So when I try to do future analysis, one of the biggest contrary assumptions or scenarios that I focus on is what if we end up creating a strong world government that strongly regulates investments and reproduction and other sorts of things and then prevents the sort of competitive environment in which the evolutionary analysis applies? And I’m very concerned about this scenario. That is my best judgment of our biggest long-term risk in terms of the thing we should focus attention on is this creation of a strong civilization-wide government, say solar system-wide government, that is going to be wary of competition and wary of allowing independent choices and probably wary of allowing interstellar colonization. That is, this vast expansion into the universe could well be prevented and stopped by that. So I’m very into sort of a baseline set of assumptions and the conclusions and then asking what’s the most likely or the biggest thing that could go wrong there? And that would be my guess for the biggest thing that could go wrong. It seems more useful to have a concrete scenario like that that I can get my hands on and think about than just the general idea that, hey, who knows, you might be wrong.
David Deutsch
I agree. The concreteness is what I would call the explanatory-ness. I mean, that is explaining how things could go wrong. There’s also the fact that, though, that things could go wrong in a way that our existing explanations don’t predict. By the way, you haven’t yet spoken of a type of prediction that I don’t think you should be making or that I…
Robin Hanson
Good.
David Deutsch
All the specific examples you’ve given are… If you like, it’s a matter of terminology, but it isn’t a matter of terminology. It’s a matter of epistemology and fundamental theory. But the bottom line is you haven’t yet given an example where I would disagree with you that something is worth investigating. But I also think that if, let’s say, there’s going to be a conditional on our civilization or our species not surviving for the next century, I think it is overwhelmingly likely that the reason will be one that we haven’t thought of yet. We should work on thinking of more things.
Robin Hanson
Yes, absolutely. That’s my basic conclusion. I’m all over that. I’m into generating new scenarios for things to go wrong that other people haven’t been talking about. I have been generating such scenarios. Great.
David Deutsch
But this is just a special case of the fact that what we really should be doing is creating more knowledge, more general purpose knowledge, knowledge that is potentially applicable to applications that we don’t yet know. So one of them is the one you’ve just mentioned, but there’s also just doing pure mathematics or working on things that don’t even have names yet.
Robin Hanson
I agree we should have some broad just exploration of all possible areas of knowledge, which isn’t terribly directed by any particular goals. But then we should also focus our attention on particular things that seem promising for particular reasons. For example, I have this book, The Age of Em, which you may know takes a very particular technology scenario, which some people might find unlikely and just tries to work through in great detail all the consequences of that. My justification would be even if you think it only has a 1% chance, if it’s worth having 100 books on the future, then it’s worth having a book like this because with this and the other 99 books, we could lay out a picture of a wide range of options. And so that seems to me that we should just take particular scenarios that we can think are not crazy and work out their consequences to think about what would happen if.
David Deutsch
We should, but just as a side remark, you’ve caught me in an illegitimate use of probability when I said that I think it’s overwhelming. Given that on the assumption that the civilization or our species will be destroyed, then it’s overwhelmingly likely that it will be. I shouldn’t have said that. And this just shows how deeply this mistaken notion of probability, of ideas having probabilities has permeated our culture. Even though I hate it, I can’t help using it.
Robin Hanson
So as you know, there are betting markets out there and you can interpret betting markets as representing probabilities. And in fact, there’s standard theorems in finance, how you can represent basically all prices through some set of probabilities. And I’ve worked on mechanisms to create conditional probabilities or combinatorial probabilities. We can elaborate pretty large, complicated combinatorial sets. And we have a lot of literature that at least says these probability estimates are well calibrated. That is, you know, when it says 20%, 20% of the time that happens and that they are relatively accurate in the sense that compared to other ways that make these forecasts, they are, you know, giving you the truth more often. I mean, this seems to be pretty useful to me to have these systems that produce calibrated, accurate forecasts according to a probabilistic measure. You don’t approve of them, but you probably still want to use them. I mean, when you get a weather forecast that says 70% chance of rain, then that weighs on you more than 10% chance of rain. Don’t you listen to these things? Aren’t they useful?
David Deutsch
Yes. And I admit that I think the connection between what you might call risk and probability, that is the reason why risks can be approximated by probabilities and also the reason why frequencies in certain situations can be approximated by probabilities is an unsolved problem. And I think it’s a very important problem. And if I wasn’t working on other things, I would be working on that. I mean, I think-
Robin Hanson
Okay. So until you figure that out, it’s not too crazy for the rest of us to continue to use them. Right? I mean, since they seem to be useful so far, this appearance of use isn’t just an illusion, right? I mean, we are actually getting substantial value out of weather forecasts and financial prices.
David Deutsch
Assuming that there are no asteroids out there heading on a collision course is also a very useful assumption in practice. Because the chance is low, most people would say. Yes, but they’re wrong. We don’t know what the chance is. And we need this idea that the chance is low is a brake on the urgency of working out what the chance actually is. So we need to work on that. And the fact that life continues productively on the opposite assumption isn’t an argument. We need knowledge about asteroids. We need that. If there is an asteroid, there either is or isn’t an asteroid out there heading towards us in the next hundred years or in the next year. And that’s a fact. It has nothing to do with probability. Probabilities can’t help us. Once it’s in the sky, it’s going to be no good with this. We can’t sue anybody and say the probability of that was very low. God will just say, no, it wasn’t. It was one. It always was one. I know.
Robin Hanson
But when you have the right theory of asteroids, then it can make a stochastic prediction and you have chance. You need a theory of asteroids with a theory that predicts their distribution. We do actually have such theories. I mean, people have actually calculated the rate of asteroids with stochastic models like that. That’s a common thing. So our best theories of planetary formations say that gas giants ought to form in the outer stellar system. And indeed, they have in our solar system. But all the other solar systems we look at have them be formed near to their star and nobody knows why.
David Deutsch
And therefore, I don’t hold much with stochastic theories of asteroids either.
Robin Hanson
You might remember that the first steam engines didn’t have thermodynamics as a basis. Thermodynamics was generated after the steam engines. Once you have a thing that seems to work, we might be justified in just using it. And if you say it doesn’t have a good theoretical foundation, we might say, well, then go find it. But we’re just still going to use it until you prove theoretically that it’s badly founded, that we’re misjudging.
David Deutsch
I agree that not only can we use it, but we should use it. So as I said, everything you’re doing is not just legitimate, but at least with the things you’ve told me, not just legitimate, but morally required. And it’s a bit of a scandal that more people aren’t doing it. But I think the same is true of all fundamental theories, all fundamental branches of knowledge. We don’t know how to make quantum mechanics and gravity coherent or general relativity coherent. Nevertheless, we’re completely justified in using each of them in their typical current areas of application. Well, using them for predicting things like satellite orbits, but not for predicting things like how much information you can get into a finite volume. There, the theory itself says that it doesn’t know. And that’s exactly what we choose.
Robin Hanson
So if you could tell me where exactly is probability most problematic, then I will try to avoid it there, because you’ve thought more about where probability is problematic. But if you don’t mention a particular area is especially problematic, then I’m probably just going to keep using it because it’s just probability.
David Deutsch
Yes. In short, the place where it’s dangerous is where the thing you’re predicting depends on the future growth of knowledge. So you’ve given some example. It still works even then, like when you’re saying that the stock price of something will be a random walk. Even when knowledge accumulates, the stock price follows a random walk.
Robin Hanson
Yes. Yes, exactly.
David Deutsch
So there are cases, but there are cases where it’s very misleading. So for example, when you say that all long lived civilizations in the past have failed within. Well, they have failed all long lived civilizations. Say all civilizations that have lasted longer than 400 years, if you count ours as having lasted 300 years, all civilizations have lasted for more than 400 years have already failed. And therefore, ours will. That’s illegitimate. Because it’s making an assumption that the frequency is a probability. And here I have a substantive theory that says why it isn’t, namely that our civilization is different from all the others in the relevant way. But it doesn’t matter. Even if I didn’t know that theory, I would still say it was illegitimate to extrapolate the future of our theory based on our civilization based on past civilizations, because all of them depended on the future growth of knowledge. If you look in detail, actually, at how they failed, they all failed in different ways. But one thing you can say about it is that in all cases, more knowledge would have saved them.
Robin Hanson
But the growth of knowledge was in fact the major reason for growth for all of our past history. That is, even biology, the major reason for innovation in biology was the growth of knowledge for forager humans slowly over 2 million years. They slowly grew by accumulating culture, the growth of knowledge, the farming culture, accumulated doubling roughly every thousand years through the growth of knowledge, primarily via sort of different kinds of plants and animals were domesticated. The growth of knowledge has always been the dominant factor in growth for all of history. That’s still true today. So that fact doesn’t distinguish us from our ancestors. Maybe the way we grow knowledge or the rate at which we can grow is the distinction. But the fact that knowledge was growing and was the main long term force has not changed.
David Deutsch
Yeah, it’s the way we grow knowledge.
Robin Hanson
And then we might find that our descendants will grow knowledge even faster. And I’ve thought a lot about sort of the ways that our descendants might grow knowledge faster than we do. And what are the obstacles to our growing knowledge now? Yeah, I’ll take a difference, not just the difference of scale or rate. Yes, it’s qualitative.
David Deutsch
Again, just like the life as a chemical process is different from all other chemical processes. And thinking as an information processing process is different from all other information processing processes. Therefore, when you try to predict one in terms of frequencies derived from the other, you’re going to make mistakes. You’re going to make bad mistakes. Always, by the way, in the direction of pessimism. Because it’s not just that we are unlike previous civilizations. We’re unlike them in a systematic way. Go ahead, Daniel.
Daniel Martin
It’s a big topic and people use probability for more than just matching to frequencies. Robin Hanson has papers on priors and stuff and there’s more there. But it’s a big topic. It sounds like we agreed to talk for about an hour. We’ve talked for about an hour. I feel like we’re at a sort of a resting… We’re not in the middle of some dispute or something like that.
Robin Hanson
Yeah, well, very interesting. I mean, I think we should publish this and see what people say.
Daniel Martin
Absolutely. If they like it, maybe we can do it again. And we’ll make them take sides. Who’s the winner here, right?
Robin Hanson
I’m not sure we need a live vote. I don’t think we need a winner.
Daniel Martin
I was just teasing you about the previous debate.
Robin Hanson
Okay, it was a nice conversation. So nice to see you. All right. Thank you for hosting, Daniel.
Daniel Martin
Welcome.
Robin Hanson
Bye-bye.
Markdown