He claims that not-3 implies that 1 is useless and 4 will only give incorrect results
By that, it seems like he is saying that being right is useless, and that the right answer yields incorrect results… that first bit would be rather horrible if true, and the last bit is just silly outside of the context of trying to nail down exactly how and why not-3 is true.
Could you rephrase ‘philosophically wrong’ (as your friend means it) more clearly? I don’t want to attack a straw man, but it seems like it could be going in a ridiculous direction.
I think he is trying to say that it’s more important to understand how humans currently perform cognitive tasks than to try to formulate a robust way to perform cognitive tasks that has mathematical guarantees about accuracy. This same friend has some vague, new-agey type beliefs about “emotions” and I think he takes the evidence about emotions being major players in decision making a little too far. He sees attempts to quantify what he calls “human decision making” with the same tools that we use for “computational decision making” as fruitless. For him, aesthetics, for example, cannot be analyzed down to the level of preference orderings, utility functions, and Bayesian decision rules. Aesthetics simply are what they are, directly encoded by emotions and not “understandable” in any symbolic sense.
I don’t think he would object to using Bayesian reasoning to make the best choices for, say, financial planning. But if you claimed that future neuroscientists will ever be able to quantitatively understand what “love” is (or even consciousness), he would reply that since human cognition is sub-conscious and non-Bayesian, you’ll never explain it in a way such that knowledge of Bayesian decision theory can help a human make “human decisions” any better than their emotions would by default.
I think the problem is that he makes a dichotomy between “human computation” and “machine computation” (owing maybe to the fact that his background is in philosophy and psychology, with a lot less emphasis on the math and physics relevant to cognition). Not only this, but he then further claims that neuroscience evidence in favor of emotions as major players in human cognition is also in favor of treating the two sides of his dichotomy with fundamentally different tools, and that Bayesian reasoning is not a successful normative model for human computation. To him, comparing the goodness of a Bayesian decision in place of a human-emotion decision is a “silly” thing to do.
Yes, I already try to adhere to that. The reason I came to LW to ask around was because I just wanted to make a succinct reply containing some relevant reading materials and a short summary of the error going on. Emotions do play a significant role in human cognition, and we are not by a long stretch good Bayesian reasoners. But there’s no special reason why we have to treat emotions and our current modes of cognition as if they are innately good or fundamentally not understandable. Some evidence supports this as mentioned in many of the comments above. I’m very grateful to have LW as a resource when it comes to cases like this. I think my own explanation of all of the above would probably have been mostly “correct” but horribly imprecise and nebulously flowing around lots of peripheral topics that won’t directly bear any fruit.
By that, it seems like he is saying that being right is useless, and that the right answer yields incorrect results… that first bit would be rather horrible if true, and the last bit is just silly outside of the context of trying to nail down exactly how and why not-3 is true.
Could you rephrase ‘philosophically wrong’ (as your friend means it) more clearly? I don’t want to attack a straw man, but it seems like it could be going in a ridiculous direction.
I think he is trying to say that it’s more important to understand how humans currently perform cognitive tasks than to try to formulate a robust way to perform cognitive tasks that has mathematical guarantees about accuracy. This same friend has some vague, new-agey type beliefs about “emotions” and I think he takes the evidence about emotions being major players in decision making a little too far. He sees attempts to quantify what he calls “human decision making” with the same tools that we use for “computational decision making” as fruitless. For him, aesthetics, for example, cannot be analyzed down to the level of preference orderings, utility functions, and Bayesian decision rules. Aesthetics simply are what they are, directly encoded by emotions and not “understandable” in any symbolic sense.
I don’t think he would object to using Bayesian reasoning to make the best choices for, say, financial planning. But if you claimed that future neuroscientists will ever be able to quantitatively understand what “love” is (or even consciousness), he would reply that since human cognition is sub-conscious and non-Bayesian, you’ll never explain it in a way such that knowledge of Bayesian decision theory can help a human make “human decisions” any better than their emotions would by default.
I think the problem is that he makes a dichotomy between “human computation” and “machine computation” (owing maybe to the fact that his background is in philosophy and psychology, with a lot less emphasis on the math and physics relevant to cognition). Not only this, but he then further claims that neuroscience evidence in favor of emotions as major players in human cognition is also in favor of treating the two sides of his dichotomy with fundamentally different tools, and that Bayesian reasoning is not a successful normative model for human computation. To him, comparing the goodness of a Bayesian decision in place of a human-emotion decision is a “silly” thing to do.
I predict that the lesson behind this exchange will turn out to be “Don’t argue with people who think consciousness is fundamental”.
Yes, I already try to adhere to that. The reason I came to LW to ask around was because I just wanted to make a succinct reply containing some relevant reading materials and a short summary of the error going on. Emotions do play a significant role in human cognition, and we are not by a long stretch good Bayesian reasoners. But there’s no special reason why we have to treat emotions and our current modes of cognition as if they are innately good or fundamentally not understandable. Some evidence supports this as mentioned in many of the comments above. I’m very grateful to have LW as a resource when it comes to cases like this. I think my own explanation of all of the above would probably have been mostly “correct” but horribly imprecise and nebulously flowing around lots of peripheral topics that won’t directly bear any fruit.