Concept learning set-size functions for Clark's nutcrackers.

Researchers taught Clark's nutcrackers to tell if two pictures were the same or different. They kept adding new picture pairs until the birds got every test right without extra training.

The team wanted to know how many examples the birds needed before the concept clicked. They tracked the exact number of training pairs it took to reach perfect performance.

Most birds needed 64 to 128 different picture pairs before they could sort brand-new pairs correctly. After that point, they rarely made mistakes, even with pictures they had never seen.

This number range matches what monkeys need and is smaller than what pigeons require. The birds showed they could form an abstract same/different rule, not just memorize each pair.

Fields et al. (2002) first showed that people also learn categories faster when you give them many examples, but warned that too many choices at once can slow the process. The nutcracker study keeps the choices simple and still finds the same "more examples help" pattern.

Luciano et al. (2007) used multiple-exemplar training with a baby and got symmetry learning with far fewer items. The difference makes sense: human infants already have language-like brains, while birds need extra examples to reach the same abstract level.

Carey et al. (2014) remind us that every trial matters when we graph learning. Their warning about sparse data lines up here; the nutcracker curves would look jumpy if the researchers had skipped trials while counting toward that 64-128 range.

If you run stimulus equivalence lessons, plan for at least 60-120 exemplars before expecting true concept mastery. Start with simple two-choice arrangements to keep the task clear, and collect every trial so your visual analysis matches the real learning curve. This count gives you a realistic timeline for programming generalized skills in learners who need more repetition.