Tversky and Kahneman believe in such ideas as “evolved mental behaviour” and “bounded rationality”. These beliefs exist right enough. If you read The Beginning of Infinity by David Deutsch you will see arguments against these sort of things.
That sounds pretty awful. Bounded rationality is a standard concept. Surely if you argue against it, you are confused, or don’t understand it properly. I’m not sure what an “evolved mental behaviour” is, but that sounds pretty uncontroversial too. Looking at Deutsch on video about 28:00 in he is using the term “bounded rationality” to refer to something different—so this seems like a simple confusion based on different definitions of terms.
If you are going to assume that people are confused for arguing against “standard concepts” or because you think something is uncontroversial, then that is just argument from authority.
The supposed heuristics which Herbert Simon and others propose which give rise to our alledged cognitive biases are held by them to have evolved via biological evolution, to be based on induction, and to be bounded. Hard-coded processes based on induction that can generate some knowledge but not all knowledge goes against the ideas that Deutsch discusses in The Beginning of Infinity. For one thing, induction is impossible and doesn’t happen anywhere including in human brains. For another, knowledge creation is all or nothing; a machine that can generate some knowledge can generate all knowledge (the jump to universality) - halfway houses like these heuristics would be very difficult to engineer, they would keep jumping. And, for another, human knowledge and reasoning is memetic, not genetic, and there are no hard-coded reasoning rules.
This is just an argument over the definition of the phrase “bounded rationality”. Let’s call these two definitions BR1 and BR2. The definition that timtyler, Kahnemann and Tversky, and I are using is BR1; the definition that you, curi, and David Deutsch use is BR2.
BR1 means “rationality that is performed using a finite amount of resources”. Think of this as bounded-resource rationality. All rationality done in this universe is BR1, by definition, because you only get a limited amount of time and memory to think about things. This definition does not contain any claims about what sort of knowledge BR1 can or can’t generate. A detailed theory of BR1 would say things like “solving this math problem requires at least 10^9 operations”. More commonly, people refer to BR1 to distinguish it from things like AIXI, which is a mathematical construct that can theoretically figure out anything given sufficient data, but which is impossible to construct because it contains several infinities. A mind with universal reasoning is BR1 if it only has a finite amount of time to do it in.
BR2 means “rationality that can generate some types of knowledge, but not others”. Think of this as bounded-domain rationality. Whether this exists at all depends on what you mean by “knowledge”. For example, if you have a computer program that collects seismograph data and predicts earthquakes, you might say it “knows” where earthquakes will occur; this would make it a BR2. If you say that this sort of thing doesn’t count as knowledge until a human reads it from a screen or printout, then no BR2s exist.
BR1 is a standard concept, but as far as I know BR2 is unique to Deutsch’s book Beginning of Infinity. BR1 exists, tautologically from its definition. Whether BR2 exists or not depends on how you define some other things, but personally I don’t find BR2 illuminating so I see no reason to take a stance either way on it.
I’m pretty sure there’s a similar issue with the definition of “induction”. I know of at least two definitions relevant to epistemology, but neither of them seems to make sense in context so I suspect that Deutsch has come up with a third. Could you explain what Deutsch uses the word induction to mean? I think that would clear up a great deal of confusion.
All your points are wrong, though. Induction has been discussed to death already. Computation universality doesn’t mean intelligent systems evolve without cognitive biases, and the fact that human cultural knowledge is memetic doesn’t mean there are not common built-in biases either. The human brain is reprogrammable to some extent, but much of the basic pattern-recognition circuitry has a genetically specified architecture.
Many of the biases in question are in the basic psychology textbooks—this is surely not something that is up for debate.
The conjunction fallacy: explanations of the linda problem by the theory of
hints
Empirical research has shown that in some situations, subjects tend to assign
a probability to a conjunction of two events that is larger than the probability
they assign to each of these two events. This empirical phenomenon is
traditionally called the conjunction fallacy. One of the best-known experiments
used to demonstrate the conjunction fallacy is the Linda problem introduced by
Tversky and Kahneman in 1982. They explain the “fallacious behavior” by their
so-called judgemental heuristics. These heuristics have been criticized heavily
as being far “too vague to count as explanations”. In this article, it is shown
that the “fallacious behavior” in the Linda problem can be explained by the
so-called theory of hints.
Why do you argue from authority saying things like something surely cannot be up for debate because it’s in all the textbooks? curi and I are fallibilists: nothing is beyond question.
You say you’re a fallibist, but you’re actually falling into the failure mode described in this article. Suppose you’ve got a question with positions A and B, with a a bunch of supporting arguments for A, and a bunch of supporting arguments for B. Some of those arguments for each side will be wrong, or ambiguous, or inapplicable—that’s what fallibilism predicts and I think we all agree with that.
Suppose there are 3 valid and 3 invalid arguments for A, and 3 valid and 3 invalid arguments for B. Now suppose someone decides to get rid any of the arguments that are invalid, but they happen to think A is better. Most people will end up attacking all the arguments for B, but they won’t look as closely at the arguments for A. After they’re finished, they’ll have 3 valid and 3 invalid arguments for A, and 3 valid arguments for B—which looks like a preponderance of evidence in favor of B, but it isn’t.
Now read the abstract of that paper you linked again. That paper disagrees with where K&T draw the boundary between questions that trigger the conjunction fallacy and questions that don’t, and describe the underlying mechanism that produces it differently. They do not claim that the conjunction fallacy doesn’t exist.
Empirical research has shown that in some situations, subjects tend to assign a probability to a conjunction of two events that is larger than the probability they assign to each of these two events. This empirical phenomenon is traditionally called the conjunction fallacy. One of the best-known experiments used to demonstrate the conjunction fallacy is the Linda problem introduced by Tversky and Kahneman in 1982. They explain the “fallacious behavior” by their so-called judgemental heuristics. These heuristics have been criticized heavily as being far “too vague to count as explanations”. In this article, it is shown that the “fallacious behavior” in the Linda problem can be explained by the so-called theory of hints. source
It seems as though they acknowledge the conjunction fallacy and are proposing different underlying mechanisms to explain how it is produced.
Why do you argue from authority saying things like something surely cannot be up for debate because it’s in all the textbooks? curi and I are fallibilists: nothing is beyond question.
If you want to argue with psychology 101, fine, but do it in public, without experimental support, and a dodgy theoretical framework derived from computation universality and things are not going to go well.
If citing textbooks is classed as “arguing from authority”, one should point out that such arguments are usually correct.
They have put fallacious behaviour in quotes to indicate that they don’t agree the fallacy exists. I could be wrong, however, as I am just going from the abstract and maybe the authors do claim it exists. However they seem to be saying it is just an artifact of hints. I’ll need to read the paper to understand better. Maybe I’ll end up disagreeing with the authors.
Textbook arguments are often wrong. Consider quantum physics and the Copenhagen Interpretation for example. And one way of arguing against CI is from a philosophical perspective (it’s instrumentalist and a bad explanation).
I looked through the whole paper and don’t think you’re wrong.
I don’t agree with the hints paper in various respects. But it disagrees with the conjunction fallacy and argues that conjunction isn’t the real issue and the biases explanation isn’t right either. So certainly there is disagreement on these issues.
If citing textbooks is classed as “arguing from authority”, one should point out that such arguments are usually correct.
Do you mean in the context of arguments in textbooks? This seems like a very weak claim, given how frequently some areas change. Indeed, psychology is an area where what an intro level textbook would both claim to be true and would even discuss as relevant topics has changed drastically in the last 60 years. For example, in a modern psychology textbook the primary discussion of Freud will be to note that most of his claims fell into two broad categories:untestable or demonstrably false. Similarly, even experimentally derived claims about some things (such as how children learn) has changed a lot in the last few years as more clever experimental design has done a better job separating issues of planning and physical coordination from babies’ models of reality. Psychology seems to be a bad area to make this sort of argument.
Do you mean in the context of arguments in textbooks?
Yes.
This seems like a very weak claim, given how frequently some areas change.
It is weak, in that it makes no bold claims, and merely states what most would take for granted—that most of the things in textbooks are essentially correct.
That sounds pretty awful. Bounded rationality is a standard concept. Surely if you argue against it, you are confused, or don’t understand it properly. I’m not sure what an “evolved mental behaviour” is, but that sounds pretty uncontroversial too. Looking at Deutsch on video about 28:00 in he is using the term “bounded rationality” to refer to something different—so this seems like a simple confusion based on different definitions of terms.
If you are going to assume that people are confused for arguing against “standard concepts” or because you think something is uncontroversial, then that is just argument from authority.
The supposed heuristics which Herbert Simon and others propose which give rise to our alledged cognitive biases are held by them to have evolved via biological evolution, to be based on induction, and to be bounded. Hard-coded processes based on induction that can generate some knowledge but not all knowledge goes against the ideas that Deutsch discusses in The Beginning of Infinity. For one thing, induction is impossible and doesn’t happen anywhere including in human brains. For another, knowledge creation is all or nothing; a machine that can generate some knowledge can generate all knowledge (the jump to universality) - halfway houses like these heuristics would be very difficult to engineer, they would keep jumping. And, for another, human knowledge and reasoning is memetic, not genetic, and there are no hard-coded reasoning rules.
This is just an argument over the definition of the phrase “bounded rationality”. Let’s call these two definitions BR1 and BR2. The definition that timtyler, Kahnemann and Tversky, and I are using is BR1; the definition that you, curi, and David Deutsch use is BR2.
BR1 means “rationality that is performed using a finite amount of resources”. Think of this as bounded-resource rationality. All rationality done in this universe is BR1, by definition, because you only get a limited amount of time and memory to think about things. This definition does not contain any claims about what sort of knowledge BR1 can or can’t generate. A detailed theory of BR1 would say things like “solving this math problem requires at least 10^9 operations”. More commonly, people refer to BR1 to distinguish it from things like AIXI, which is a mathematical construct that can theoretically figure out anything given sufficient data, but which is impossible to construct because it contains several infinities. A mind with universal reasoning is BR1 if it only has a finite amount of time to do it in.
BR2 means “rationality that can generate some types of knowledge, but not others”. Think of this as bounded-domain rationality. Whether this exists at all depends on what you mean by “knowledge”. For example, if you have a computer program that collects seismograph data and predicts earthquakes, you might say it “knows” where earthquakes will occur; this would make it a BR2. If you say that this sort of thing doesn’t count as knowledge until a human reads it from a screen or printout, then no BR2s exist.
BR1 is a standard concept, but as far as I know BR2 is unique to Deutsch’s book Beginning of Infinity. BR1 exists, tautologically from its definition. Whether BR2 exists or not depends on how you define some other things, but personally I don’t find BR2 illuminating so I see no reason to take a stance either way on it.
I’m pretty sure there’s a similar issue with the definition of “induction”. I know of at least two definitions relevant to epistemology, but neither of them seems to make sense in context so I suspect that Deutsch has come up with a third. Could you explain what Deutsch uses the word induction to mean? I think that would clear up a great deal of confusion.
All your points are wrong, though. Induction has been discussed to death already. Computation universality doesn’t mean intelligent systems evolve without cognitive biases, and the fact that human cultural knowledge is memetic doesn’t mean there are not common built-in biases either. The human brain is reprogrammable to some extent, but much of the basic pattern-recognition circuitry has a genetically specified architecture.
Many of the biases in question are in the basic psychology textbooks—this is surely not something that is up for debate.
Looks to me that those biases are very much up for debate and not just by curi and myself:
Why do you argue from authority saying things like something surely cannot be up for debate because it’s in all the textbooks? curi and I are fallibilists: nothing is beyond question.
You say you’re a fallibist, but you’re actually falling into the failure mode described in this article. Suppose you’ve got a question with positions A and B, with a a bunch of supporting arguments for A, and a bunch of supporting arguments for B. Some of those arguments for each side will be wrong, or ambiguous, or inapplicable—that’s what fallibilism predicts and I think we all agree with that.
Suppose there are 3 valid and 3 invalid arguments for A, and 3 valid and 3 invalid arguments for B. Now suppose someone decides to get rid any of the arguments that are invalid, but they happen to think A is better. Most people will end up attacking all the arguments for B, but they won’t look as closely at the arguments for A. After they’re finished, they’ll have 3 valid and 3 invalid arguments for A, and 3 valid arguments for B—which looks like a preponderance of evidence in favor of B, but it isn’t.
Now read the abstract of that paper you linked again. That paper disagrees with where K&T draw the boundary between questions that trigger the conjunction fallacy and questions that don’t, and describe the underlying mechanism that produces it differently. They do not claim that the conjunction fallacy doesn’t exist.
It seems as though they acknowledge the conjunction fallacy and are proposing different underlying mechanisms to explain how it is produced.
If you want to argue with psychology 101, fine, but do it in public, without experimental support, and a dodgy theoretical framework derived from computation universality and things are not going to go well.
If citing textbooks is classed as “arguing from authority”, one should point out that such arguments are usually correct.
They have put fallacious behaviour in quotes to indicate that they don’t agree the fallacy exists. I could be wrong, however, as I am just going from the abstract and maybe the authors do claim it exists. However they seem to be saying it is just an artifact of hints. I’ll need to read the paper to understand better. Maybe I’ll end up disagreeing with the authors.
Textbook arguments are often wrong. Consider quantum physics and the Copenhagen Interpretation for example. And one way of arguing against CI is from a philosophical perspective (it’s instrumentalist and a bad explanation).
I looked through the whole paper and don’t think you’re wrong.
I don’t agree with the hints paper in various respects. But it disagrees with the conjunction fallacy and argues that conjunction isn’t the real issue and the biases explanation isn’t right either. So certainly there is disagreement on these issues.
Do you mean in the context of arguments in textbooks? This seems like a very weak claim, given how frequently some areas change. Indeed, psychology is an area where what an intro level textbook would both claim to be true and would even discuss as relevant topics has changed drastically in the last 60 years. For example, in a modern psychology textbook the primary discussion of Freud will be to note that most of his claims fell into two broad categories:untestable or demonstrably false. Similarly, even experimentally derived claims about some things (such as how children learn) has changed a lot in the last few years as more clever experimental design has done a better job separating issues of planning and physical coordination from babies’ models of reality. Psychology seems to be a bad area to make this sort of argument.
Yes.
It is weak, in that it makes no bold claims, and merely states what most would take for granted—that most of the things in textbooks are essentially correct.
Nice post.