Analogical Reasoning: A Human Superpower

How the human understanding of relationships sets us apart from animals

Look at the puzzle below. Which of the images in the second row best fits in the box with the question mark?

If you guessed D, you solved an analogical reasoning problem: bird is to nest as dog is to doghouse, or bird:nest::dog:doghouse. You judged that things were similar not because of anything about them that can be directly perceived, like color, shape, or size, but because they have a similar relationship to each other. The nest is the home of the bird, just as the doghouse is the home of the dog. Such similar relationships are what constitute analogies.

A new book by psychologist Keith J. Holyoak, The Human Edge: Analogy and the Roots of Human Intelligence, makes the case that analogical reasoning enables humans to be creative problem solvers. Analogies are the source of many scientific innovations, as well as the metaphors on which poetry is built. Analogical reasoning ability is a key component of intelligence. It is also a distinctively human ability. Most four-year-old children can find the right answer to the puzzle above, and by nine, virtually all of them can.¹ However, there is no convincing evidence that non-human animals can solve this analogy problem or even simpler ones.

Analogical reasoning and creative problem solving

Imagine that a surgeon has a patient with a malignant stomach tumor. A certain kind of radiation will kill the tumor. In small doses, this radiation is harmless, but the large and intense dose required to kill the tumor would also destroy the flesh along the path to the tumor, which would result in the patient’s death. How can the surgeon use the radiation to cure the patient without killing them?

I was flummoxed when I read the problem, as most people are. But let’s consider another case. You’re a general who wants to capture a fortress that lies at the center of many different roads. If you send your whole army down one road, it will be destroyed because there are mines in the roads that will be set off by that many troops. So you decide to divide your army into multiple small groups and send them down different roads, and you are successful in capturing the fortess.

Do you see how you can kill the tumor without harming the patient now? The doctors should send small amounts of radiation along several pathways that converge on the tumor, just as the general sends small numbers of troops by many roads.

When experimental subjects are given the radiation problem in isolation, only 10% can solve it. However, if they are given the general problem beforehand, about 30% more of them get the right answer. If they are given a hint that the general problem might be relevant to solving the radiation problem, an additional 30% solve the puzzle.²

The solution to this problem involved analogical reasoning, or detection of similar relationships among entities. What makes the cases of the stomach tumor and the assault on the fortress similar is that the elements have similar relationships to each other. The general is like the doctor, the tumor is like the fortress, and the small radiation doses are like the troops. Both the troops and the radiation doses converge on the target from many different directions.

From Holyoak, The Human Edge, p. 50.

The analogy enables people to come up with a creative solution to the tumor problem that they usually cannot think of unaided. Not only that, but the analogy leads to the formation of a new concept or schema: one way of applying force to a point is to divide the force into smaller forces that converge on the point. This convergence schema may enrich our conceptual repertoire if we hadn’t thought of it before.

By using analogies, we recruit information from one scenario to provide us with information about another. Analogies can be used not only to generate novel hypotheses, but to predict the existence of unobservable properties and to explain difficult concepts.

The analogy gap

Most scientists who compare human and animal cognition believe that cognitive abilities lie on a continuum. Humans may vastly exceed animals in their ability to understand analogies, the mental states of others, causation, and other problems, but the difference is one of degree, not of kind. Human abilities, on this view, exist in a rudimentary form among apes and other animals.

Holyoak challenges this paradigm, however. In The Human Edge and this article, he argues that human cognition is fundamentally different from animal cognition because humans can form explicit concepts of the relationships between things, and animals can’t.³ Explicit concepts of relations are the basis of analogy. We need to form the concept “is the home of” before we can understand that bird:nest::dog:doghouse.

There is some experimental evidence that appears to confirm that animals can understand the very simplest of relationships: sameness and difference. However, Holyoak and his colleagues convincingly argue that this appearance is deceiving. Take the exercises below in which animals are prompted to judge which of the two figures in the bottom row is the best match for the figure at the top. The answer, which is obvious to us, is that the best match is the first item: the two cubes are the same, as are the two puzzle pieces, and the clocks are all the same, like the envelopes. We answer the question easily because we have explicit concepts of sameness and difference.

From Holyoak, The Human Edge, p. 85.

Many types of animals, including pigeons, chimpanzees, and baboons, can be trained to answer correctly on exercise B, but they usually fail to do the same on exercise A. Rather than using explicit concepts of sameness and difference, the animals turn out to be judging the amount of variety in an image. There is a great deal of variety among the images in the second square of the bottom row in exercise B, so animals can distinguish it as different from the uniform other images in the exercise. In example A, there is less variety among the images, so animals have a harder time detecting similarity and difference.

Humans have explicit concepts of a countless number of relations: we can understand not only that one thing is the same as another, but also that one thing is on top of another or one thing causes another or one thing is chasing another. Because of these concepts, we can form analogies between different sets of things that are related in the same way.

Because humans have explicit relational concepts, we can apply them easily across domains. We can judge that two images, two ideas, or two sounds are the same. When one pool ball slams into another, we can judge that the first ball causes the second to move. But we can also use this notion of causality in radically different domains, such as when we judge that poverty causes crime or that necessity causes invention.

Until the age of four, human children perform much like animals on analogy tasks, in that they answer correctly on exercise B, but not on A. However, around the age of four, children start pulling away from animals on these tasks, and they begin to perform successfully on creative reasoning tasks by using analogy.⁴ Increasing skill in relational reasoning is driven by the maturation of the brain, which brings with it improvement in executive functions like working memory and inhibition that Holyoak argues are crucial to relational reasoning.

Holyoak and colleagues have argued that a whole slew of cognitive differences between human and nonhuman animals are explained by the human talent for relational reasoning. This talent explains not only human superiority in analogical reasoning, but also our superiority in reasoning about causation, rules, spatial relationships, transitivity, hierarchy, and the mental states of others.

Analogical reasoning and intelligence

Not only is analogical reasoning one of the abilities that set us apart from animals, but it is a key component of intelligence. Intelligence tests like IQ tests are largely, perhaps even mostly, measuring analogical reasoning abilities. So when you say someone is bright, you may just be saying that they have a talent for analogy.

Here’s an example of a question from the Raven’s Progressive Matrices IQ test, generally considered to be among the most reliable measures of intelligence. What answer is best?

If you managed to guess 4, you did so through analogical reasoning. You figured out that the figures in the right-most column are what remains of the middle row after you subtract the first row from it. Many intelligence tests test verbal analogies as well. Take this question:

Such tests of analogical reasoning are among the best tests of intelligence. Psychologists have defined a factor, called the “g factor,” that accounts for performance on diverse intelligence tests. The best tests of intelligence capture this g factor and are said to be the most g-loaded. The chart below ranks different types of intelligence tests, with the most g-loaded at the center. Most of the tests at the center measure analogical reasoning ability.

Snow, R.E., Kyllonen, P.C., & Marshalek, B. (1984). The topography of ability and learning correlations. In R. J. Sternberg (Ed.), Advances in the psychology of human intelligence, Vol. 2 (pp. 47–103). Erlbaum. Quoted in Holyoak, Human Edge, p. 29.

Intelligence is based on the brain’s executive functions (EFs), which are “a family of top-down mental processes needed when you have to concentrate and pay attention, when going on automatic or relying on instinct or intuition would be ill-advised, insufficient, or impossible.”⁵ Executive functions are essential for reasoning, problem solving, and planning. The basic EFs are inhibition, working memory, and cognitive flexibility. Scores on analogical reasoning correlate with scores on other EFs. Not only that, but analogical reasoning ability predicts variations in intelligence even when EFs are controlled for. For this reason, some psychologists consider analogical reasoning to be an independent EF in its own right.⁶

Analogical reasoning and scientific genius

Analogical reasoning is one means by which scientists conceive of novel theories. Physicists developed their theories of light and sound by making an analogy between these phenomena and waves of water. Darwin’s theory of natural selection was spurred by an analogy between human breeders’ selection of animals for desirable traits and the selection of the fittest organisms by the natural environment.⁷

Psychologist Dedre Gentner wrote a case study of analogy in the writings of Johannes Kepler (1571-1630), who founded the science of astrophysics.⁸ Kepler’s primary scientific achievement was transforming Copernicus’s mathematical theory of the heliocentric universe into a physical theory. Kepler observed that planets moved more slowly the further away they were from the sun. To explain this fact, he made use of a brilliant analogy. Just as a light is dimmer the further away it is, so there must be some force that operates between astronomical bodies that weakens the further they are from each other. The sun exerts this force, which Kepler called the “moving power” (vis motrix), on the planets, but more weakly on the distant planets, which explains why they orbit more slowly. Through this analogy with light, Kepler had begun to conceive of gravity, and he came close to formulating the mathematics of how the strength of gravity varies with distance. Kepler developed his theory of the moving power through further analogies. He sometimes argued that it was akin to magnetism and other times, to a great water current that was generated by the sun’s rotation.

Kepler was very fond of analogies and used them often. He wrote, “I especially love analogies, my most faithful masters, acquainted with all the secrets of nature.” His writings teem with analogies, such as this one describing his reluctance to work on astrology, which he understood was a pseudoscience:

A mind accustomed to mathematical deduction, when confronted with the faulty foundations [of astrology] resists a long, long time, like an obstinate mule, until compelled by beating and curses to put its foot into that dirty puddle.

Kepler devoted great time and effort to developing, refining, and interrogating the analogies that generated his scientific theories. Since he saw an analogy between the moving power and magnetism and light, he intensively researched both subjects. He thought in detail about the consequences and limitations of his analogies. For example, he wondered whether the analogy between light and the moving force meant the latter could be eclipsed and concluded that it could not, as this part of the analogy did not hold up.

Gentner compares Kepler’s thinking to that of contemporary scientists, who also regularly use analogy in their thinking. Today’s scientists typically use near analogies, or analogies between closely related domains. As an example, they might compare the operation of the Ebola virus to that of the Herpes virus.⁹ By contrast, Kepler was skilled in creating distant analogies between radically different entities, like light, gravity, and magnetism, or scientific investigation and the progress of a mule.

Gentner speculates that skill in conceiving of distant analogies aids in the invention of radically novel theories that cause paradigm shifts. Her work also suggests that a taste for analogies may be an essential trait of scientific genius.

The neuroscience of analogical reasoning

An array of neuroscience work has found that analogies are processed in the frontoparietal control network, which is the brain region involved in directing attention and breaking complex tasks down into sequential steps. As the chart shows, four hubs in this network are particularly relevant for analogies.

Say that you’re trying to solve this problem: insect:bee::fish:? The main hubs required for soliving this problem are:

• The posterior parietal cortex (PPC), which is responsible for representing a relation, like insect:bee.

• The dorsolateral prefrontal cortex (DLPFC), which maintains relations in working memory so they can be further processed.

• The ventrolateral prefrontal cortex (VLPFC), which suppresses interference from information that is irrelevant to a reasoning problem. The term “fish” is closely related to many other concepts, like “water,” but these relations that aren’t relevant to the problem must be suppressed to prevent confusion.

• The rostrolateral prefrontal cortex (RLPFC), which integrates relations with others to form an analogy like insect:bee::fish:halibut.

The locations of the different hubs involved in analogical reasoning tell scientists much about how it works. For example, the PPC is normally involved in the representation of space. It seems that the brain might be representing relationships spatially by placing them in positions relative to each other. The RLPFC or “Brodmann Area” is much larger and has different features than it does in the brains of primates and changes substantially over the course of maturation. So this area may account for much of what makes human intelligence different from non-humans’ and adult intelligence different from children’s.¹⁰

Analogical reasoning and ChatGPT-3

Holyoak and his colleagues have found that the large-language model ChatGPT-3 has approximately human-level abilities in most forms of analogical reasoning.¹¹ When you give them matrix tasks, a:b::c:d tasks, or problem-solving tasks like the doctor/general task above, the LLM scores much like human beings. Oddly, ChatGPT shows weaknesses in analogical reasoning that are similar to human weaknesses. For example, when you give them two analogous stories, both humans and the LLM sometimes need a hint before they can see the analogy. Also, both intelligences find it easier to solve simple than complex analogies. As Holyoak puts it:

Basically, when the program was compared with human participants who had been asked to solve the same sets of analogies, ChatGPT looked like ‘one of the gang’—nothing in its performance clearly betrayed its nonhuman identity. Moreover, whatever factors made analogy problems easy or hard for people had a comparable impact on ChatGPT.¹²

ChatGPT underperforms humans on two types of analogy tasks. Since LLMs can’t see or otherwise sense the world, they are worse than us at tasks that require spatial understanding. LLMs also don’t have any long-term memory, so they are unable, as humans are, to retrieve information from long-term memory to form analogies.

Analogical reasoning and the hard problem of consciousness

Many people believe that there is an unbridgeable explanatory gap between our understandings of the physical world and of consciousness, such that the first cannot in principle explain the second. It’s hard to imagine how physical systems operating according to deterministic laws could give rise to something as flexible, protean, and creative as our conscious experience. The concept of a machine seems entirely incompatible with the concept of creativity. This contradiction is expressed in the “ghost in the machine” metaphor, according to which a disembodied mind haunts our biological brain and renders it capable of distinctively human cognition. Because of this contradiction, many are skeptical that AI can ever achieve human levels of creative genius.

The work on analogical reasoning that I have reviewed here might help resolve this contradiction and bridge the explanatory gap. Analogies are one means by which brains achieve creative solutions to problems, and we know that at least one type of machine, AIs, are capable of this kind of creativity. Scientists still don’t understand everything about how analogical thought emerges from the brain, but they have begun to map out how the different subtasks necessary for analogy are distributed in the brain. I continue to hold out hope that neuroscientists will develop an ever more precise and encompassing, and perhaps even complete, understanding of how a deterministic brain can give rise to flexible and spontaneous thinking and action.

If you got something out of this piece, please hit the like button, as that will make it more likely that others will see it.

1. Goswami, U. (2001). Analogical reasoning in children. In D. Gentner, K. J. Holyoak, & B. N. Kokinov (Eds.), The analogical mind: Perspectives from cognitive science (pp. 437-470). MIT Press. The puzzle is from this article. Throughout, I have tried to include links to the full texts of my sources in the body of the essay.

2. Holyoak, K. J. (2025). The Human Edge: Analogy and the Roots of Creative Intelligence (MIT Press), p. 50.

3. Penn, D. C., Holyoak, K. J., & Povinelli, D. J. (2008). Darwin's mistake: Explaining the discontinuity between human and nonhuman minds. Behavioral and Brain Sciences, 31(2), 109-178.

4. Holyoak, Human Edge, p. 87.

5. Diamond, A. (2013). Executive functions. Annual Review of Psychology, 64, 135-168. https://doi.org/10.1146/annurev-psych-113011-143750

6. Starr, A. et al. (2023). Relational thinking: An overlooked component of executive functioning. Developmental Science, 26(3), e13320. https://doi.org/10.1111/desc.13320

7. Holyoak, K. J., & Thagard, P. (1995). Mental leaps: Analogy in creative thought. MIT Press, pp. 186-88.

8. Gentner, D. (2002). Analogy in scientific discovery: The case of Johannes Kepler. In L. Magnani & N. J. Nersessian (Eds.), Model-based reasoning: Science, technology, values (pp.21-39). New York: Kluwer Academic/Plenum Publisher.

9. Dunbar, K. (1997). How scientists think: On-line creativity and conceptual change in science. In T. B. Ward, S. M. Smith, & J. Vaid (Eds.), Creative thought: An investigation of conceptual structures and processes (pp. 461-493). American Psychological Association. https://doi.org/10.1037/10227-017

10. All the information in this section is from Holyoak, Human Edge, pp. 65-76.

11. Webb, T., Holyoak, K. J., & Lu, H. (2023). Emergent analogical reasoning in large language models. Nature Human Behaviour, 7, 1526-1541. https://doi.org/10.1038/s41562-023-01659-w

12. Holyoak, Human Edge, p. 184.