I find that all questions fall into one of 3 categories:
Well defined: These questions are clear, and contain all the basic information you need to answer them, with little or no need to infer what the questioner meant. Word problems are a good example, and so is someone asking for directions or asking what you would like for dinner.
Poorly defined: These are problems that you don’t know how to solve, at least at first. Maybe you have to learn what the questioner means, or maybe you have to acquire some fundamental understanding in order to evaluate several possible methods of coming to a solution. Most engineering R&D problems fit in this box, but other examples include the chicken/egg problem and asking whether a tree falling in a forest makes a sound.
What is the square root of a potato: At least that’s how I’ve always described such questions to people. These are questions that are so poorly defined that they are demonstrably completely incoherent. There is no possible way of interpreting these questions which could possibly have an answer. It’s when a question is “not even wrong”. This seems quite similar to what Yudkowsky is describing here, although maybe he is referring to things somewhere between #2 and #3. I guess you could create a spectrum running from #1 to #3 type questions, but I’ve found that problems tend to fall pretty neatly into these 3 categories, so I haven’t found it useful to think of them as lying along a spectrum.
I find that all questions fall into one of 3 categories:
Well defined: These questions are clear, and contain all the basic information you need to answer them, with little or no need to infer what the questioner meant. Word problems are a good example, and so is someone asking for directions or asking what you would like for dinner.
Poorly defined: These are problems that you don’t know how to solve, at least at first. Maybe you have to learn what the questioner means, or maybe you have to acquire some fundamental understanding in order to evaluate several possible methods of coming to a solution. Most engineering R&D problems fit in this box, but other examples include the chicken/egg problem and asking whether a tree falling in a forest makes a sound.
What is the square root of a potato: At least that’s how I’ve always described such questions to people. These are questions that are so poorly defined that they are demonstrably completely incoherent. There is no possible way of interpreting these questions which could possibly have an answer. It’s when a question is “not even wrong”. This seems quite similar to what Yudkowsky is describing here, although maybe he is referring to things somewhere between #2 and #3. I guess you could create a spectrum running from #1 to #3 type questions, but I’ve found that problems tend to fall pretty neatly into these 3 categories, so I haven’t found it useful to think of them as lying along a spectrum.