reinforcement

Terms from Artificial Intelligence: humans at the heart of algorithms

The glossary is being gradually proof checked, but currently has many typos and misspellings.

Reinforcement is the process of increasing or decreasing the weightings or parameters that gave rise to a positive or negative behaviour. The idea is based on Skinner's experiments with pigeons in the 1950s. If the mapping between stimulus, behaviour and reward is simple and direct, for example a look-up table with weights, then reinforcement is relatively straightforward. More commonly however, the causal relationship is complex and there are problems of credit assignment.

Used in Chap. 22: page 351