Rewards strengthen behaviors they follow. This principle, articulated in Thorndike's Law of Effect, is foundational to behavioral science-but it obscures two different ways behaviors may be strengthened. First, and axiomatic, is that the frequency of rewarded behaviors increases. Second, but less well established, is that the variability of rewarded behaviors decreases. Indeed, since Thorndike's pioneering research, investigations into behavioral variability have posed a challenging theoretical puzzle: Some indicate a winnowing of behavioral variants, whereas others suggest an ongoing waxing and waning of even dominant behavioral variants throughout rewarded training. We devised a new experimental task which allowed us to monitor pigeons' sequential pecking of five visually distinctive touchscreen buttons, with all sequences delivering food reward. The 120 possible five-peck sequences we monitored over 250 daily sessions produced a rich set of data to help solve this puzzle. We found that pigeons did decrease the diversity of the sequences they performed. Nevertheless, pigeons continued to perform many different sequences-with even the most dominant sequences frequently rising and falling. These findings and others in diverse realms of behavioral science, neuroscience, and computer science suggest a far-reaching solution to this puzzle: The form and frequency of consistently rewarded behavior involve a dynamic adaptive balance between stability produced by the Law of Effect and variability induced by a persistent exploratory predisposition. Called "the edge of chaos," this balance point preserves responses that reliably secure rewards and engenders behavioral flexibility in the event that the prevailing contingencies change or better options arise. (PsycInfo Database Record (c) 2026 APA, all rights reserved).
Wasserman et al. (Mon,) studied this question.