Latent learning is learning that takes place without any obvious reward or motivation and that stays hidden until a reason to use it appears. The term has been part of psychology since the 1930s, when maze experiments with rats showed that animals could quietly absorb information about their surroundings during unrewarded exploration and then deploy that knowledge almost immediately once a reward was introduced. The concept challenged the then-dominant idea that learning requires reinforcement every step of the way, and it continues to reshape how researchers think about curiosity, memory, and the brain’s ability to build internal models of the world.
Where the Idea Came From
The psychologist Edward Tolman is most closely associated with latent learning. In his best-known experiments, rats were allowed to wander through a complex maze for several days with no food waiting at the end. A second group received food at the goal box from the start. The unrewarded rats seemed to learn nothing; they made just as many errors as rats wandering aimlessly. But once food was placed in the goal box, the previously unrewarded rats reduced their errors dramatically within a single session, performing as well as the rats that had been rewarded all along. Something had clearly been learned during those aimless-looking sessions. It just hadn’t shown up in behavior until there was a reason for it to.
Tolman argued that these rats had built what he called a “cognitive map,” a mental representation of the maze’s layout, during their unrewarded exploration. The map was there the whole time; only the performance was missing. This distinction between learning and performance has remained central to psychology for decades, spurred by exactly this kind of evidence that learning can happen even when no visible change in behavior is observed.1PubMed. Learning versus performance: an integrative review
How Textbooks Get It Wrong
If you learned about latent learning in an introductory psychology class, there’s a decent chance some of the details you absorbed were inaccurate. A review of 48 introductory psychology textbooks published between 1948 and 2004 found persistent errors in how the topic was presented. Among the most common: the claim that Tolman’s experiments proved, without any room for debate, that reinforcement is unnecessary for learning; the assertion that all behavioral theories collapsed in the face of these results; and the specific mischaracterization of B. F. Skinner as someone who insisted reinforcement was required for any learning to occur.2Europe PMC. Behaviorism, latent learning, and cognitive maps: needed revisions in introductory psychology textbooks
The reality is messier than those textbooks suggest. Skinner’s brand of behaviorism was not the same as the simpler stimulus-response framework that earlier behaviorists had proposed, and he didn’t argue that reinforcement was necessary for all forms of learning in the way textbooks sometimes imply. Meanwhile, Tolman’s cognitive map was not the only possible explanation for the maze results. Other behaviorist accounts, including ones involving secondary reinforcement or stimulus changes, could partially explain the findings. The experiments were important, but presenting them as a clean knockout blow to behaviorism oversimplifies a complicated and still-evolving debate.3Europe PMC. Behaviorism, latent learning, and cognitive maps: needed revisions in introductory psychology textbooks
Cognitive Maps and Why They Matter
The idea behind latent learning is that, even without a goal in mind, an organism processes and stores information about the structure of its environment. That stored information acts as a cognitive map, a flexible internal model that can later be used to guide behavior when a goal does emerge. This is more than a simple record of “turn left here, turn right there.” Cognitive maps encode relationships between locations, the general layout of a space, and even abstract connections between experiences that didn’t occur at the same time.4PubMed Central. Latent learning, cognitive maps, and curiosity
Recent computational work has explored exactly what kind of spatial information gets encoded during unrewarded exploration and how that information helps when a reward finally shows up. Simulations of latent learning using grid-world environments confirm what Tolman saw: exploration aligned with a future reward location significantly improves learning speed once that reward is introduced, compared to random or no exploration. In other words, the structure of where you’ve wandered matters. If your past exploration happened to cover the terrain near a future goal, you benefit more than if you wandered somewhere irrelevant.5PubMed Central. Accounting for sensitivity of latent learning to behavioral statistics with successor representations
Humans build these maps too, and not just for physical spaces. In one study, people were exposed to the edges of a graph structure in a randomized, piecemeal fashion with no explicit instructions to learn the layout. Afterward, they were able to reason about shortest-path distances across the full graph, assembling a coherent structure from disordered experiences. This is latent learning in a more abstract sense: picking up the hidden architecture of a complex system just by passively encountering its parts.6PubMed Central. Humans can navigate complex graph structures acquired during latent learning
What the Brain Is Doing During Latent Learning
The hippocampus, the brain region most associated with memory and spatial navigation, is central to latent learning. One influential theory reframes the hippocampus not primarily as a spatial mapper but as a sequence-learning engine. Under this view, the mental representation of space is an emergent property of learning sequences of sensory experiences. As you move through an environment, the hippocampus learns compressed representations of those sequential experiences, and the resulting internal models function as cognitive maps suitable for planning, memory consolidation, and abstract reasoning.7PubMed Central. Space is a latent sequence: A theory of the hippocampus
Sleep appears to play a role in cementing latently learned information. Research on hippocampal neurons in mice has identified distinct subpopulations of cells in a region called CA1 that respond differently during and after unrewarded exploration. Some cells are “strongly spatial,” firing when the animal is in a specific location. Others are “weakly spatial,” responding not to particular places but to patterns of activity in other cells. These weakly spatial cells can serve as bridges between distinct locations, effectively knitting together separate spatial memories into a coherent map during sleep.8Cell Reports. Latent learning drives sleep-dependent plasticity in distinct CA1 subpopulations
Neurochemistry matters as well. In zebrafish, dopamine receptors turn out to be critically involved in both acquiring and consolidating latent learning. Drugs that activate dopamine receptors impaired learning and memory in a spatial task, while drugs that block certain dopamine receptor subtypes actually improved performance, with a particularly prominent role for D2-type receptors. These findings suggest that the dopamine system, long associated with reward-driven learning, is also tuned into the quieter, unrewarded kind.9PubMed Central. Dopamine receptors participate in acquisition and consolidation of latent learning of spatial information in zebrafish (Danio rerio)
Curiosity as the Engine
If latent learning doesn’t require reward, what drives it? A growing body of research points to curiosity itself as the motivating force. One framework proposes that the brain is intrinsically motivated to engage in activities where learning is actively happening, essentially maximizing the rate at which it reduces prediction errors about the world. Under this account, you’re drawn to situations that are just beyond your current level of knowledge or skill. If something is too easy (prediction errors already near zero) or too hard (errors aren’t decreasing no matter what you do), curiosity fades. The sweet spot is where your internal model of the world is improving fastest.10Trends in Cognitive Sciences. Bridging information gain and learning progress in curiosity-driven exploration
This framework helps explain why latent learning varies between individuals. A study tracking how people explored virtual rooms found that curiosity before entering a room predicted how extensively participants wandered inside it. But this relationship was moderated by personality traits. People with higher stress tolerance, meaning a greater self-reported ability to cope with the anxiety and uncertainty of new situations, showed a stronger link between curiosity and exploration. In other words, being curious isn’t enough on its own; you also need to feel comfortable enough with uncertainty to act on that curiosity.11Communications Psychology. Curiosity shapes spatial exploration and cognitive map formation in humans
Latent Learning Across the Lifespan
If latent learning depends on implicit, automatic absorption of environmental structure, it’s worth asking whether some ages are better at it than others. The picture is somewhat complicated, because it depends on exactly what kind of learning you’re measuring and how you measure it.
One large study testing implicit sequence learning across ages 4 to 85 found that sensitivity to raw probabilities in a sequence dropped significantly around age 12. Young children were especially good at picking up on the basic statistical patterns in a sequence, a finding that aligns with the intuition that young kids soak up environmental regularities almost effortlessly. The researchers suggested this reflects a developmental shift: before early adolescence, the brain heavily weights raw probability information; after that transition, it favors more complex interpretations of events.12PubMed Central. The best time to acquire new skills: age-related differences in implicit sequence learning across the human lifespan
That doesn’t mean adults are worse at all forms of implicit skill learning, though. Other research using different tasks has found that adolescence and adulthood are actually the most efficient periods for certain kinds of skill learning, with performance improving from childhood into adulthood before declining in old age.13PubMed. Development of different forms of skill learning throughout the lifespan The takeaway is that latent learning isn’t a single monolithic ability. Different components of it, from raw pattern detection to more structured skill acquisition, peak at different life stages.
How Sensory Channels Affect Incidental Pickup
Another factor that shapes how much you learn without trying is the number of senses engaged at once. A study of incidental learning in children found that participants scored significantly higher on tests of information they’d been passively exposed to when that exposure was audiovisual, combining sound and images, compared to auditory-only or visual-only conditions.14PubMed Central. Incidental learning in a multisensory environment across childhood This has practical implications for education and environment design: if you want people, especially children, to absorb information passively, presenting it through multiple channels at once gives it a better chance of sticking, even if nobody is explicitly studying.
Latent Learning in Other Species
Tolman’s original experiments used rats, but latent learning is far from a mammal-only phenomenon. Some of the most striking recent demonstrations come from ants. In one set of experiments, ants were allowed to explore an arena with certain objects in it. Later, when they needed to navigate obstacles to retrieve food, ants that had previously explored the arena made more efficient foraging decisions, using their prior knowledge of object placement to plan better routes. The researchers described this as evidence that ants “latently learn the affordance of their surroundings,” an unexpected cognitive ability for an invertebrate with a brain containing roughly 250,000 neurons.15PubMed Central. Ants combine object affordance with latent learning to make efficient foraging decisions
A separate study went even further, demonstrating that ants learn the routes they travel continuously and can memorize them in a single trial, without reward or punishment, even when those routes are meandering and don’t lead anywhere useful. This is textbook latent learning, but occurring in a creature with no hippocampus and a brain that operates on fundamentally different hardware from a mammal’s. Interestingly, the researchers found that the ants accomplished this without forming map-like representations of space in the way vertebrates are thought to, suggesting that latent learning may rely on different neural mechanisms across the animal kingdom while achieving functionally similar results.16bioRxiv. Latent learning without map-like representation of space in navigating ants
Everyday Examples You Probably Don’t Recognize
Latent learning is happening around you constantly, in ways that rarely get labeled as such. If you’ve ever taken a slightly different route to work and then, weeks later, effortlessly navigated to a new restaurant on that street, you were drawing on a latently learned cognitive map. The same goes for absorbing the layout of a friend’s house during a party and then knowing exactly where the bathroom is the next time you visit, or for picking up the structure of a foreign city’s subway system just by riding it a few times with no intention of memorizing anything.
Less obvious examples extend beyond spatial navigation. Children who overhear conversations in a second language at home may not produce any of that language for years but then learn it remarkably fast when they begin formal study. People who watch cooking shows without any plan to cook are often surprised by how much they already know when they finally try a recipe. In each case, information was encoded passively, without reinforcement, and remained dormant until a goal activated it. The learning was latent; only the performance was delayed.
Latent Learning and Artificial Intelligence
The concept of latent learning has found a second life in machine learning and artificial intelligence. Researchers designing AI systems face a version of the same problem Tolman identified: how do you build agents that can explore and learn about an environment before they know what they’ll be asked to do? In reinforcement learning, the dominant approach for training AI agents, the system learns by receiving rewards for good actions. But just as Tolman’s rats needed to explore before rewards appeared, AI agents perform better on long-horizon tasks if they first build internal models of their environment through unrewarded exploration.
One concrete approach, called latent-space collocation, optimizes trajectories through learned internal models of the environment rather than relying on trial-and-error from scratch. This method improves on earlier techniques for tasks with sparse rewards and long-term goals, situations where the agent needs to plan far ahead rather than just react to immediate feedback.17arXiv. Model-Based Reinforcement Learning via Latent-Space Collocation The parallel to biological latent learning is direct: build a model of the world first, exploit it later when a goal emerges.
Computational models of latent learning in biological organisms are converging on similar ideas. Successor representation models, which encode not just where an agent is but where it’s likely to end up from each state, capture many features of real animal latent learning. These models show that the spatial information encoded during unrewarded exploration directly impacts how effective cognitive maps are at supporting later goal-directed tasks, confirming that the quality and structure of exploration matter, not just its quantity.18PubMed Central. Accounting for sensitivity of latent learning to behavioral statistics with successor representations
Why the Learning-Performance Distinction Still Trips People Up
The single most counterintuitive thing about latent learning is the gap between learning and performance. We tend to assume that if someone can’t demonstrate a skill, they haven’t learned it. Teachers evaluate students through tests. Coaches evaluate athletes through performance. Employers evaluate workers through output. In all of these settings, the assumption is that what you can show is what you know. Latent learning says that assumption is often wrong.
This has real consequences. A student who appears to have learned nothing during an unstructured exploratory lesson may actually have absorbed the layout of a problem space that will make later instruction far more productive. An employee who spent time in a different department and seems to have “wasted” those weeks might suddenly exhibit expertise that only makes sense if they were picking up structure the whole time. The difficulty is that latent learning is, by definition, invisible right up until the moment it becomes visible. That makes it hard to measure, hard to plan for, and easy to dismiss.
The ant studies make this point in a particularly vivid way. If you watched an ant wandering a seemingly pointless route through an arena, you would have no reason to think anything useful was happening. But that single trial of apparently aimless walking was enough to produce measurably better foraging decisions later. The learning was real; only our ability to see it was limited.