Some experiments with giving a computer program the ability to learn to play a simple game by asking advice of a human teacher
A model of learning in a simple game playing context is simulated with the aim of gaining insight into a pattern learning process rather than the subtleties of the particular game. The simulation consists of a model of a learning player able to ask help of a (human) teacher when unable to make its own decisions on the basis of previously learned knowledge. An opponent is provided by either a human player or another example of a simulated player. To begin with, the learning player is new to the game and asks many questions of its teacher. As its playing experience increases, so it relies less on external advice, making more of its own decisions based on advice given in similar game situations encountered previously. The system can be set up to allow fully automated trials between simulated players with varying learning experience in order to compare their performance. The difficulty of conveying to the learning model the intention of the teacher's advice, even in the context of a simple game, highlights more general problems in Artificial Intelligence. The problem is a fundamental difference in the mechanisms of human and computer activity. Having done this work it is possible to conclude that the world of a computer, and the natural world, impose such different constraints that they encourage the development of very different mechanisms.