= 27d567bc2d48d9f40e9ccc20ea370b15)
Wei_Dai comments on 'Is' and 'Ought' and Rationality - Less Wrong
You are viewing a comment permalink. View the original post to see all comments and the full post content.
= 783df68a0f980790206b9ea87794c5b6)
Loading…= b715d02e85f7b3b4ae65ca3d3e450868)
Subscribe to RSS Feed
Powered by Reddit
= f037147d6e6c911a85753b9abdedda8d)
You are viewing a comment permalink. View the original post to see all comments and the full post content.
Comments (70)
Suppose I hear Bob say "I want to eat an apple." Am I justified in assigning a higher probability to "Bob wants to eat an apple" after I hear this than before (assuming I don't have some other evidence to the contrary, like someone is holding a gun to Bob's head)? Assuming the answer is yes, and given that "Bob said 'I want to eat an apple.'" is a naturalistic fact, how did I learn something non-naturalistic from that?
I think this question hits the nail on the head. You are justified in assigning a higher probability to "Bob wants to eat an apple" just in case you are already justified in taking Bob to be a rational agent (other things being equal...). If Bob isn't at least minimally rational, you can't even get so far as construing his words as English, let alone to trust that his intent in uttering them is to convey that he wants to eat an apple (think about assessing a wannabe AI chatbot, here). But in taking Bob to be rational, you are already taking him to have preferences and beliefs, and for there to be things which he ought or ought not to do. In other words, you have already crossed beyond what mere natural science provides for. This, anyway, is what I'm trying to argue.
I think I kind of see what you're getting at. In order to recognize that Bob is rational, I have to have some way of knowing the properties of rationality, and the way we learn such properties does not seem to resemble the methods of the empirical sciences, like physics or chemistry.
But to me it does seems to bear some resemblance to the methods of mathematics. For example in number theory we try to capture some of our intuitions about "natural numbers" in a set of axioms, which then allows us to derive other properties of natural numbers. In the study of rationality we have for example Von Neumann–Morgenstern axioms. Although there is much more disagreement about what an appropriate set of axioms might be where rationality is concerned, the basic methodology still seems similar. Do you agree?
I appreciate this clarification. The point is meant to be about, as you say, the empirical sciences. I agree that there can be arbitrarily sophisticated scientific theories concerning rational behaviour - just that these theories aren't straight-forwardly continuous with the theories of natural science.
In case you haven't encountered it and might be interested, the underdetermination problem associated with inferring to beliefs and desires from mere behaviour has been considered in some depth by Donald Davidson, eg in his Essays on Actions and Events
In your OP, you wrote that you found statements like
implausible.
It seems quite possible to me that, perhaps through some method other than the methods of the empirical sciences (for example, through philosophical inquiry), we can determine that among the properties of "want" is that "want to eat an apple" correctly describes the brain state ABC or computational state DEF (or something of that nature). Do you still consider that statement implausible?
This seems reasonable, but I have to ask about "correctly describes". The statement
"want to eat an apple" implies being in brain state ABC or computational state DEF (or something of that nature)
is plausible to me. I think the reverse implication, though raises a problem:
being in brain state ABC or computational state DEF (or something of that nature) implies "want to eat an apple"
But maybe neither of these is what you have in mind?
I think I mean the latter. What problem do you see with it?
I don't see the problem with the latter either.
I do accept that 'wants' imply 'oughts'. It's an oversimplification, but the thought is that statements such as
are intuitively plausible. If wanting carries no implications for what one ought to do, I don't see how motivation can get off the ground.
Now, if we have
1) wanting that P implies one ought to do Q,
and
2) being in physical state ABC implies wanting that P
then, by transitivity of implication, we get
3) being in physical state ABC implies one ought to do Q
And this is just the kind of implication I'm trying to show is problematic.
Would it be fair to say that your position is that there could be two physically identical brains, and one of them wants to eat an apple but the other doesn't, or perhaps that one of them is rational but the other isn't. In other words that preference-zombies or rationality-zombies could exist?
(In case it's not clear why I'm saying this, this is what accepting
while denying
would imply.)
I think your question again gets right to the nub of the matter. I have no snappy answer to the challenge -here is my long-winded response.
The zombie analogy is a good one. I understand it's meant just as an analogy -the intent is not to fall into the qualia quagmire. The thought is that from a purely naturalistic perspective, people can only properly be seen as, as you put it, preference- or rationality-zombies.
The issue here is the validity of identity claims of the form,
My answer is to compare them to the fate of identity claims relating to sensations (qualia again), such as
Suppose being in pain is found empirically always to correlate to being in brain state DEF, and the identity is proposed. Qualiaphiles will object, saying that this identity misses what's crucial to pain, viz, how it feels. The qualiaphile's thought can be defended by considering the logic of identity claims generally (this adapted from Saul Kripke's Naming and Necessity).
Scientific identity claims are necessary - if water = H2O in this world, then water = H2O in all possible worlds. That is, because water is a natural kind, whatever it is, it couldn't have been anything else. It is possible for water to present itself to us in a different phenomenal aspect ('ice9'!), but this is OK because what's essential to water is its underlying structure, not its phenomenal properties. The situation is different for pain - what's essential to pain is its phenomenal properties. Because pain essentially feels like this (so the story goes), it's correlation with being in brain state DEF can only be contingent. Since identities of this kind, if true, are by their natures necessary, the identity is false.
There is a further step (lots of steps, I admit) to rationality. The thought is that our access to people's rationality is 'direct' in the way our access to pain is. The unmediated judgement of rationality would, if push were to come to shove, trump the scientifically informed, indirect inference from brain states. Defending this proposition would take some doing, but the idea is that we need to understand each other as rational agents before we can get as far as dissecting ourselves to understand ourselves as mere objects.
There are underdetermination problems all over the philosophy of science. I don't see how this poses a special problem for norms, or rationality. When two domains of science are integrated, it is often via proposed bridge laws that may not provide an exactly intuitive match. For example, some of the gases that have a high "temperature" when that is defined as mean kinetic energy, might feel somewhat colder than some others with lower "temperature". But we accept the reduction provided it succeeds well enough.
If there are no perfect conceptual matches by definitions of a to-be-reduced term in the terms of the reducing domain, that is not fatal. If we can't find one now, that is even less so.
I agree that underdetermination problems are distinct from problems about norms -from the is/ought problem. Apologies if I introduced confusion in mentioning them. They are relevant because they arise (roughly speaking) at the interface between decision theory and empirical science, ie, where you try to map mere behaviours onto desires and beliefs.
My understanding is that in philosophy of science, an underdetermination problem arises when all evidence is consistent with more than one theory or explanation. You have a scientist, a set of facts, and more than one theory which the scientist can fit to the facts. In answer to your initial challenge, the problem is different for human psychology because the underdetermination is not of the scientist's theory but supposedly of one set of facts (facts about beliefs and desires) by another (behaviour and all cognitive states of the agent). That is, in contrast to the basic case, here you have a scientist, one set of facts -about a person's behaviour and cognitive states- a second set of suppposed facts -about the person's beliefs and desires- and the problem is that the former set underdetermine the latter.
You seem to be introducing a fact/theory dichotomy. That doesn't seem promising.
If we look at successful reductions in the sciences, they can make at least some of our underdetermination problems disappear. Mean kinetic energy of a gas is a more precise notion than "temperature" was in prior centuries, for example. I wouldn't try to wrestle with the concepts of cognitive states and behaviors to resolve underdetermination. Instead, it seems worthwhile to propose candidate bridge laws and see where they get us. I think that Millikan et al may well be onto something.
As I understand it, the problem of scientific underdetermination can only be formulated if we make some kind of fact/theory distinction - observation/theory would be better, is that ok with you?
I'm not actually seeing how the temperature example is an instance of underdetermination, and I'm a little fuzzy on where bridge laws fit in, but would be open to clarification on these things.
Well, scientific underdetermination problems can be formulated with a context-relative observation/theory distinction. But this is compatible with seeing observations as potentially open to contention between different theories (and in that sense "theory-laden"). The question is, are these distinctions robust enough to support your argument?
By the way, I'm confused by your use of "cognitive states" in your 08 July comment above, where it is contrasted to beliefs and desires. Did you mean neural states?
Temperature was underdetermined in the early stages of science because the nebulae of associated phenomena had not been sorted out. Sometimes different methods of assessing temperature could conflict. E.g., object A might be hotter to the touch than B, yet when both are placed in contact with C, B warms C and A cools it.
You are quite right -sorry about the confusion. I meant to say behaviour and computational states -the thought being that we are trying to correlate having a belief or desire to being in some combination of these.
I understand you're referring here to the claim -for which I can't take credit- that facts about behaviour underdetermine facts about beliefs and desires. Because the issue -or so I want to argue- is of underdetermination of one set of potential facts (namely, about beliefs and desires) by another (uninterpreted behaviour), rather than of underdetermination of scientific theory by fact or observation, I'm not seeing that the issue of the theory-ladenness of observation ultimately presents a problem.
The underdetermination is pretty easy to show, at least on a superficial level. Suppose you observe
You infer:
But couldn't one also infer,
or
You may think that if you observe enough behaviour, you can constrain these possibilities. There are arguments (which I acknowledge I have not given), which show (or so a lot of people think) that this is not the case - the underdetermination keeps pace.