Insight learning: what Köhler's chimpanzees actually showed

Insight learning is learning in which the solution to a problem appears suddenly and as a whole. It happens because the learner sees how the parts of the problem fit together. Köhler offered it as an alternative to slow trial and error. The concept comes from the psychologist Wolfgang Köhler. He studied chimpanzees on Tenerife from 1913 to 1917 and published his results in 1917.

The best-known example is Sultan, a chimpanzee who joined two sticks to reach a banana. Köhler’s own report of that scene is less tidy than many summaries. Sultan failed for over an hour, and he joined the sticks for the first time while he played with them. This article sets out what Köhler described and how insight differs from trial and error and from latent learning. It also shows where the four stages come from and what later studies changed.

• Insight learning means a sudden, complete solution that takes account of the whole situation. It is not a lucky hit after many separate attempts.
• Sultan first joined the two sticks in play, and only a keeper saw it. Köhler wrote that the accident was not the important fact. What Sultan did next was.
• Edward Thorndike looked for sudden solutions in cats, dogs and chicks in 1898 and found none. Köhler replied that the animals could not see how Thorndike’s boxes worked.
• The four stages that many courses teach come from Graham Wallas’s 1926 book on creative thought, not from Köhler.
• Pigeons solved Köhler’s box problem after they had learned its parts one at a time. Later studies show that sudden solutions often build on earlier learning.

What insight learning means

The APA Dictionary of Psychology defines insight learning as “a cognitive form of learning involving the mental rearrangement or restructuring of the elements in a problem to achieve a sudden understanding of the problem and arrive at a solution”. The dictionary adds that Köhler offered it “as an alternative to trial-and-error learning”. It dates his description to the 1920s. The English editions of his book appeared in 1925 and 1927, but the German original came out in 1917.

Köhler gave his own test for insight. In his book, he called it “the appearance of a complete solution with reference to the whole lay-out of the field”. An insightful solution is not a chain of chance movements that happens to end at the goal. From the start, it takes account of how the parts of the situation relate to each other.

The moment when the solution arrives has its own name. The APA Dictionary defines the aha experience, also called the aha moment, as “the emotional reaction that typically occurs at a moment of sudden insight into a problem or other puzzling issue”.

Köhler belonged to the Gestalt school. The APA Dictionary names Köhler, Kurt Koffka and Max Wertheimer as the German psychologists who argued for Gestalt psychology. It treats experience as an organised whole, not as a sum of separate parts. That view explains why Köhler cared so much about whether an animal could see the whole problem. Gestalt is one of the German words that English psychology kept, as our list of psychology terms that do not translate cleanly shows.

Köhler’s experiments on Tenerife

The Prussian Academy of Science kept a research station for apes on Tenerife from 1912 to 1920. Köhler’s studies there ran from 1913 to 1917. His book appeared in German in 1917 and in English, as The Mentality of Apes, in 1925. The page numbers in this article come from the revised English edition of 1927.

Most tasks had the same pattern. Food, usually fruit, lay out of reach, and the direct way to it was blocked. The animals had to take a detour, use a stick as a tool, make a tool, or move boxes under fruit that hung high up.

Some scenes match the textbook picture exactly. On 17 February 1914, food lay beyond the bars, out of Sultan’s reach. A castor-oil bush stood at the back of the room. Sultan looked out through the bars, turned, went straight to the bush, broke off a thin branch, ran back and reached the food. Köhler called it “one single quick chain of action, without the least ‘hiatus’”. A repeat soon after went less smoothly.

The double stick: what Sultan actually did

The most famous task is a different one. Sultan got two hollow bamboo rods. The thinner rod fitted into the end of the thicker one. Fruit lay outside the bars, just too far for either rod alone. Köhler describes what happened next:

1. 1.Sultan tried to reach the fruit with one rod, then with the other.
2. 2.He pulled a box to the bars and pushed it away again at once.
3. 3.He pushed one rod forward with the other rod until it touched the fruit. Köhler called this a “good error”: useless, but sensibly aimed at the goal.
4. 4.The observer put a finger into the opening of one rod. This hint had no effect.
5. 5.After more than an hour, Köhler stopped the test. Sultan kept the rods, and a keeper stayed to watch him.

Köhler knew the start of the solution only from the keeper’s report. In the keeper’s words, Sultan sat down on the box with the two rods and “plays carelessly with them”. By chance, he held them in a straight line and pushed the thinner one into the thicker one. At once he jumped up, ran to the bars and pulled in a banana with the double stick. The report covers “a period of scarcely five minutes” after the test stopped.

Then Köhler watched for himself. When the rods came apart, Sultan joined them again at once and pulled in all the fruit, one piece after another. The next day, he pushed one rod with the other for a few seconds. Then he joined both and reached the goal. When the fruit lay farther away, the solution came “quite suddenly”: Sultan added a third tube and reached the fruit “with the triple pole”.

Köhler did not hide the accident. In a footnote, he wrote that Sultan “had first of all connected the sticks in play and without considering the objective”. He judged the accident differently from his critics:

For Köhler, the point was what Sultan did with the double stick after the first success. He wrote that Sultan’s behaviour was then “exactly the same as if he had discovered the new procedure in a genuine solution”. So one of the best-known examples of insight learning starts with play and chance, and the man who reported it wrote that down himself.

Is insight learning really learning?

On an MCAT forum, a student asked how insight counts as learning, because it looks like using knowledge that you already have. Köhler’s records give two answers.

The first answer is that the change lasted. Before the first join, Sultan failed for over an hour. After it, as the record above shows, he used the double stick again and again. The solution stayed. It did not need many rewarded repetitions to grow.

The second answer is that the parts were not new. Köhler noted that the chimpanzees often poked straws and small sticks into holes and cracks in their play. Putting a stick into a hole was an old action for Sultan. What was new was the use of that action for the goal. Later research, below, took this point much further.

Köhler also warned against seeing insight everywhere. When the apes built towers of boxes, the general idea of “higher box up” was clear to them. But he wrote that the final result “was achieved only by chance, and after an almost entirely unintelligent muddling around”. In his records, insight was a feature of some solutions, not a label for all ape behaviour.

Insight learning vs trial and error

Köhler wrote against a well-known rival view. In his 1898 study of animal intelligence, Edward Thorndike tested cats (in his words, “rather kittens”), dogs and chicks. He put hungry cats and dogs into boxes with food outside. Each box opened with a simple action, such as a pull on a loop. Thorndike measured how long each animal needed to get out on each trial. Over many trials, the useless movements dropped away. The successful movement, in his words, was “stamped in by the resulting pleasure”.

Thorndike looked for insight in these data, and he said what it would look like. If an animal understood the box, it would do the right act at once from then on. “There ought, that is, to be a sudden vertical descent in the time-curve,” he wrote. He found none:

The APA Dictionary traces the term trial-and-error learning to Thorndike’s theory and notes that he first preferred the phrase “trial-and-accidental success”. Thorndike later stated the law of effect: responses that bring satisfaction become more firmly connected to the situation. Our comparison of classical and operant conditioning explains how that idea became operant conditioning.

Köhler thought Thorndike’s boxes were the wrong test for insight. The animals often saw only one end of the mechanism. Thorndike’s own description of one box says that its string ran outside the box and was “inaccessible and invisible from within”. Köhler asked: “For if essential portions of the experimental apparatus cannot be seen by the animals, how can they use their intelligence faculties in tackling the situation?” In his own tasks, the goal and the tools were in full view.

Köhler also argued that the apes did not try things blindly. He wrote that a solution often followed “a period of perplexity or quiet”, and then ran as “one continuous smooth action”. He gave an example. A local colleague who believed in the chance theory of animal behaviour came to watch Sultan. What impressed him most was a pause in which Sultan “slowly scratched his head and moved nothing but his eyes and his head gently, while he most carefully eyed the whole situation”.

Insight learning vs latent learning

Insight and latent learning are often confused, because neither follows the pattern of reward and repetition. The difference is in timing and purpose. Insight solves a problem that the learner has now, and the solution appears suddenly. Latent learning stores something that does not show in behaviour until there is a reason to use it later.

The classic studies also differ. Insight comes from Köhler’s apes and their tools. Latent learning comes from the rats of Hugh Blodgett and Edward Tolman. They learned a maze with no food at the end and showed it only when food appeared. Our article on latent learning describes those experiments and compares the two terms with observational learning.

The stages of insight

Many courses and websites teach insight as a series of stages. One well-known set has four stages, and it comes from the political scientist Graham Wallas. In his 1926 book The Art of Thought, Wallas started from the physicist Hermann von Helmholtz’s account of how his best ideas came to him. Wallas named three stages from it: Preparation, Incubation and Illumination. Then he added “a fourth stage, of Verification”.

Wallas described human creative thought, with examples from Helmholtz and the mathematician Henri Poincaré. He did not write about apes, and Köhler’s book has no numbered stages. Wallas also named a moment inside Illumination, which he called Intimation. It is the feeling that a solution is coming before you know what it is.

The table also shows where the model and the scene do not match. In Wallas’s model, the new idea arises in the thinker’s mind. In Köhler’s report, the first join happened in play, by chance and without attention to the goal, and Sultan used it at once. The stages are a useful memory aid for an exam. They are not a description of what happens in every case of insight.

What later research changed

One question was open from the start: what had Köhler’s chimpanzees already learned about sticks and boxes? According to Cook and Fowler (2014), little was known about it. H. G. Birch tested the question in 1945. As Ebel and Call (2018) summarise it, most of his chimpanzees first failed to pull in food with a hoe. After a period of free play with sticks, they succeeded.

In 1984, Robert Epstein and colleagues repeated Köhler’s box problem with pigeons. All the pigeons learned to peck a toy banana. The researchers trained the other parts separately, in different combinations for different birds. Some pigeons learned to push a box towards a target. Some learned to climb onto a box and peck the banana above it. Some learned to ignore the banana when it hung out of reach. Then each bird saw the full problem for the first time: the banana out of reach, and the box somewhere else.

The pigeons with all three skills first looked back and forth between the box and the banana. Then they suddenly pushed the box under the banana, climbed up and pecked. According to Epstein (1999), the whole performance typically took about a minute. The 1984 paper said the birds solved the problem “in a remarkably chimpanzee-like (and, perforce, human-like) fashion”.

When a skill was missing, the result changed. One pigeon had learned to climb but not to push the box towards a target. It pushed the box around and rarely looked up. After 14 minutes, it looked up by chance when the box was under the banana, and it climbed at once. Birds that had never learned to push, or never learned to climb, did not solve the problem.

Cook and Fowler repeated the pigeon study in 2014. According to their summary, Epstein’s three pigeons had solved the problem in less than 2 minutes. Their own pigeons solved it too. Then they gave the pigeons two boxes with the same training history, and a pigeon could stand on only one of them. The pigeons pushed both boxes equally and without direction. The authors concluded that the birds did not understand what the box was for. They explained the solution by the layout of the task and the behaviour the birds had learned before.

Ebel and Call added a similar result for great apes. Their study had 25 apes. The apes that first explored an empty apparatus later solved the baited task faster than the apes that met the food first.

None of this proves Köhler wrong, but it moves the question. A sudden solution can come from new combinations of parts that the learner learned earlier. Whether the animal understands what it does is still debated.

Insight in people

In human research, insight is now its own field. In a 2014 review, John Kounios and Mark Beeman wrote that insight occurs when a person “suddenly reinterprets a stimulus, situation, or event to produce a nonobvious, nondominant interpretation”. That can be the solution to a problem, the point of a joke, or a new way to see an ambiguous picture.

Aha solutions are not only a feeling. In four experiments by Salvi and colleagues (2016), people solved word and picture puzzles and said for each answer whether it came by insight. Their insight solutions were, on average, more accurate than their step-by-step solutions. These were puzzles in a laboratory, so the result does not mean that every aha feeling is correct.

Not every researcher thinks that the ape studies and the human studies measure the same thing. In a 2024 paper, Eli Shupe argued that the insight in human studies is “fundamentally different” from the insight in animal studies. For an exam, the textbook definition still applies.

Keeping the learning theories apart

Insight learning becomes clear when you compare it with the other learning theories. In operant conditioning, consequences shape behaviour over many trials, as our article on schedules of reinforcement shows. In insight learning, a view of the whole problem shapes a solution that appears at once.

The studies by Birch and Epstein suggest one careful lesson for your own studying: the sudden solutions built on parts that were learned before. If you want to see how terms connect, it helps to know each term well first. That is a reasonable reading of the studies, not a tested study technique.

In Psychology World, you can save the learning terms that you mix up to one list. Examples are insight learning, trial and error, latent learning, observational learning and operant conditioning. Then 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 on how to memorise psychology terminology explains which study methods have the strongest evidence.