What Is Intelligence?

A scientific guide to what we mean by intelligence—in humans, other animals, artificial systems and the study of cognitive ability.

Intelligence is one of the most familiar ideas in everyday life and one of the hardest to define precisely. We use the word for people, animals and machines; sometimes for groups, organisms and other adaptive systems. Yet those uses do not necessarily refer to one identical property. To understand intelligence scientifically, we have to ask what kinds of capacities the word is trying to capture, how those capacities can be measured, and where observation ends and interpretation begins.

What do we mean by intelligence?

Ask ten people what intelligence is and the answers will overlap without being identical. They may mention learning quickly, seeing patterns, reasoning well, solving unfamiliar problems, understanding difficult ideas, remembering, adapting, or knowing what to do when the usual solution fails. Scientific definitions vary for much the same reason. Intelligence is not a single behaviour that can be pointed to directly; it is a concept used to organize a family of capacities and patterns of performance.

Across psychology and artificial-intelligence research, definitions repeatedly return to a few themes: acquiring and using information, learning from experience, reasoning, adapting to changing circumstances, pursuing goals and solving problems. But the boundaries remain contested. A survey by Shane Legg and Marcus Hutter collected more than 70 definitions from psychologists and AI researchers, illustrating both the recurring themes and the lack of a single definition accepted across disciplines (Legg & Hutter, 2007).

That does not make the concept scientifically empty. Many important scientific constructs are defined partly by the patterns they explain and the measurements that relate to them. What matters is being precise about which form of intelligence is under discussion, what evidence is being used, and which claims are empirical findings rather than interpretations.

Three useful definitions show why the concept is both recognizable and difficult to compress into a single sentence.

Psychology. The American Psychological Association defines intelligence in terms of deriving information, learning from experience, adapting to the environment, understanding, and using thought and reason effectively. In this tradition, intelligence is principally a construct for describing and explaining cognitive capacities and individual differences in human functioning.

Cognitive and developmental approaches. Cognitive accounts tend to emphasize the processes that make intelligent behaviour possible: representing information, learning, reasoning, remembering, solving problems, transferring knowledge and adjusting strategies when circumstances change. Piaget’s developmental account, for example, placed adaptation at the centre of intelligence.

Artificial intelligence. In AI, definitions often need to be less human-specific. Legg and Hutter’s influential cross-disciplinary work extracted recurring features from dozens of definitions and developed a formal account centred on an agent’s ability to achieve goals across a wide range of environments.

These definitions emphasize different things, but they overlap around a recognizable core: information must be acquired or represented; experience must be usable; behaviour or reasoning must be adjustable; and the system must be capable of producing effective responses to problems, goals, or changing conditions.

Drawing these traditions together, we will use the following working definition throughout this article: Intelligence is the capacity of a system to acquire, represent and use information in ways that enable learning, reasoning, problem solving and flexible adaptation across changing or unfamiliar conditions.

This is a synthesis, not an official definition adopted by psychology, cognitive science or AI. It is intentionally broad enough to ask comparable questions about humans, other animals and artificial systems without assuming that intelligence must be biological, conscious, verbal or organized like the human brain. It is also restrictive enough to avoid calling every complex or adaptive process intelligent.

For human intelligence specifically, the definition can be made more precise: human intelligence refers to general and specific cognitive capacities that support reasoning, learning, acquiring and using knowledge, solving novel problems and adapting effectively to cognitive demands. These capacities are partly distinct but systematically related.

Intelligence beyond humans

Human intelligence is only one place where the word is used. Comparative cognition studies learning, memory, categorization, problem solving and other cognitive capacities across species. Research on cephalopods is especially instructive because octopuses, cuttlefish and squid possess nervous systems that differ radically from vertebrate brains yet show sophisticated perception, learning, memory and flexible behaviour. This makes them valuable for asking whether comparable cognitive capacities can arise through very different evolutionary routes (Schnell et al., 2021).

The same research also illustrates the need for caution. Flexible or impressive behaviour is not, by itself, proof of a complex mental mechanism. Comparative-cognition experiments are designed to test whether behaviour that looks sophisticated can be explained by simpler learning processes or whether stronger inferences are warranted. Calling a behaviour intelligent is therefore the beginning of an investigation, not the end of one.

Artificial intelligence creates a different boundary problem. A machine can perform tasks that once seemed to require human intelligence—recognizing patterns, producing language, planning, searching large problem spaces or learning from data—without sharing the biological architecture through which human intelligence developed. Formal definitions of machine intelligence therefore often emphasize an agent’s capacity to achieve goals across environments rather than resemblance to the human brain (Legg & Hutter, 2007). Whether such performance implies understanding, consciousness or human-like cognition is a separate question.

Researchers also use terms such as collective intelligence and, more controversially, biological or basal cognition to describe adaptive problem solving at levels ranging from groups of organisms to cellular collectives. These approaches are active research programmes rather than proof that every adaptive process in nature is literally intelligent. Complexity, adaptation, information processing, cognition and intelligence are related concepts, but they should not be treated as synonyms (Levin et al., 2024).

The safest conclusion is therefore broad but modest: intelligence may not be one thing implemented in one way. Different biological and artificial systems can display capacities that scientists describe as intelligent, while the extent to which these capacities share underlying mechanisms remains an empirical and philosophical question.

What psychologists mean by human intelligence

When psychologists study human intelligence, the question is narrower. The focus is usually on individual differences in cognitive abilities: why people differ in how effectively they reason, learn, solve problems, understand complex material and perform across a range of mental tasks.

There is no single sentence that every researcher accepts as the definitive verbal description of intelligence. But there is considerably more agreement about the empirical phenomena that intelligence research must explain. More than a century of psychometric research has shown systematic individual differences in cognitive performance, substantial correlations among different cognitive tests, reliable measurement of many of those differences, and meaningful associations with educational and other life outcomes (Deary, 2012; Deary, Penke, & Johnson, 2010).

This is an important distinction. Disagreement about the best theory of intelligence is not the same as disagreement about whether stable and measurable differences in cognitive performance exist.

How can scientists study intelligence if they cannot see it directly?

Intelligence is not observed directly in the way height is observed with a ruler. Researchers observe what people do: solve a novel reasoning problem, remember information, understand words, manipulate spatial relationships, detect patterns, learn a rule or process information under time constraints.

From repeated measurements across many tasks and many people, researchers infer underlying cognitive abilities. In psychometrics, intelligence and its component abilities are therefore commonly treated as latent constructs: variables inferred from patterns in observed performance rather than objects directly visible inside the brain.

This makes measurement inseparable from interpretation. A test score is evidence about cognitive performance under specified conditions. A well-designed intelligence test can provide reliable and useful evidence about cognitive abilities, but the score is not the ability itself. The distinction becomes especially important when interpreting a single result, comparing scores across tests, or making claims about an individual.

The remarkable pattern: cognitive abilities are related

One of the most replicated observations in intelligence research is known as the positive manifold. People who perform relatively well on one cognitive test tend, on average, to perform relatively well on other cognitive tests too. The correlations are far from perfect, but they are reliably positive across many kinds of cognitive tasks.

Charles Spearman used this pattern in the early twentieth century to propose a general factor, later called g, representing what different cognitive tests have in common. In modern psychometric models, a general factor can summarize part of the shared variation across diverse tests, while broad and narrow abilities account for additional structure.

The positive manifold is an empirical finding. What causes it is a theoretical question. Some accounts interpret g as reflecting a broad general capacity. Other models explain the same pattern through overlapping cognitive processes, sampling, developmental interactions or network dynamics. Mutualism and process-overlap theories are prominent examples of alternatives to a simple single-cause interpretation (van der Maas et al., 2006; Kovacs & Conway, 2016).

That distinction is crucial: the statistical regularity is highly robust; its best causal explanation remains debated.

Is intelligence one thing or many?

The best-supported answer for human intelligence is: both. Cognitive abilities are related strongly enough for a general dimension to be measured, but differentiated enough for broad and specific abilities also to be measured meaningfully.

The evidence does not fit comfortably into a simple choice between “one intelligence” and “many unrelated intelligences.” Human cognitive abilities show both commonality and differentiation.

At a broad level, performance across many cognitive tasks is correlated. At the same time, people can show meaningful relative strengths and weaknesses across domains. Factor-analytic research has repeatedly identified broad abilities such as fluid reasoning, acquired knowledge, visual-spatial processing, memory-related abilities and processing speed, along with narrower skills within them.

Hierarchical models capture both levels at once. Carroll’s three-stratum model and the Cattell–Horn tradition were later synthesized in the Cattell–Horn–Carroll, or CHC, framework, which has become an influential taxonomy for organizing human cognitive abilities (McGrew, 2009). The exact architecture, labels and interpretation of factors continue to be refined; the larger point is that general and specific abilities need not be competing ideas.

A person can therefore be relatively strong overall and still have a distinctive cognitive profile. Conversely, two people with similar overall scores can differ in the pattern of abilities that contributed to those scores.

The major dimensions of human cognitive ability

Different theories divide cognitive ability in somewhat different ways, so no short list should be mistaken for a final map. Still, several broad domains recur across contemporary measurement frameworks.

Fluid reasoning concerns solving unfamiliar problems and identifying relations when prior knowledge alone is not enough. Crystallized or acquired knowledge concerns information and language learned through education and experience. Visual-spatial abilities involve analyzing, transforming or reasoning with visual and spatial information. Memory-related abilities concern learning, retaining, retrieving or actively maintaining information. Processing speed concerns the efficiency with which relatively simple cognitive operations can be carried out under particular conditions.

These domains are related but not interchangeable. Their relative importance also depends on the task. A vocabulary problem, a novel matrix-reasoning problem and a timed symbol-comparison task all require cognition, but they place different demands on it. This is one reason a single overall score can be informative without exhausting what there is to know about a person’s cognitive performance.

Intelligence, knowledge, achievement and performance are not the same thing

Everyday language often blends together concepts that psychology tries to distinguish. Intelligence is not identical to how much someone knows. Knowledge accumulates through learning and experience, although the ease and speed with which people acquire knowledge can be related to cognitive ability.

Intelligence is also not identical to educational achievement. School grades and achievement tests reflect learning, opportunity, instruction, motivation, subject-specific knowledge and other influences as well as cognitive ability. A large meta-analysis found a substantial association between intelligence-test scores and school grades, but a substantial association is not identity: the two constructs remain distinguishable (Roth et al., 2015).

Expertise is another case. An expert may outperform a generally bright novice because years of domain-specific knowledge and pattern recognition radically change the problem being solved. Creativity, wisdom, personality and practical judgement likewise intersect with intellectual functioning without simply being alternative names for psychometric intelligence.

Finally, intelligence should be distinguished from performance on a particular occasion. Performance is an event: what happened when a person attempted a specific task at a specific time. Ability is an inference about capacities that help explain patterns of performance across occasions.

Where does IQ fit?

IQ is not intelligence itself. It is a standardized score derived from performance across selected cognitive tasks and designed to estimate aspects of cognitive ability relative to an appropriate comparison group.

Modern intelligence tests are measurement instruments designed to estimate latent cognitive abilities from performance across a standardized battery of tasks. Their IQ scores are norm-referenced: they locate a person’s test performance relative to an appropriate comparison group. Multiple tasks are used precisely because intelligence is not inferred from a single puzzle or observation.

A good intelligence test can provide reliable, predictive and clinically or educationally useful information. But “IQ” and “intelligence” are not interchangeable concepts. Different tests sample somewhat different tasks and abilities; every score contains measurement error; interpretation depends on the purpose of testing and the population for which the test was designed.

The useful formulation is therefore simple: an IQ score is evidence about cognitive ability, not intelligence made visible as a number.

Is intelligence stable—or can it change?

Intelligence is both stable and changeable, depending on what kind of change is being discussed. Cognitive abilities develop and age, measured scores can change, and moment-to-moment performance fluctuates; at the same time, people’s relative standing in cognitive ability shows substantial long-term stability.

The question “Can intelligence change?” bundles together several different questions. Do cognitive abilities develop across childhood? Yes. Do different abilities follow different trajectories across adulthood? Yes. Can test scores change? Yes. Are individual differences also stable enough to be meaningful? Also yes.

Longitudinal research shows substantial rank-order stability: people’s relative standing in cognitive ability tends to persist, increasingly so from later childhood onward, even while average levels of particular abilities change with development and ageing. An unusually long follow-up in the Lothian Birth Cohort found a sizeable association between intelligence measured at age 11 and the same test administered at age 90 (Deary, Pattie, & Starr, 2013).

Stability is therefore not immutability. Development, education, health, ageing, practice, measurement conditions and other influences can affect cognitive scores and cognitive functioning. A meta-analysis of 42 quasi-experimental datasets involving more than 600,000 participants found consistent evidence that an additional year of education was associated with gains of roughly 1 to 5 IQ points, depending on the design and context (Ritchie & Tucker-Drob, 2018). This is strong evidence that measured cognitive abilities are environmentally responsive, although it does not mean that every intervention produces broad gains in general intelligence or that every score change reflects an equivalent change in a single underlying quantity called intelligence.

Intelligence is not the same as intelligent performance

No single performance is identical to intelligence. Cognitive ability contributes to performance, but what a person does on a particular occasion also depends on the task, relevant knowledge, familiarity, current state, available time and other conditions.

A person’s underlying cognitive abilities contribute to what they can do, but observed performance is produced in a particular situation. The demands of the task matter. So do relevant knowledge, familiarity, available time and the processes the task recruits.

Current state can matter as well. Acute stress has been shown in meta-analytic work to impair working memory and cognitive flexibility on average, although effects differ across executive functions and conditions. Sleep loss reliably affects vigilance and can impair aspects of attention and executive functioning. These findings do not imply that a tired or stressed person has suddenly acquired a different enduring intelligence. They show that the performance through which cognitive ability is observed can be state-sensitive (Shields, Sazma, & Yonelinas, 2016; Alhola & Polo-Kantola, 2007).

The same caution applies in the opposite direction. A strong performance on one familiar task does not establish broad intellectual ability, and a poor performance on one demanding occasion does not by itself establish a global deficit. Scientific interpretation depends on repeated patterns, appropriate comparison standards, measurement quality and context.

This distinction is one of the reasons intelligence is best understood as a construct inferred from evidence rather than as a label attached to isolated successes or failures.

What does intelligence predict—and what remains unresolved?

Yes, intelligence predicts consequential outcomes, especially learning and educational achievement, but prediction is probabilistic rather than deterministic. Cognitive ability is one contributor among many, and the strength of its association varies by outcome, population and method.

Intelligence matters scientifically in part because measures of cognitive ability predict outcomes beyond the test itself. The strongest and most consistent associations are with learning and educational achievement. In the 2015 meta-analysis of intelligence and school grades, 240 independent samples and more than 105,000 participants produced a corrected population correlation of .54, with substantial variation across subjects, grade levels and test types (Roth et al., 2015).

Associations also exist with occupational performance and with some health and longevity outcomes, but their magnitude and interpretation require care. Recent re-analysis of the job-performance literature has produced lower estimates than the classic figures often repeated in textbooks, underscoring how conclusions can change when datasets, correction methods and historical samples are reconsidered (Demeke et al., 2026). Observational associations with health or mortality likewise do not imply a simple direct causal path from intelligence to health (Calvin et al., 2011).

Important theoretical questions remain open. Researchers continue to debate what produces the positive manifold, how best to interpret g, how cognitive abilities develop and interact, how intelligence should be compared across species or substrates, and which features of artificial systems deserve the same conceptual label used for biological cognition.

So what is intelligence? In this article, we have defined intelligence as the capacity of a system to acquire, represent and use information in ways that enable learning, reasoning, problem solving and flexible adaptation across changing or unfamiliar conditions. That is a working synthesis rather than a universally accepted scientific definition. Across species and artificial systems, these capacities may be implemented in very different ways. In humans, intelligence has both general and specific structure: cognitive abilities are systematically related, yet meaningful differences among broad and specific abilities remain.

Intelligence can be measured indirectly through carefully designed observations of performance. Those measurements are useful, sometimes strongly predictive, and scientifically consequential. But no single score, task, behaviour or theory is identical to intelligence itself. The enduring scientific task is to understand the capacities behind intelligent behaviour, the structure of their differences, and the conditions under which our measurements justify the conclusions we draw.

References

• Alhola, P., & Polo-Kantola, P. (2007). Sleep deprivation: Impact on cognitive performance. Neuropsychiatric Disease and Treatment, 3(5), 553–567.
• Calvin, C. M., Deary, I. J., Fenton, C., Roberts, B. A., Der, G., Leckenby, N., & Batty, G. D. (2011). Intelligence in youth and all-cause mortality: Systematic review with meta-analysis. International Journal of Epidemiology, 40(3), 626–644. DOI
• McMillen, P., & Levin, M. (2024). Collective intelligence: A unifying concept for integrating biology across scales and substrates. Communications Biology, 7, 378. DOI
• Deary, I. J., Pattie, A., & Starr, J. M. (2013). The stability of intelligence from age 11 to age 90 years: The Lothian Birth Cohort of 1921. Psychological Science, 24(12), 2361–2368. DOI
• Deary, I. J., Penke, L., & Johnson, W. (2010). The neuroscience of human intelligence differences. Nature Reviews Neuroscience, 11, 201–211. DOI
• Demeke, S., Sackett, P. R., Nye, C. D., Li, Y., et al. (2026). Is general mental ability still the best predictor of job performance? Integrating contemporary meta-analytic evidence. Journal of Business and Psychology. DOI
• Kovacs, K., & Conway, A. R. A. (2016). Process Overlap Theory: A unified account of the general factor of intelligence. Psychological Inquiry, 27(3), 151–177. DOI
• Legg, S., & Hutter, M. (2007). A collection of definitions of intelligence. In Advances in Artificial General Intelligence: Concepts, Architectures and Algorithms. DOI
• Legg, S., & Hutter, M. (2007). Universal intelligence: A definition of machine intelligence. Minds and Machines, 17, 391–444. DOI
• McGrew, K. S. (2009). CHC theory and the human cognitive abilities project: Standing on the shoulders of the giants of psychometric intelligence research. Intelligence, 37(1), 1–10. DOI
• Roth, B., Becker, N., Romeyke, S., Schäfer, S., Domnick, F., & Spinath, F. M. (2015). Intelligence and school grades: A meta-analysis. Intelligence, 53, 118–137. DOI
• Schnell, A. K., Amodio, P., Boeckle, M., & Clayton, N. S. (2021). How intelligent is a cephalopod? Lessons from comparative cognition. Biological Reviews, 96(1), 162–178.
• Shields, G. S., Sazma, M. A., & Yonelinas, A. P. (2016). The effects of acute stress on core executive functions: A meta-analysis and comparison with cortisol. Neuroscience & Biobehavioral Reviews, 68, 651–668. DOI
• van der Maas, H. L. J., Dolan, C. V., Grasman, R. P. P. P., Wicherts, J. M., Huizenga, H. M., & Raijmakers, M. E. J. (2006). A dynamical model of general intelligence: The positive manifold of intelligence by mutualism. Psychological Review, 113(4), 842–861. DOI
• American Psychological Association. Piagetian development materials: intelligence as adaptation. https://www.apa.org/ed/precollege/topss/lessons/life-development.pdf
• Ritchie, S. J., & Tucker-Drob, E. M. (2018). How much does education improve intelligence? A meta-analysis. Psychological Science, 29(8), 1358–1369. DOI