Factors influencing rate and extent of learning in the presence of misinformative feedback.

Publisher Website
Google Scholar
Abstract
On each of 240 trials 1 of 8 white stimulus lights arranged in a circle came on and the subject (S) selected 1 of 2 response keys. Selection of the correct key lit a green feedback light and the wrong key lit a red feedback light, except on misinformative trials. On these trials S was misinformed about the adequacy of his response and the opposite feedback light came on. The left key was correct (most often rewarded) for 4 stimulus lights and the right key was correct for the other 4. Groups were given 0, 10, 20, 30, 40, or 50% misinformative feedback. A 2d categorization depended on instructions and the presence or absence of a correction signal. The principal findings were as follows: 1. When Ss were aware of misinformative feedback they learned significantly faster and more went to optimal solution. The effect was most pronounced at 10% misinformation but was also apparent at the 20% and 30% levels. 2. Use of an unexplained correction signal near the end of misinformative trials significantly improved performance. 3. A further significant improvement was obtained when the correction signal was explained. 4. In general, the difficulty of learning optimal behavior increased as the percentage of misinformative feedback increased. 5. Considering runs of at least 3 consecutive errors to a single stimulus light, the average length of such runs was significantly greater when Ss were aware of possible misinformative feedback. In other words, awareness also increased persistence in incorrect modes of response. 6. Most individuals and groups were not showing probability matching behavior at the termination of practice. In some cases this may have been because practice was not carried far enough, but frequently it was because individuals or groups had surpassed a probability matching solution or reached a stable state below the matching level.
Keywords

This publication has 2 references indexed in Scilit: