23 Aug 04
Last night I found a perfect strategy for winning at an idealized form of twenty questions, on which I may start a P&P thread later. However, human beings generally do not use perfect strategies, which is why artificial intelligence has little use for them. Instead, www.20q.net can play much as a person does--and very well. I was skeptical, but when it identified a honey badger in 17 questions I was impressed. Do check it out.
23 Aug 04
Originally posted by royalchicken
Last night I found a perfect strategy for winning at an idealized form of twenty questions, on which I may start a P&P thread later. However, human beings generally do not use perfect strategies, which is why artificial intelligence has little use for them. Instead, www.20q.net can play much as a person does--and very well. I was skeptical, but when it identified a honey badger in 17 questions I was impressed. Do check it out.
Ooops.... Took 29 to get to a Playstation2.
Way cool, thanks royalchicken.
23 Aug 04
Originally posted by royalchickenLet me guess: Your method involves repeated violations of the "use/mention" distinction.
You're quite welcome.
Actually, if the word in question is allowed to be any English word, then my strategy will take exactly 20 questions, or fewer with some corrections.
23 Aug 04
Originally posted by bbarrIt gave me mathematics in 29.
Apparently, it couldn't guess "mathematics" either.
There are many possibilities:
1. The initial questionnaire affects the questions asked (ie ''male, 16, USA'' lead to a set of questions that could identify maths in that instance while ''male, 28, USA'' did not)
2. You can't answer the questions in line with common perceptions of mathematics. In that case, a mathematician would be less likely to have maths identified than anyone else, except for the fact that the type of person who would visit 20q is probably on average going to have more correct prejudices about something like maths than the average person. The jury's out here.
3. Had it been allowed to go past 30, it would have given you mathematics in short order...
etc.
23 Aug 04
Originally posted by royalchickenIn other words, your method involves repeated violations of the "use/mention" distinction.
No; my method uses a search algorithm and is purely syntactic--I can just search the entire English language in twenty or fewer questions.
Suppose that I'm thinking of an aardvark. Now, what you are supposed to guess is that I'm thinking of an aardvark, not that there is a word 'aardvark' that English-speakers mention when they want to express something about aardvarks.
You, Sir, are a cheater.
Anyway, I bet I can beat your algorithm.