In Conversation with Dr. Paul Bartha: Precautionary Reasoning and Analogical Reasoning
In this conversation, I speak with Dr. Paul Bartha, Professor of Philosophy at the University of British Columbia.
CiTR 101.9 FM / The Blue Hour
Recorded live on June 23, 2026
You can listen to the episode on CiTR.
The Blue Hour, hosted by Farha Guerrero, airs live every Tuesday at 2 p.m. on CiTR 101.9 FM in Vancouver and at citr.ca.
Paul Bartha works in Logic, Decision Theory and Philosophy of Science, and is one of the leading philosophers of analogical reasoning. He is the author of By Parallel Reasoning: The Construction and Evaluation of Analogical Arguments, published by Oxford University Press, as well as the Stanford Encyclopedia of Philosophy entry “Analogy and Analogical Reasoning.” Together, these works examine how analogies help us draw conclusions in science, philosophy, and everyday life.
More recently, his work has turned to precautionary reasoning, especially decisions involving environmental risk. In his presentation “Modeling Precautionary Decisions,” he examines how we should reason when the possible outcomes may be catastrophic, the probabilities uncertain, and the usual tools of decision-making may be inadequate.
Transcript
Farha: Paul Bartha, thank you for coming to The Blue Hour.
Paul: Thank you very much, Farha. I have a big smile on my face because it is always fun to talk about these esoteric ideas. I appreciate your patience, and I hope I can explain things to you and maybe to some of the listeners.
Farha: Absolutely. I have no doubt you will.
Paul: I guess anybody who likes philosophy is drawn to questions, because philosophy is often better at giving questions than giving answers, although I am interested in answers.
I thought you might ask me what connects these topics. I think it is the fact that we have pretty well-developed models of what is called rational choice, or modelling belief change, but there are these topics that live in the dark shadows. They are modes of reasoning that people use a lot, and it is much harder to capture them. Analogical reasoning is one of them. It does not fit very nicely in the standard models of reasoning.
Precautionary reasoning is another one. There are lots of missing pieces in your knowledge, and it is much harder to fit it into the paradigm of good decision-making, from the models we have.
Farha: What strikes me is what you said. Analogical reasoning and precautionary reasoning seem to emerge from the same human predicament, that we rarely have complete information. That is what you are after, right?
Paul: That is part of it. When people talk about the precautionary principle and precautionary reasoning, part of the motivation for this principal is that sometimes we need to make decisions quickly, and there are huge consequences at stake.
The standard cost-benefit model, or standard decision theory, there are different names for this mode of reasoning, works like this. You carefully enumerate all the options. Then you identify what possible states the world could be in. You assign various values to the different outcomes. You can think of it almost like a table. The states are the columns, the decisions, possible acts are the rows, and at each intersection there is an outcome. You say, this is the value I would assign. You use an abstract number called a utility, or a monetary value for economists.
Then you assign probabilities. What is the chance of this outcome arising if I take this act and it is this state? If you are thinking of something like, should I bring an umbrella, it works very well. It is either going to rain or it isn’t. I would assign it a very low value if it rains and I forget my umbrella.
You figure out the probabilities and then calculate what are called expected values for each act. It is a kind of statistical average, where you weight the utilities by the probabilities. It is a bit like gambling: is it worth doing this act? Is it worth doing that act? Which one has the highest expected value?
That is a really good model for most decisions. But you need to know a lot. You need to be able to estimate the probabilities and the values. But often in life we are faced with huge uncertainty, and we may not even know our values. One motivation for precautionary reasoning is that there are cases where we are faced with enormous looming catastrophe, like climate temperature rise. We do not know exactly what the probabilities are. We are starting to get pretty decent estimates, and the probabilities may be higher than people thought in the past, but in these cases we do not want to wait until we have all this information. So we need tools for reasoning.
There are tools in decision theory for reasoning under uncertainty, but the precautionary reasoning case is sort of in the middle. You know something about the probabilities and something about the values, but you need a way to make a decision.
Farha Guerrero: You recently gave a lecture about precautionary reasoning. But before we get into that, what first drew you to this research? Not what drew you to philosophy, but what drew you to this topic?
Paul: I came into this through a back entrance, I guess. I was interested in Pascal's wager, which is a classic argument in philosophy. It is one of the most famous arguments in philosophy, and I do not think the argument succeeds.
It is an argument that you ought to wager, as it were, for God. If you have a choice between trying to adopt a religious point of view or not, and either God exists or does not, it is the same kind of table I was talking about a minute ago.
Pascal famously said there is a possibility of an infinite reward if you believe and God exists. If you wager and God exists, there is an infinite reward. All the other outcomes in the table are at most finite. Maybe there is a small gain. You are certainly better off not believing if God does not exist, but it is a finite gain. As long as there is some probability that God exists, the expected value of wagering is going to be infinite.
I was fascinated by this argument, but most people are not that excited about it. They are much more interested in infinitely bad things.
Farha: I like that.
Paul: That is where I realized exactly the same kind of structure comes in with these precautionary arguments. Because there is a chance of some terribly bad thing happening, and that seems to outweigh the good.
Is that wise, to reason in that way? All the same objections that have been raised against Pascal's wager apply to this precautionary reasoning about the environment. I did not get into this through environmental ethics, but now I am interested in environmental ethics.
Farha: You absolutely are. In your recent presentation, you bring up things that are very familiar to us: the pandemic, debates about pipelines, and certainly climate change. Your new research is moving in interesting directions that are really pertinent to today's world.
Paul: Years ago, I was a fan of Buffy the Vampire Slayer, and there is some line about the plural of apocalypse being apocalypses. We are faced with a whole bunch of apocalypses now.
Farha: Apparently.
Paul: There is the pandemic, AI extinction, wars, nuclear risk, genetic tinkering with designer babies, and all this strange stuff. It is true. We are in a golden age for disaster and apocalypse.
I will admit right off the bat that this work is at the alpha stage. It is not something I would dare to apply too directly, although there are some examples where I have tried to apply it to decisions. One of the things I argued in this paper was that there might be cases where we want to use normal reasoning. Most of the time, normal cost-benefit reasoning is perfectly appropriate. But when there is uncertainty about probability and when there is something catastrophic, and I think this is true even if you do have a pretty good grasp of probabilities, you might still want to use some kind of precautionary reasoning, because there is this sense that standard cost-benefit reasoning does not help protect the things we care about the most.
This came out at the beginning of my talk. By the way, that talk was my presidential address. I was the president of the Canadian Philosophical Association until about a week ago.
Farha: Did you retire?
Paul: It is a one-year term. I am now past president. My year of living administratively is coming to a close. I will not be department head, and I will not be president of the CPA anymore.
The idea I tried to get across was that this notion of catastrophe is a relative one. It can be applied to existential threats, but it also applies to local perceptions of something being catastrophic. There can be something like a local catastrophe. I used the example of Northern Gateway.
Farha: Which is a really good one.
Paul: It has come back now. When I first started thinking about this about ten years ago, it was just before the Liberal government killed this Northern Gateway idea, which was building a pipeline from Alberta to Kitimat. on the Douglas Channel. It would carry bitumen, then it would be loaded onto tankers, and the tankers would steer around this treacherous channel.
I had a quotation in the paper I sent you from a local Indigenous leader, Nadleh Whut’en Larry Nooski from the region. He said that nothing would ever persuade them to take this chance. “We do not care about the value, this is valueless for us.”
I thought that, this kind of thinking, people who will not make trade-offs, who will not compromise, is characteristic of precautionary reasoning.
Even more than ignorance of probabilities, I think the hallmark of precautionary reasoning is when there is some discontinuity in value. Some outcome is so bad that you are not prepared to take any gamble.
Many people would say that is irrational. We make these gambles every day. When we get in a car, we take the risk of a car accident. You could be paralyzed. I understand that objection, and I think I have a way to answer it.
But there are certain big one-off decisions where you do a much better job of capturing the reasoning by modelling it in this absolute no trade-offs, no compromise way. I think it would be wrong to adopt a formal decision theory framework that makes that irrational right from the get-go. I am trying to develop a formalism that allows that kind of reasoning to be potentially rational.
Farha: That is a very important point, because these are Indigenous communities that often object to things like pipelines, and they have a different worldview. There is a disconnect between cost-benefit thinking, or thinking driven by business or incentives, and the way they see things.
What I really like about what you are trying to investigate is how to understand that type of thinking that seems illogical, but is not to them, because they see the potentiality, even if the potentiality of an oil spill is very small. They see it as something worth protecting, to ensure that rivers and streams, animals and wildlife and fish, do not suffer.
Even if it seems illogical based on the numbers or the chances of catastrophe, and you give other examples, like the Manhattan Project, you can see how human beings wrestle with these types of decisions. But they are not easy to reduce into something simple.
Paul: I should say that if you take a cost-benefit approach, you can get the same recommendation. You can get the recommendation not to build the pipeline based on cost-benefit analysis. You might think there is too high a risk.
The Manhattan Project is a great example. There was something that would have been an absolute global catastrophe. During the work on the Manhattan Project, the scientists were worried that a nuclear explosion might ignite the nitrogen in the atmosphere and destroy the whole world. They paused the work and spent a few days doing calculations, and they realized there was no chance. In the line from Arthur Compton that I included in the talk, he speaks very much like the Indigenous leader. If there is a chance, he does not care how small, we cannot go on with this.
So in cases of a clear global existential threat, it seems like this type of reasoning has a place. But I think it even has a place for smaller decisions, or at least I do not want to rule it out. I still have not made up my mind. There are cases where, if you have a certain set of structured preferences already in place, you should just use cost-benefit reasoning. But if all you have is an intuition that this thing is catastrophically worse than that thing, and you do not have well-formulated preferences, then you are reaching for a different decision framework. That is where this comes in.
Farha: That has to excite you as a philosopher, because this is a territory you want to chart.
Paul: Yes. The overriding framework for rational choice is called Bayesian decision theory, Bayesian epistemology. It uses probabilities and utilities just the way I described it, and it is great. Most of what is done in this framework is wonderful. But it does seem to leave certain dark corners that you cannot explore. I would like to make sense of this type of reasoning.
I feel a kind of humbleness with Indigenous perspectives. I do not fully understand them, but it comes up in a number of places, the importance of relationships with nature. The peoples I have talked to are not thinking in terms of rights and duties. They are thinking of our relationships with nature and parts of nature. This might be a way of forging a link. I do not think they would formalize it the way I have done it, but I have looked at attempts to think about rights of nature, and those are also attempts to apply “Western concepts”, concepts from the tradition of rights that evolved in the Enlightenment, applying these to Indigenous perspectives.
Maybe we can apply rights to nature.
This is a different approach.
There is an idea of infinite value lurking in the background; that nature has infinite value. In a way, precautionary reasoning is a way of capturing that infinity, but in a way that allows you to do some trade-offs.
For example, the paper shows how to deal with simple problems. I call them simple problems, where there is only one possible catastrophe, like Northern Gateway. Those are not so hard, and you might not even need the precautionary principle for those. But what happens in a case like COVID, where you are dealing with more than one catastrophe? There is a looming health catastrophe if you do not lock down, but there is also a potential economic catastrophe if you lock down too aggressively. There may be more than one catastrophic outcome, and that turns out to be common.
I have seen people object to the precautionary principle with an example like research at CERN. What if they created a black hole accidentally and it swallowed up the whole earth?
Farha: I have been to CERN.
Paul: I have never been there. I would love to go. If even one scientist is worried about this black hole, should we halt the research? On the other hand, we might need the research to prevent some other kind of catastrophe. What are we supposed to do?
For many years, the problem with the precautionary principle was that it was just a qualitative idea. If there is a looming catastrophe that exceeds a certain threshold of seriousness, then you should take the precautionary act. I talked about this tripod: a harm threshold, a knowledge threshold, and a precautionary action. But the criticism was, what are you supposed to do when there is more than one catastrophe? You get paralyzed.
The machinery, as I call it, of this lexical approach gives you a way to do trade-offs among catastrophes. It focuses your lens on the catastrophic angle of things and ignores the lesser concerns, just focusing on preventing the catastrophes. It is not meant to be used everywhere.
A good example where it might apply is logging. I looked at a discussion of old growth forest logging in B.C. If you use the cost-benefit model, sometimes called ecosystem services, the idea is that you assign values to each tree. You try to monetize the aesthetic value of having a forest there, then you apply cost-benefit reasoning. A logging company might say, we should log a lot of this old growth, but leave something like Cathedral Grove on Vancouver Island. It is a little patch of old-growth forest, and we get this aesthetic value. You can go to Cathedral Grove, and that is one answer, a very aggressive decision in terms of the environment.
At the other extreme, you might have somebody like Aldo Leopold with this land ethic, where anything that harms the beauty of nature is wrong. Basically, no logging that mars nature. Some environmentalists might advocate that approach.
I am trying to give honest answers about what my theory would show. If you make the catastrophe something like the disappearance of a viable old-growth forest, then it turns out that the principle recommends halting logging at the point where the probability of that occurring starts to go up. So it is somewhere in the middle. It allows some logging, but you should halt the logging at the point where you are starting to raise the chance that it will not be a viable forest, that it will disappear. You cannot know this 100 percent, but that is the point where you should halt.
It is a more conservative position than the cost-benefit approach. It may be a little more favourable to other considerations than a position like Aldo Leopold's would be. I think you get recommendations that are at least conceptually somewhat balanced. Does that make sense?
Farha: I think maybe we will backtrack just a bit, because your undergraduate degree was in mathematics.
Paul: Yes.
Farha: Listeners are hearing the mathematical side of your research, and you use words like models. Could you give us a little bit of, from a philosopher's perspective, how you understand something that feels very scientific and mathematical? I could be wrong, but it sounds to me that you started off in math, and something drew you to philosophy. Could you bring us there for a moment?
Paul: That is correct. I did a math specialist degree first, and I actually started a PhD in math.
Farha: Really?
Paul: Yes, and then I dropped out.
Farha: Oh, wow.
Paul: I loved the style of thinking and the precision. It is kind of ironic that what got me interested in philosophy was actually a math course. It was not philosophy of mathematics. When people talk about philosophy of mathematics, they are often concerned with set theory and number concepts, and what those concepts mean.
For me, it was a second-year course where the professor had this really innovative way of almost torturing the math to get an intuition. He was trying to make sense of a geometric intuition about what is called the area of a manifold. Imagine a kind of surface area on a plane that is curvy, and you are asking how to define its surface area. There is a standard definition in math textbooks, and he wanted to give another way, as a limiting construction involving enclosing three-dimensional surfaces that gradually shrink.
I loved this, and I have put this same idea to work in some philosophy papers. If you really want to capture a concept precisely, you need these formal tools.
That interested me in philosophy. I took a year off, did a master's, and decided I wanted to move into philosophy. I knew I was giving up something in terms of depth of precision and the level you can go. But in return, you get a very broad range of problems that you can think about. That is what I like about philosophy. There is a huge range of topics.
I did not realize it at the time, but this puts me a little bit in company with the history of philosophy in the 20th century. In the early 20th century, there was a lot of optimism in philosophy of science, coming from new developments in logic and science. People thought we could make philosophy scientific and use scientific models to think about things. There are famous names from this period, maybe not famous to people who are not philosophers, people like Carnap and Reichenbach, who tried to develop these formal tools.
People later realized the limitations of these approaches. The real world is not so exact, and any mathematical model you adopt is going to be wrong in certain ways. That is absolutely true of my work. It works sometimes and does not work other times, but it is always interesting and instructive.
In the 1960s, a name more familiar to people might be Thomas Kuhn, with The Structure of Scientific Revolutions. This was a very famous work in 20th-century philosophy. He looked at the actual history of science and said that if you look at the way science actually evolves, it is very different from the way some philosophers tend to think of science. Then there was a reaction and a move toward much more historically oriented philosophy.
Now there is a spectrum. Some people are still doing very formal work. Others are doing historically oriented work, and working with scientists to figure out what their methods are. The work I am doing on analogies, and on this, is in the middle. I see value in a middle ground, where you are abstracting a little bit from a particular topic in science. You are not just looking at what Darwin was doing in the 1840s, but you are looking at broader patterns. At the same time, you are not trying to give some logic of science.
The success of basic logic in mathematics was so great that it inspired people to say maybe we can give a logical approach.
I see what I am doing as a little tongue-in-cheek, trying to carry out some of this program while recognizing its limits. I have that background from my own history, starting out in mathematics, loving this formal approach, and trying to develop models that are clear and sharp.
Farha: Can describe what these models are.
Paul: The precautionary one is only a little bit different from the standard model, which I described earlier. You do a decision table where you put out utilities and probabilities and do the calculation.
The precautionary principle has been advanced as a kind of rival principle, but decision theory is quite mathematical. It is not very hard mathematics, at least. Economists use this, and once it seems reasonable, it gets absorbed by other disciplines.
The way I think of the precautionary principle is that instead of utilities being a single number, like 10 or 100 or negative 1,000, it is like a vector. It has two components. The first flags whether there is a catastrophe or not. That is the catastrophe dimension. The second is everything else, all the other values.
Here is a simple example to illustrate how this would go. Suppose you are trying to rent an apartment near UBC, and you absolutely value being within five kilometres of UBC above all else. It is locally a catastrophe, so to speak, if your apartment is more than five kilometres away. I am deliberately making it look bad for these vector utilities.
You would then say that anything over five kilometres is negative one. It is a disaster. Anything less than five kilometres is zero. It is fine. There is no catastrophe. The other vector component has all the other things: how nice the apartment is, how new it is, and so on.
When you make your decision and compare two apartments, you first look at the first component. Is it catastrophe or not? Only after that do you look at these other values. That is the idea of a lexical utility or vector utility. You first compare the first component, and only after you are sure the two things are equal do you look at the second component.
It turns out that if you do this with decisions involving catastrophe, you do not actually need a new precautionary principle. It is still maximizing expected utility, the same idea as standard decision theory, but using vector utilities instead of single-number utility. It is elegant. It is a generalization of normal decision theory.
It is math, and it does get harder once you get into trade-offs. We have one paper where we are trying to do this in as neutral a way as we can. So we are very liberal about what counts as catastrophe. We looked at a paper about climate change temperature targets for global temperature rise. We said that what somebody regards as catastrophic may depend on whether they are exporting fossil fuels. We have to allow that for some countries, sadly like Canada, which depend on this, certain outcomes might be considered an economic catastrophe. But against this there is the climate catastrophe.
This is one of those trade-off cases. In the paper, we argued that even fossil-fuel-exporting countries have reason to seek a pretty aggressive mitigation target. There is some math there because you are looking at graphs of probabilities of different catastrophes, and finding a point at which the marginal probabilities of the two catastrophes are equal. It is a little bit like economics. The basic concept is a slight generalization of normal decision theory.
Farha: That makes sense. Let's go back to COVID for a bit, because something in your paper was quite interesting. You looked at deaths from the illness versus deaths that did not necessarily have anything to do with the illness, but may have been caused by the lockdowns: people who were isolated, people who suffered tremendously in terms of mental health. Those could have ended up in statistics you would not expect when you are thinking about COVID. COVID is a classic example of something that is not so cut-and-dried when it comes to applying these models.
Paul: That is certainly true. As I said earlier, I am wary of claiming too much for this. It is still conceptual, but COVID is a great example. Even if you focus on mortality, this approach says let us not try to take everything into account. Let's not look at all quality of life.
In a cost-benefit analysis, if you are thinking of lockdown versus no lockdown, or how long the lockdown should last, if you use standard cost benefit analysis, you have to look at everything, like quality of life. Maybe that will strike people as exactly right. But this other approach, which focuses on precaution and catastrophe, says you actually zero in on something like mortality.
But not all the mortality is coming from COVID. Some of it is coming from people who do not get medical treatment, or people who commit suicide. Even if you just focus on mortality, there is mortality that is not directly attributable to COVID. It might even be attributable to the precaution, to the lockdown.
The fact is, you still need a tool that balances the prevention of these different catastrophes. That is what this theory does. I think it will give you a recommendation that it will be more aggressive about preventing mortality due to COVID than you would get from a standard cost-benefit approach, which would give a lot more weight to other factors, like quality of life.
You might disagree with the recommendation the principle gives, but it is at least an interesting alternative. It lets you get to a decision more quickly because you do not have to think about all costs and benefits. You are focusing on dimensions you have identified as the most important, in terms of what the catastrophe is.
Farha: It is a very interesting example, because even when we use the word lockdown, we can understand how that looked in British Columbia, under Bonnie Henry for example, but then you move to another place, go down to Argentina, which had one of the longest lockdowns ever, and you get a very strong variation in how a pandemic was understood by various governments all over the world, even inter-provincially. It interesting because it shows how difficult it is to understand even something like death numbers.
Paul: I immediately have to back off and say that I am humbled when I think about any actual example. There are a gazillion questions. When do you halt this? It is not just lockdown or no lockdown. It is a variable. How long is the lockdown? How severe is the lockdown? How do you justify this to people whose livelihoods are being harmed?
I fully admit that this is not something to be applied in a simplistic way. It is a concept that allows you to think in a slightly new way, but it needs a lot of work if it is ever going to be applied.
Farha: That is what you are onto, which is interesting, because even the words themselves, precautionary reasoning, are interesting philosophically.
You got me so interested that I said we would not talk about analogical reasoning, and then I changed my mind this morning when I started to read about analogical reasoning a little more. They are connected, but they are different. Do you feel that we can leave precautionary reasoning here, or is there something else you want to add?
Paul: I did not explain everything, but that is fine. The only thing I will add is that the latest work I am doing is trying to develop a series of more sophisticated models. You might start with COVID and this very simple model, lockdown or not, and then get one answer. Then you treat it as a variable and get a much more sophisticated answer.
Some problems will probably pop up in people's minds about how sensitive the decision is to where you arbitrarily draw the line and what counts as catastrophe. It turns out that if you go to a more sophisticated model, that disappears. It is cheap for me to say that because we do not have time, but that is what the paper I sent you is trying to argue.
Once you move to these more sophisticated models, you can answer worries about how sensitive the recommendation is to some arbitrary cut-off point. I am trying to develop a view that this is not the right way to think about the precautionary principle. It is not that there is some sharp cut-off where one more person dies and it is catastrophe, one fewer and it is not. That is totally wrong. Each one is a kind of little catastrophe. This is mainly just trying to head off obvious objections that anyone listening might think of. But I am happy to move on to analogical reasoning.
Farha: That is a big part of your life's work, and you are one of the leading experts in the world. Was I right when I introduced you like that?
Paul: What I will say is that lots of people have talked about models of belief revision, and what I call Bayesian epistemology and Bayesian philosophy of science, modelling the way science evolves in terms of probability and how we test. But not many people have thought about analogical reasoning. It is very hard. There are some people who have very good books on it, but not a lot.
It is used all over the place. We all use it in daily life. It is used in legal reasoning. When you appeal to a precedent, you are reasoning by analogy.
A simple example I like is life on other planets.
Farha: I interviewed back-to-back two astrophysicists from UBC, Jaymie Matthews and Michelle Kunimoto who study exoplanets.
Paul: That inference is becoming more interesting. Thomas Reid made this argument about life on Mars in the 18th century, or something like that. People have been interested in the idea that there is life on Earth, so could there be life elsewhere? In the absence of solid evidence, we use analogies. We give our best effort at explaining life on Earth, then we look for the factors that were there in that explanation and see whether they are present on Mars, or certainly exoplanets. That is an example.
I am fascinated with what I call remote targets. Sometimes there is an object we are trying to reason about that is inaccessible because it is gone. It is in the historical past, like an ancient civilization or ancient biology. It is remote historically. But it could also be remote in terms of distance. With black holes, there is a huge debate now about whether we can confirm some of our models of black holes using analogies.
It all started at UBC because of Bill Unruh. I do not know if you have interviewed him.
Farha: I should. Is he still around?
Paul: He is still around. He has this dumb hole idea. Think of a waterfall. Imagine that you put a radio speaker behind the waterfall, and now the water is flowing faster than the speed of sound. Now the sound cannot escape. It is a sonic analogue of a black hole.
Maybe we could study these sonic analogues and learn things about real black holes. That is very crudely explained, but the idea has been refined. They have more sophisticated analogues now that use Bose-Einstein condensates, quantum fluids, or whatever. They think they might have the same structure in some ways as a real black hole.
One of the predictions for a real black hole is Stephen Hawking's prediction of radiation. Very weak Hawking radiation is leaking from a black hole. Black holes actually evaporate over billions of years. It is not true that nothing escapes, because of quantum effects. There is some leakage, as it were.
Hawking's model had some mathematical weaknesses. Bill Unruh and a student of his made some amendments to the model. In the last few years, people have created tabletop analogues where they have observed analogues of Hawking radiation. So suddenly philosophy is important.
I talked to Unruh, and I do not think he would call this confirmation. He said it is a proof of concept. It shows that maybe there is something to this model, that it could work for this special case. But there are philosophers who say this actually provides evidence that if you were able to get close enough to a real black hole, you would observe this.
This is an example of a remote target. In an analogy, you have a source domain, which is the familiar, close object that you know something about. Then you have a target, which is the object you are trying to draw a conclusion about. If the target is not accessible any other way, sometimes analogy is the only way you can make inferences about it.
One last example of a remote target would be something ethically remote. Humans are, in some ways, a remote target for certain kinds of tests, like drug testing. Eventually they test drugs on real humans, but initially humans are off limits because of ethics. We need to use surrogates.
Farha: I like that word.
Paul: Surrogates?
Farha: Why not?
Paul: There is actually a concept in philosophy called surrogative reasoning.
Farha: Of course.
Paul: You use some other device to reason about something. It is more general than analogical reasoning. Basically, all modelling is surrogative reasoning. You are studying an animal model, or a bit of cell tissue, or maybe a computer model, to do analogical reasoning.
Remote targets are a fascinating case where analogical reasoning may be the best we have. So I am interested in that. Sometimes people wonder whether philosophy has any use, and I think it comes up in cases like these, at the margins of good science, where we are trying to extend our methods and we are not sure whether the standard approaches to evidence are valid. Sometimes scientists will talk to philosophers about this and ask, what do you think? I do not think they take it too seriously, but they are interested in hearing what philosophers think.
Farha: In the past, a lot of these disciplines were one. You could make an argument that the crossover was very much a thing. Your research in analogical reasoning touches physics, but it also touches things like archaeology, right?
Paul: Yes, archaeology. Absolutely.
Farha: Law, 100 percent. That makes sense. But the key word is also plausible. Is that right?
Paul: This is an excellent question. Analogical arguments vary in strength. When I wrote my book, about 15 years ago now, I decided I was going to focus on a very weak use of analogical arguments, which is what I called plausibility.
I was focusing on scientific uses of analogy, and I said that there is a stage of scientific reasoning where, before you put a lot of work into testing and setting up experiments, you need to be convinced that you have an idea that has a chance. It is plausible. That is where analogies come in.
The greatest example of this is Darwin. He has a book where he accumulates massive amounts of evidence for natural selection, but he starts by talking about artificial selection. There is scholarly debate about whether they are even different, but I think he was guided by a philosopher of science. I think it was especially Herschel, though Whewell is relevant here too. They were scientist-philosophers who articulated what is called the vera causa principl, which is any new theory in science, to be taken seriously, needs to be a true cause. It needs to exhibit analogy with something already known to operate causally in the world.
Darwin is showing that selection is a real force in the world. We know it is real because we see it at work with pigeon breeders. He was very influenced by philosophers of science. He was using this stage of plausibility reasoning.
In the book, I focused on analogies as a way of demonstrating plausibility. I also have a chapter on mathematical analogies, which are used all over the place in mathematics. You would not think so, because you have this lovely deductive reasoning, but no mathematician is going to spend hours and hours thinking about a hard problem unless they think, this is kind of like this other thing. Plausibility reasoning is very important in mathematics, and it is clearly not the same as a solid proof.
I was very interested in this, but that does not mean there are not really strong, compelling analogies. There are much stronger analogies, and I did not really focus on them. People have scolded me for this. The book does not really look at really powerful analogies.
A good example would be in physics, like scale models. If you want to test how an airplane might function, you might use a wind tunnel with a little scale model of the plane. If this is done right, you get really strong evidence for how the real plane is going to behave. I thought this was not too difficult, but I was told, no, you are completely wrong. This is really hard stuff. It is an interesting use of analogy, and it is left out of my book. I am coming back to that. I would like to look at really compelling analogies.
Farha: Now we have only a few minutes. I did mention AI, and you mentioned it in your email to me. What can you say in three minutes? Because that is exciting stuff.
Paul: It is hair-raising. It is asking questions that were not really asked before.
There is a long tradition, and part of the evidence that analogical reasoning is important is that machine learning people have been studying it since the 1980s: models of analogical reasoning, computational models. But this was the old-style AI, where you had symbolic manipulation and algorithms to do analogical reasoning.
As recently as 2020, I read a paper by Melanie Mitchell, who was Douglas Hofstadter's student. Hofstadter is an eminent computer scientist with philosophical interests. Mitchell said that none of these computational programs really work that well. They are all interesting, but we do not have anything.
Then, three or four years later, there were papers out about how well ChatGPT, and this is not even the latest version, does on analogical reasoning. They go through all these benchmark tests, with shapes and so on, and the AI outperforms college students. The last page of the article says this is an emergent ability, and we do not really know how it does it.
I went to a talk two weeks ago. One was about basic logic, and one was about my work on analogy. They were saying that AI does pretty well in analogical reasoning, but it does not use Paul Bartha's theory. I don’t think it is doing good analogical reasoning. It is using LLM methods, statistical prediction of the next token or whatever, and yet it does it really well.
So do we even need a normative philosophical theory? One answer is, you do not need them. You can throw them out the door. You can just start using these models and not worry about the philosophy anymore. In the book, I argued we need both. We should keep doing these computational models, but we should still think about what makes them work. I would like to think about this in the coming year.
Obviously, as a philosopher, I want to be honest. In the end, I may concede that maybe we do not need philosophy.
Farha: Did you just say that?
Paul: What did I say?
Farha: We do not need philosophy?
Paul: I may never be coming back again. But one cheap argument is that these models are not 100 percent reliable. If you were a lawyer relying on AI, you might get a very bad answer. But maybe we should set that aside, because they are pretty good.
Another argument I have seen is about these centaur systems. Have you heard this term? You pair a human expert with a program, and sometimes they can outperform the program. The human is fine-tuning. But I do not have a lot of faith in that answer either.
The third answer is that human flourishing is about reasoning and thinking hard and coming up with new ideas. I think analogies are used in that way too. For that, you need a kind of model that is not the way the AI thinks. It is a way a human could think.
The last answer is just me. I am curious. I want to understand why an analogy is good. Why is an analogical argument good? AI is never going to provide that. It may spit out arguments that we think are intuitively good, but what makes them good? For that, you need a philosophical theory. That is my quick answer.
Farha: That is a really great way to end the show. I love this conversation. I hope you enjoyed this.
Paul: I did. I hope it makes sense to people.
Farha: This is great. Ending with some question marks about AI is a really good way to end.
But it is exciting stuff, and I hope it keeps you motivated, because that is ultimately the goal of any philosopher. It is the why.
Paul: Asking questions.
Farha: It is the continuous questioning. I do love that Socrates, I love him.
Thank you, Paul, for coming in.
Paul: Thank you very much, Farha.
For readers interested in Dr. Paul Bartha’s work and the ideas discussed in this conversation:
Analogy and Analogical Reasoning — Stanford Encyclopedia of Philosophy
Hypersensitivity and the Lexical Precautionary Principle — Synthese