Learning in a Probabilistic Environment: A New Approach, and Some Preliminary Findings.
Abstract
Many studies of 'probability learning' have led to the conclusion that human learners cannot find the 'rule' amidst the 'noise' (Brehmer, 1980). It is hypothesized that under more natural conditions, learners do develop rules which are probabilistically predictive, and improve chiefly through the addition of new predictive variables. The present study represents natural learning situations by including: instructions and rewards that emphasize gradual development of understanding, rather than discovery of the right rule; and a large number of cues, which must be discovered, rather than a few cues explicitly given. Results with 12 college-student subjects indicate significant learning in a computer-displayed task, over approximately 10 hours of experience. Learning was incremental, and was accompanied by the addition of valid factors to existing rules. These results contrast with findings that people fail to utilize information effectively in probabilistic situations. Earlier studies do not, however, model situations in which learning requires the discovery and validation of predictive cues, processes critical for the development of real-world expertise.
Document Details
- Document Type
- Technical Report
- Publication Date
- May 01, 1983
- Accession Number
- ADA137031
Entities
People
- J. Klayman
Organizations
- University of Chicago