Samuel's Checkers Program
Arthur Samuel wrote an early (1959) program that played
Checkers at championship level. This
program had several interesting
- The program learned its heuristic function from experience playing
humans and itself. The heuristic function was a weighted average
of several features (e.g. piece advantage,
mobility, control of the center).
Coefficients of features were ``rewarded'' or ``punished'' depending on
whether the program won or lost.
- If the program wins (or loses), it is difficult to determine which
move(s) were responsible and should be rewarded or punished. This is
called the credit assignment problem .
- Low-pass filtering of coefficients was used to smooth out the
effects of unjust rewards and punishments.