Latent learning: what Blodgett's and Tolman's rats actually showed

Latent learning is learning that happens without an obvious reward and does not show in behaviour until there is a reason to use it. The classic evidence comes from rats that ran a maze for days with no food at the end. They kept making wrong turns until food appeared, and within two days of finding it they ran the maze almost without errors. They had learned the maze all along. They had no reason to show it.

• The term comes from Hugh Blodgett, whose experiment was published in 1929. Edward Tolman, who made latent learning famous, gave Blodgett the credit for both the first experiment and the concept.
• In Tolman and Honzik’s 1930 study, rats that first found food on day 11 cut their errors sharply the next day. The authors were more careful about what that meant than most summaries are.
• “Latent” means hidden in performance. In the rat experiments it describes what the error counts did not show, not what the rats were aware of.
• The experiments separate learning from performance: the reward changed what the rats did with what they knew. Albert Bandura later found the same split in observational learning.
• Whether any learning is truly free of reward was argued for decades. The reinforcement side pointed to possible hidden rewards, such as curiosity and escape from the maze.

What latent learning means

The APA Dictionary of Psychology defines latent learning as learning “that is acquired without conscious effort, awareness, intention, or reinforcement”. The learning “is not manifested as a change in performance until a specific need for it arises”. The dictionary’s human example is a student who can quote a passage accurately in an exam without ever having tried to learn it. Its animal example is a rat that explores a maze without reward. Later, that rat finds the goal faster than a rat that has never seen the maze.

Two ideas sit inside that definition. The first is learning without reinforcement. In operant conditioning, a behaviour changes because of what follows it. Our comparison of classical and operant conditioning explains this, and the schedules of reinforcement show how the timing of rewards matters. Latent learning is the claim that some learning happens when no consequence strengthens it. The second idea is the gap between learning and performance. What an animal knows and what it does are not the same. A test that measures only behaviour can miss what was learned.

The word “latent” refers to the second idea. A 1951 review by the psychologist Donald Thistlethwaite noted two senses of the term. The first is learning “not manifest in performance scores”. The second is learning when the incentive is irrelevant to the animal’s current drive. Blodgett, who coined the term, used it in the first sense. The APA definition adds that the learning happens without conscious effort, awareness or intention, which fits the human examples. In the rat experiments, nobody measured awareness. What stayed hidden was the performance.

Tolman’s name for what the rats acquired was a cognitive map. The APA Dictionary defines a cognitive map as “a mental understanding of an environment, formed through trial and error as well as observation”. It credits the concept to Tolman.

Blodgett’s 1929 experiment: where the term comes from

Most summaries start with Tolman, but Tolman did not start with himself. In his 1948 paper Cognitive maps in rats and men, he wrote: “The first of the latent learning experiments was performed at Berkeley by Blodgett. It was published in 1929. Blodgett not only performed the experiments, he also originated the concept.” Blodgett’s paper is The effect of the introduction of reward upon the maze performance of rats. It is a shortened version of his doctoral report, which he deposited at the University of California in 1925.

Blodgett ran rats through a maze once a day and counted their errors. One error was an entry into a blind alley by a full body length. The main comparison used three groups:

• Group I, the control group: 36 rats found food at the end of every run and ate there for three minutes.
• Group II: 36 litter mates of Group I. For the first six days they found an empty food box, stayed in it for two minutes, and were fed later in another cage. Food was waiting at the end of the run on day 7 and on every run after that.
• Group III: the same as Group II, but the food appeared at the end of day 3 instead of day 7.

While the experimental rats found no food, their errors fell only slowly. Then the reward arrived, and the error rate dropped in one day.

Group III showed the same pattern four days earlier: 2.72 errors on day 3 and 1.20 on day 4. In Blodgett’s summary, the first-day drop in mean errors after the reward “was greater than that made by the control group on any single day”.

Blodgett was careful about what his statistics proved. First, he compared the Group II drop with the largest one-day drop of the control group. A difference that large would occur by chance 904 times out of 10,000, so he called the result “probable, although not certain”. His strongest result came from a second comparison, against the control curve at the same error level. For Group III, the chance was only 173 in 10,000, which he called “a very strong indication” of latent learning.

He also tested whether the rats had learned the route or had only got used to the apparatus. A fourth group of 10 rats ran the maze backwards, from the empty food box to the start, for seven days. When they then ran it forwards to food, they did not show the same sudden improvement. Blodgett concluded that latent learning is “something more than a general familiarity” with the maze.

Tolman and Honzik’s 1930 study: the famous graph

The latent learning experiment that most summaries describe is Tolman and Honzik’s. Their paper, Introduction and removal of reward, and maze performance in rats, appeared in 1930 in the same series as Blodgett’s. Summaries usually show it as one experiment with three groups. The paper is more complicated, and the details change what the graph can prove.

The experiment used a 14-unit T-maze and 82 male rats in two groups of 41. Each rat ran the maze once a day for 22 days. The HNR-R group (hungry, not rewarded, then rewarded) found no food for 10 days, and food was put in the end box on day 11. The HR-NR group had the opposite schedule: food for 10 days, then none. Rats that found no food in the maze were fed in their home cages at least three hours after the run.

The famous result came on day 12, the first run after the rats had found food. The errors of the HNR-R group fell sharply, and on day 13 they fell below the errors of rats that had found food every day. But the comparison curves came from a different experiment.

All groups were scored in average errors, so curves from the two experiments could share one graph. But the delayed-reward rats never ran side by side with the always-rewarded rats.

The authors drew the cautious conclusion that this design allows. The difference between the HNR-R and HR curves from day 12 on was statistically large, with a critical ratio of 12.2. Even so, they were “not ready, however, to state with any certainty that this was due to the greater efficiency of the latent learning”. They added: “It may have been that the rats of the HNR-R Group were brighter as regards maze learning than the HR Group.” Their summary keeps the hedge: the result “suggests that latent learning may be more effective than overt learning”.

Two more details rarely appear in summaries. First, the HNR-R curve was not flat before the reward. In the paper, it shows “a downward trend” during the 10 days without food, similar to the never-rewarded group. The rats improved a little without food and much faster with it. Second, the experiment also ran in reverse. When the HR-NR rats stopped finding food, their errors rose on day 12. The average rise was 1.1 errors per rat, to about the level of the never-rewarded group. The authors traced the rise to the rats searching blind alleys that they normally avoided. Reward changed performance in both directions.

Secondary sources often number these groups 1, 2 and 3, but not in the same order. “Group 2” is the never-rewarded rats in one source and the delayed-reward rats in another. The paper names each group by its condition: H for hungry, R for rewarded, NR for not rewarded.

From latent learning to the cognitive map

Tolman used these results to argue that rats learn more than responses. In 1948 he wrote of Blodgett’s rats that, during the unrewarded runs, “they had learned where many of the blinds were. They had been building up a ‘map,’ and could utilize the latter as soon as they were motivated to do so.”

He called a study by Spence and Lippitt at the University of Iowa “probably the best experiment demonstrating latent learning”. Spence was himself a reinforcement theorist. In Tolman’s account, rats ran a Y-maze with food at the end of one arm and water at the end of the other. During training, they were neither hungry nor thirsty. In the test, hungry rats chose the food arm and thirsty rats chose the water arm. Both did so “statistically more frequently” than they chose the other arm. They had learned where each goal was while they needed neither.

Was it really learning without reward?

The latent learning experiments became the centre of a long argument. Thistlethwaite’s 1951 review in Psychological Bulletin named two questions. Is reinforcement “a necessary condition for learning”? And “should a distinction be made between learning and performance”? Reinforcement theorists, among them Thorndike, Hull and Spence, held that some rewarding after-effect is necessary for learning. Tolman and others argued that the latent learning data showed that some learning does not need it.

The reinforcement side did not have to deny the data. It could argue that the unrewarded rats were rewarded after all. Thistlethwaite lists the obvious candidates for the Blodgett experiments: “satisfactions of curiosity and escape from the maze”. His reply was that “the abrupt behavioral changes following the introduction of reward simply have not been shown to be deducible from the reinforcement postulate.” In the exploration experiments, he added, such a reward comes on wrong routes as well as right ones. So it “offers no basis for the differential conditioning of correct responses”.

The evidence was not one-sided. Thistlethwaite sorted the experiments into four types. In one type, hungry or thirsty rats ran past a goal they did not need, and were then tested after their need had changed. For that type, he reported that “many investigators have failed to find evidence of latent learning”, while others reported positive results. His review is also not a neutral referee. It was part of a dissertation at Berkeley, and its first footnote thanks Tolman for advice.

Recent work reframes the question rather than ending it. In a 2020 review, the neuroscientists Maya Wang and Benjamin Hayden wrote that “latent learning is not driven by standard reinforcement processes”. They proposed that curiosity, a desire for information, is what motivates it. On this view, the exploring rat is not passive. It explores because it wants information, and the map that it builds is useful later. That is a hypothesis set out in their review, not a settled result.

Latent learning in people

Human examples are easy to describe and hard to test. A passenger who has been driven along a route many times may be able to drive it the first time they have to. These are illustrations of the definition, not experiments: in everyday life, it is rarely possible to show that no reward was involved.

Controlled evidence exists for very young children. In a 2005 study in the journal Infancy, Jennifer Campanella and Carolyn Rovee-Collier showed 3-month-old infants two objects together. The infants were too young to show what they had learned, so the researchers kept the memory active with periodic reminders. At 6 months, the researchers demonstrated some actions on one object, and the infants copied those actions on the other object a day later. Infants who had seen the two objects apart did not. The authors concluded that “latent learning by very young infants is both extensive and enduring”.

Latent learning vs observational and insight learning

Students often confuse latent learning with two other kinds of learning that also do not follow the reinforcement pattern. A typical question on an MCAT forum asks whether observational learning is a subset of latent learning. It is not, but the two can overlap.

Observational learning, which the APA Dictionary also calls vicarious learning, is learning from watching others. The bridge to latent learning is a 1965 study by Albert Bandura. Children watched a filmed model behave aggressively. The model was rewarded, punished, or left without consequences. Children who had seen the model punished copied fewer of the model’s actions. Then all the children were offered attractive rewards for copying the model. According to the abstract, “the introduction of positive incentives completely wiped out the previously observed performance differences, revealing an equivalent amount of learning”. The children had learned the actions they did not perform, which is the same split between learning and performance that Blodgett’s rats showed.

Insight is different again. The APA Dictionary defines insight as “the clear and often sudden discernment of a solution to a problem”. The classic studies are Wolfgang Köhler’s experiments with chimpanzees on Tenerife. The Prussian Academy of Sciences ran a research station there from 1912 to 1920. Insight solves a problem that the learner has now. Latent learning stores something that the learner uses later.

Classifying exam scenarios

Exam questions describe a situation and ask which kind of learning it shows. Ask three questions in order. Was there a model to watch? Did the learning show only later, when a reason appeared? Did a consequence change the behaviour while it was learned?

Two traps are common. The first is a scenario where someone watches a model and copies later. That is observational learning, even though the learning stayed hidden for a while. The second is a rat that improves while it is rewarded on every run. That is ordinary operant learning. The latent part needs a period without the reward and a clear change once there is a reason to use what was learned.

Keeping the terms apart

Latent learning, cognitive map, observational learning, vicarious learning, insight and operant conditioning blur easily, because each describes learning from a different source. In Psychology World you can save terms like these to one list. Then you can drill the list as flashcards, as multiple-choice questions once it has four or more terms, or by typing the answer. What each study mode does is set out on the home page. Our guide to memorising psychology terminology explains which study methods have the strongest evidence.