Zendo, the game of inductive logic has been discussed many times on Less Wrong. To make things easier for new players, I made a web application that generates Koans of several difficulty levels. You can find it here.
It would also be cool to write a program to play Zendo, as master. Perhaps using character-strings rather than Icehouse pieces. That would probably train induction faster than playing against humans, since there’s less social overhead. Though I’m doubtful that that training would generalize to things that aren’t Zendo. If one did this, it would probably also be best to get a larger set of possible predicates, and maybe construct rules recursively as logic sentences, so it’s harder to just learn the program’s rule-distribution.
How do you rank the difficulty of Koans? My intuition does a very good job by now, but caching it out has always resulted in obviously wrong corner cases.
“A Koan has the Buddha-nature if and only if all the pieces are ungrounded, except for blue pieces.” is unclear to me. I am not sure whether blue pieces must be grounded or may be grounded.
How do you rank the difficulty of Koans? My intuition does a very good job by now, but caching it out has always resulted in obviously wrong corner cases.
The method I used is very close to this one, but what you say is true. There are edge-cases that aren’t quite right. For instance, the most difficult “medium” Koans seem more difficult than the least difficult “hard” Koans.
“A Koan has the Buddha-nature if and only if all the pieces are ungrounded, except for blue pieces.” is unclear to me. I am not sure whether blue pieces must be grounded or may be grounded.
Would “A Koan has the Buddha-nature if and only if all the non-blue pieces are ungrounded, and all the blue pieces are grounded” be less ambiguous?
Zendo, the game of inductive logic has been discussed many times on Less Wrong. To make things easier for new players, I made a web application that generates Koans of several difficulty levels. You can find it here.
Nice!
It would also be cool to write a program to play Zendo, as master. Perhaps using character-strings rather than Icehouse pieces. That would probably train induction faster than playing against humans, since there’s less social overhead. Though I’m doubtful that that training would generalize to things that aren’t Zendo. If one did this, it would probably also be best to get a larger set of possible predicates, and maybe construct rules recursively as logic sentences, so it’s harder to just learn the program’s rule-distribution.
How do you rank the difficulty of Koans? My intuition does a very good job by now, but caching it out has always resulted in obviously wrong corner cases.
“A Koan has the Buddha-nature if and only if all the pieces are ungrounded, except for blue pieces.” is unclear to me. I am not sure whether blue pieces must be grounded or may be grounded.
Nice job!
The method I used is very close to this one, but what you say is true. There are edge-cases that aren’t quite right. For instance, the most difficult “medium” Koans seem more difficult than the least difficult “hard” Koans.
Would “A Koan has the Buddha-nature if and only if all the non-blue pieces are ungrounded, and all the blue pieces are grounded” be less ambiguous?
Thanks.