I need to build an artificial intelligence capable of solving some simple Raven's Progressive Matrices. The gist of it is that you are given an analogy "A is to B as C is to blank" and numerous answer choices.
A -> B
C -> ?
My agent can, at the moment return a list of probabilities for each answer choice in this scenario. However, the actual Raven's Matrices problems are two-dimensional. The real relationship looks something like:
A -> B
| |
v v
C -> ?
It's also possible for there to be a diagonal relationship. My agent can use the same process to find probabilities for each solution based on the vertical relationship.
I'm not sure how to combine these probabilities. It seems to me that if both relationships imply answer (1) with 75% probability, then (1) should now have a larger probability. But, all that really occurs to me here is to average the two.