The Nature of Heuristics II : Background and Examples

Machine learning can be categorized along many dimensions, an important one of which is `degree o f human guidance or forethought' . This continuum stretches from rote learning, through carefully guided concept-formation by observation, out toward independent theory formation . Six years ago, the AM program was constructed as an experiment in this latter kind of learning by discovery . Its source of power was a large body of heuristics, rules which guided it toward fruitful topics of investigation , toward profitable experiments to perform, toward plausible hypotheses and definitions . Since that time, we have gained a deeper insight into the nature of heuristics and the nature of the process of formin g and extending theories empirically . `The Nature of Heuristics I' paper presented the theoretical basi s for this work, with an emphasis on how heuristics relate to each other . This paper presents ou r accretion model of theory formation, and gives many examples of its use in producing new discoverie s in various fields. These examples are drawn from runs of a program called EURISKO, the successor to AM, that embodies the accretion model and uses a corpus of heuristics to guide its behavior . Since our model demands the ability to discover new heuristics periodically, as well as new domain objects an d operators, some of our examples illustrate that process as well . `The Nature of Heuristics III' pape r describes the architecture of the EURISKO program, and conclusions we have made from its behavior . 1 . The Learning Spectru m This paper deals with learning, by which we intend to include all processe s which result in accretion of knowledge. Section 1 surveys the spectrum of learning, from rote memorization to more complex activities . Section 2 focuse s on the far end of the learning spectrum, inductive inference . Section 3 proposes a model for the far end of that activity, empirical theory formation . That mode l accounts for the discovery of new heuristics as well as new task-specific object s and operations . Finally, Section 4 presents dozens of examples of the model i n action, producing discoveries in many different fields . The next paper in thi s Artificial Intelligence 21 (1983) 31–5 9 0004-3702/83/000-0000/$03 .00 © 1983 North-Holland