1. Software agents are invading online poker sites and relieving the humans of their money.
  2. The New York Times web site will beat you at rock scissors paper because you are a predictable human. (mm:  Courtney Conklin Knapp)
  3. Tyler Cowen and his computer make each other better at chess.

Following up on Tyler, the suggestion is that there are gains from specialization in computer/human partnerships.  But it is not enough for Tyler and his computer to beat a computer.  Could Tyler and another human player (of strength comparable to his computer partner) do even better?

Now it is interesting to observe that the other comparison is not possible. Would a team of two computers (with strengths comparable to Tyler and his machine) do even better? How would two computers make a team?  If the two computers came up with different ideas how would they decide which one was better?

010101:  I think we should play Re1. I rate it +.30
1110011: I considered that move and at 22 ply I rate it at +0.05, instead I suggest we sac the Knight.
010101: I considered that move and at 22 ply I rate it at -1.8.
1110011: Here take a look at my analysis.
010101:  Yes I am aware of that sequence of moves, I already considered it.  It’s worth +0.05.
1110011: No, +0.30
010101:  No, +0.05

etc.  Any protocol for deciding which is the right analysis should already have been programmed into the original software.  Put differently, if there was a way to map the pair of evaluations (.3,.05) into a better evaluation y, then the position should already have been evaluated at y by each machine individually.

The only benefit of the two computers would be the deeper search in the same amount time.  That is, a two computer team is just two parallel processers but exactly the same evaluation heuristic applied to the final position searched. In that sense the human’s unique ability is to understand when to switch heuristics.  (But why can’t this understanding be programmed into software?)

About these ads