Successful strategies for prediction games have often historically involved
combinations of other strategies.
For example, the "iocane
powder" algorithm - which won the RSP tournament - combined the results of
multiple different prediction algorithms, and then chose the play of the
algorithm that had previously produced the best overall results. It contained
a random algorithm to act as a fall-back. This was an attempt at a defense
against any other program cracking its other algorithms.
In our matching pennies tournaments, similar aggregation techniques have been
used to combine the estimates of other algorithms - in the hope of taking
advantage of the principle of collective intelligence.
Another possibility is voting algorithms - which weigh the opinions of the
algorithms, according to how successful they have been in the past, and then
use a weighted combination of their choices to decide which move to make.
Minsky once proposed that minds worked like societies, with multiple different
sub-agents playing different roles.
One hope for the matching pennies tournaments is that they can be used to
explore the possible algorithms of collective intelligence - and help to
decide which ones are the most effective.
Tim Tyler |