Tolman’s rats, latent learning, and a fight the textbooks report as settled. · published 13 September 2026
Tolman and Honzik’s 1930 rats performed almost perfectly the day after reward arrived — real data that still needs explaining — but the textbook retelling turns an unresolved stalemate with Hull and Guthrie into a clean victory over behaviourism, and quietly omits that errors also fell in the group that was never fed at all and that the 1946 short-cut study the phrase ’cognitive map’ comes from could not be replicated.
These are the working notes, unedited. Everything the video
states as fact comes from here, and nothing else may be stated as fact.
The sections on limits and on what the finding does not show are
part of the brief, not an afterthought — see
the method.
The study: Tolman and Honzik, 1930. Food-deprived rats ran a fourteen-unit T-maze once a day for over three weeks, in three groups. One group found food at the end from the first day. One never found food at all. The third found nothing for ten days, and then on the eleventh day food was placed in the end box before the rat set off.
The result everyone quotes. The third group’s errors fell and its speed rose almost overnight once the food appeared, matching and by some measures beating the group that had been fed from the start. Tolman’s reading: those rats had been learning the maze all along and had no reason to show it. Reward changed performance, not learning. He called it latent learning, and in “Cognitive Maps in Rats and Men” (1948) argued the animals were building an internal layout of the maze rather than a chain of turns.
What the retelling gets wrong. Arthur Jensen reviewed how this is taught across forty-eight introductory textbooks published between 1948 and 2004 and found the same errors repeated. Textbooks say the experiment proved reinforcement is not necessary for learning; the question was argued for decades and never resolved. They say behavioural theories could not account for the result; Hull and Guthrie each produced coherent stimulus-response explanations at the time. They attribute to Skinner the claim that reinforcement is necessary for learning, which he did not make. And they present the cognitive map as the only available explanation without mentioning that it was disputed.
The detail that complicates the story. Over the first five days the error rate fell in all three groups, including the one that never found food at all. Something was being learned without reward everywhere in the experiment, which makes the clean two-group morality tale harder to tell.
And the follow-up failed. The study the cognitive map is really named for — Tolman, Ritchie and Kalish (1946), where rats appeared to take a direct short-cut toward a goal from a new starting point — could not be replicated.
Limits
This is one maze, one laboratory, 1930, with the statistical conventions of its time. The Tolman-versus-Hull argument is usually reported as a victory and is better described as a stalemate: no side produced the decisive result, and the field moved on rather than concluding.
What this does not show
Not that Tolman faked anything or that latent learning is a myth — the overnight change in that third group is real data, and it still needs explaining. Nor that cognitive maps turned out to be nonsense: place cells in the hippocampus were discovered decades later and the work on the brain’s spatial system won a Nobel Prize in 2014. But that is a separate line of evidence from neuroscience, not a vindication of a 1946 maze study that nobody could repeat. The narrow point is about how a live, unfinished argument gets taught as a finished one.