Search Papers | Poster Sessions | All Posters
Poster C84 in Poster Session C - Friday, August 9, 2024, 11:15 am – 1:15 pm, Johnson Ice Rink
Learning of predictable rules mixed with random reinforcement in humans
Leo Jin1 (), Greg Jensen2, Jacqueline Gottlieb1, Vincent Ferrera1; 1Columbia University, 2Reed College
Humans as well as animals are constantly learning novel predictable relationships to better adapt to the environment. However, such “learnable” patterns are often intermixed with noisy “unlearnable” randomness. It is not known if, when they are presented simultaneously, humans are capable of differentiating them, so that more energy can be invested in learnable rules. Here, we exposed humans with two pictorial sets: a “learnable” set in which the stimuli were implicitly ordered and the correct response was always to choose the higher-rank stimulus, and an “unlearnable” set in which stimuli were unordered and feedback was random regardless of the choice. The behavior patterns under the two sets were extremely polarized: Some participants ordered the stimuli in neither set (non-learners). Others ordered the stimuli in both sets, learning the correct order from the learnable set while behaving as though some ordering also existed from the unlearnable set, consistent with our previous finding from monkey behavior. Only when subjective ordering of the unlearnable set was strongly discouraged did many participants start to behave differently toward the two sets. Our results suggest that under the neutral condition humans did not differentiate well between real (learnable) patterns as opposed to random reinforcement, which contributes to deeper understanding of multi-rule learning and the formation of persistent superstitious biases.
Keywords: learnability randomness superstitious learning transitive inference