RESEARCH DOSSIER
Creative Ideation and Divergent Thinking
Abstract. Creative ideation concerns the generation of candidate alternatives capable of becoming original and contextually appropriate responses. This fundamental research monograph examines the history of divergent-thinking research, associative and executive explanations, cognitive search dynamics, experimental tasks, fluency and originality scores, semantic networks, psychometric validity, developmental and cultural constraints, neural investigations, educational interventions, and comparative human–artificial generation. The evidence base integrates historically foundational theory, controlled behavioral research, original computational experiments, neurocognitive studies, psychometric investigations, and contemporary meta-analyses and methodological reviews. It differentiates divergent-thinking test performance from creative potential, evaluation, and achieved products; identifies variation in scoring procedures as a major threat to comparability; and examines divergent conclusions from recent human–language-model comparisons. The synthesis supports a multicomponent account of candidate retrieval, regulated search, representational transformation, and context-sensitive selection while specifying unresolved problems of causal inference, measurement invariance, and cross-domain generalization.
Keywords: creative ideation; divergent thinking; creativity research; originality; idea generation; Alternative Uses Task; semantic distance; associative search; cognitive flexibility; executive functions; serial order effect; creativity measurement; construct validity; cognitive neuroscience; human–artificial creativity
The Scientific Object of Creative Ideation
Creative ideation is the generation of candidate representations that may become original and appropriate responses to an open-ended problem. Divergent thinking is the capacity and set of operations through which a person produces multiple, varied, and potentially unusual responses under specified task conditions. The two concepts overlap in research on possibilities, associations, and alternatives, yet their empirical identities depend on what the investigator observes: response sequences, semantic relations, contextual suitability, or products selected for further development. The enduring definition of creativity requires originality and effectiveness in a relevant context Runco & Jaeger (2012). Consequently, producing many improbable words, by itself, provides evidence about divergent association rather than proof of accomplished creativity. The present monograph treats ideation as a psychological process, divergent thinking as a family of operational tasks and constructs, and creative achievement as a distinct outcome requiring independent evidence.
The research object encompasses the construction of alternatives, the retrieval and recombination of concepts, switching between semantic categories, executive monitoring, changes in search across time, and the evaluation pressures that accompany idea production. It includes the effects of instructions, task materials, individual knowledge, cultural and linguistic contexts, and measurement methods on the ideas observers record. Divergent-thinking test scores offer indicators of creative potential, but their predictive meaning varies with the external criterion Runco & Acar (2012). The central scientific questions are how useful variations arise, what drives the transition from conventional to remote possibilities, how search becomes organized, and how well laboratory measures characterize enduring capacities. The boundary intentionally excludes comprehensive theories of artistic recognition, mature product evaluation, and the entire life course of creative accomplishment.
This is a critical narrative synthesis of accessible original experiments, scholarly theory, psychometric investigations, computational research, and explicitly identified reviews. Primary-source evidence receives priority when experimental manipulations or numerical findings are discussed. A historical work anchors the origin of a hypothesis; a controlled study supports claims within its experimental conditions; an observational correlation supports covariation; and a meta-analysis evaluates heterogeneous effects under its own inclusion criteria. A literature-based synthesis cannot replace new prospective experiments, and it is not designated a systematic review because no exhaustive registered search, screening protocol, or independent dual extraction was conducted. This evidence framework permits direct comparison of explanations while keeping each empirical claim proportional to its method.
Originality, Appropriateness, and the Space of Possible Ideas
A candidate idea becomes psychologically interesting when it departs meaningfully from a context's anticipated responses while remaining intelligible or functional in that context. Originality can mean statistical rarity in a sample, distance from common semantic associations, novelty relative to a person's prior repertoire, or value recognized by domain-informed judges. These meanings can diverge. A rare response may be nonsensical, an unfamiliar solution may be routine to experts, and a conventional idea in one field may be innovative in another. The originality-and-effectiveness formulation Runco & Jaeger (2012) supplies an important evaluative condition, but effectiveness during an ideation task can be provisional: early ideas are often incomplete and later require testing, revision, and feasibility checks. Researchers must therefore announce whether they assess unusualness, plausibility, usefulness, judged creativity, or several separately.
Ideational fluency measures how many permissible responses are generated. Flexibility describes variety across conceptual categories or strategies. Originality assesses uncommonness or judged inventiveness. Elaboration concerns the detail added to an idea. These traditional dimensions were operationalized through different families of creativity tasks following the psychometric program associated with Guilford Guilford (1950). They should be treated as outcomes of an administration-and-scoring procedure rather than interchangeable facets with invariant meaning. Participants who produce many responses have more opportunities to generate a rare one, so aggregate originality can rise simply because more ideas were attempted. Time limits, instructions, duplicate-response rules, and the number of presented objects alter score distributions and therefore influence apparent individual differences.
Creativity research thus studies an organized possibility space: the person's remembered concepts, perceived problem constraints, available transformations, and anticipated standards of success. The breadth of that space is partly determined by knowledge and language, while the route through it depends on search operations. A well-known object may evoke familiar affordances first and remote analogies later, and the same individual may excel when asked for surprising uses but struggle when asked to choose an economical design. The interpretive challenge is to infer latent operations from responses without treating every unusual product as the product of a single psychological faculty. Response quality, response diversity, and functional fit need evidence from distinct, transparent scoring rules.
From Guilford to the Experimental Program
J. P. Guilford's 1950 presidential address published in American Psychologist brought creativity into the agenda of differential psychology and proposed studying abilities that ordinary intelligence tests inadequately captured Guilford (1950). His factor-analytic orientation encouraged researchers to construct tests that elicited multiple responses rather than a single predetermined correct answer. The historical achievement was methodological: creative performance could be decomposed into observable kinds of responses and subjected to experimental and psychometric investigation. Guilford's program did not establish a permanent partition between intelligence and creativity, and its original factor claims should be examined within the measurement assumptions of their period. What endured was the requirement to sample a broader repertoire of thinking than traditional closed-response testing usually demanded.
Alternative-uses and consequences tasks became especially influential because an ordinary stimulus can elicit diverse outcomes without requiring expert artistic training. The participant might propose possible uses for a commonplace object, while an investigator analyzes the resulting list. These tasks were attractive for repeated administration, group comparison, and controlled manipulation. Their apparent simplicity also created a theoretical problem: the recorded number and quality of answers are jointly shaped by the person's capabilities, willingness to persist, familiarity with the object, comprehension of instructions, and the scoring rules used by researchers. Runco and Acar's later reassessment Runco & Acar (2012) makes explicit that divergent thinking should be construed as a useful indicator of creative potential with criterion-dependent predictive value.
The history of experimental ideation has proceeded from factor-analytic and response-count models toward explanations of semantic retrieval, executive search, neural coordination, and computational scoring. A major recent systematic quantitative review of divergent-thinking assessment identified 451 eligible studies from 396 articles after screening 2,066 publications dated from 1957 through 2022 Saretzki et al. (2026). Its account of extensive procedural variation shows that the historical record cannot be read as though every study used the same test. The review documents a move from instructions emphasizing the number of ideas toward originality-oriented instructions and quality-sensitive scoring. Historical synthesis consequently requires reconstructing the exact task before comparing reported abilities or trends.
Associative Explanations and Remote Connections
Sarnoff Mednick proposed an associative account in which creative thinking involves bringing previously remote associative elements into useful combinations Mednick (1962). The formulation shifted attention from the quantity of generated material to the structure of connections among concepts. A person's associative hierarchy could facilitate familiar dominant responses or permit less immediately available connections, depending on experience and retrieval conditions. In contemporary language, a search process navigates representations whose accessibility is graded rather than uniform. Remote association can produce an unusual candidate because two distant domains share a relevant relation, but the association still requires contextual interpretation and often subsequent filtering.
Semantic-network investigations have translated parts of this idea into measurable structural hypotheses. Kenett, Anaki, and Faust (2014) estimated networks from free-association data and compared groups defined by creativity measures Kenett et al. (2014). Their results were consistent with differences in semantic-network organization, offering an empirical bridge between historical association theory and computational analysis. Yet such comparisons cannot establish that the observed network form caused creative performance, because experience, vocabulary, response strategy, and grouping procedures could affect both network estimates and test scores. Associations reconstructed from participant responses are model-dependent samples of semantic access, not direct images of fixed cerebral representations.
A strong associative account must specify how remote candidates are found without simultaneously making task-relevant selection inexplicable. Random combinations often have low utility, and the most surprising answer can still violate a task's constraints. Associative availability, controlled retrieval, analogy, and evaluation therefore constitute different explanatory components even when they operate together. The meaningful distinction is between access to alternative semantic materials and the operations through which those materials are shaped for a purpose. The available evidence supports semantic association as a substantial research dimension while leaving unresolved whether broad semantic networks reflect learned knowledge, habitual search policy, or more general flexibility.
Generative Cognition and Representational Transformation
The creative cognition tradition developed a process-oriented account of how people construct and then explore provisional mental forms. Finke, Smith, and Ward's 1992 book Creative Cognition organized research around the idea that initial representations can be generated and subsequently interpreted, elaborated, and constrained Finke et al. (1992). This approach helps explain why an initially ambiguous or incomplete idea can acquire usefulness through exploration. The scientific point is that a response need not emerge as a fully articulated solution. Mental synthesis, conceptual combination, analogy, and changes in perspective can create intermediate representations that become more meaningful when their properties are inspected.
In divergent-thinking experiments, the distinction between generation and exploration is visible when participants alternate between producing rough variants and checking whether the variants are genuinely new, interpretable, and responsive to the prompt. A process model predicts that the search history matters: newly discovered features can open further directions, and earlier responses can constrain subsequent ones through repetition avoidance. The model also clarifies why an immediate originality rating may understate the promise of a preliminary representation. However, any particular staged account remains a hypothesis about cognitive operations until discriminating process data are available. Verbal report, response timing, semantic changes, and experimental constraints are complementary methods of tracing stages; none should be equated with the entire mechanism.
Creative cognition is sometimes described through broad metaphors of unconstrained imagination, but useful experimental explanations require specified operations. A participant may mentally combine features from unrelated objects, reinterpret a tool's function, change the level of abstraction, or borrow a relational pattern from another domain. These transformations predict different response structures and different error modes. A claim that a brain network or linguistic model 'generates creativity' is therefore insufficient without demonstrating a connection between its operations and novel, context-appropriate products. The research frontier lies in connecting representational transformations to measurable trajectories of individual ideas rather than treating generation as an opaque event.
Competing Accounts and Their Testable Predictions
Associative theories emphasize which candidate connections are available and how easily they can be retrieved; executive-control theories emphasize maintaining goals, suppressing repetitions, testing task requirements, and navigating away from dominant associations. Generative-cognition accounts describe how provisional structures are constructed and explored. These approaches propose overlapping mechanisms but place different explanatory weight on semantic organization, controlled attention, and representational change. Their predictions can be separated experimentally: altering a participant's knowledge changes the accessible candidate pool; varying inhibition or working-memory demands changes the capacity to manage search; introducing new representational constraints may change the kinds of transformations attempted. The results of one manipulation ordinarily constrain several explanations rather than proving one exclusive theory.
An adequate theory should explain ordinary first responses, the decline in response rate across a task, the later appearance of unusual suggestions, and the sometimes weak connection between laboratory originality and recognized real-world achievement. It should also account for variations induced by task instructions without reclassifying the same person as inherently more or less creative. Empirical studies of strategy use and fluid intelligence Nusbaum & Silvia (2011), executive processes Benedek et al. (2014), and individual semantic networks Kenett et al. (2014) furnish complementary evidence for precisely these questions. Their convergence supports a multicomponent account of ideation while also exposing limits in causal identification when people self-select strategies or when cognitive traits are inferred from correlated tests.
The theoretical synthesis adopted here is deliberately functional: candidate retrieval provides materials, controlled search organizes their exploration, representational operations form novel combinations, and task-contingent evaluation shapes what is retained or reported. This is an integrative reading of the literature, not the announcement of a new validated theory. Distinguishing those functions makes the evidence auditable and opens decisive hypotheses about search trajectories, construct validity, cognitive-load effects, and transfer across domains. Later chapters test these explanatory possibilities against observable behavior, psychometric methods, neural investigations, and contemporary human–artificial comparisons.
Divergent Thinking Tasks and Experimental Paradigms
The Alternative Uses Task asks respondents to find possible uses for a familiar object beyond its ordinary role. Its scientific value comes from manipulating a common object while observing how participants expand or transform functional affordances. Variants differ in object choice, instructions, time allotment, whether responses are spoken or written, and the number of tasks combined. Those choices affect response quantity and quality. The task samples verbalized, object-centered ideation under laboratory constraints; it is not a universal assay of creative capacity. The review by Saretzki, Forthmann, and Benedek (2026) documents substantial heterogeneity in these details across published assessments Saretzki et al. (2026), making exact procedural reporting indispensable whenever results are compared or replicated.
Other paradigms ask for hypothetical consequences, improvements of objects, unusual questions, remote conceptual associations, or words as semantically unrelated as possible. Each method recruits overlapping yet nonidentical operations: conceptual fluency, abstraction, scenario generation, or dissimilarity maximization. Olson and colleagues (2021) introduced the Divergent Association Task, in which participants generate unrelated nouns and semantic distance among responses is calculated computationally Olson et al. (2021). That task reduces dependence on subjective judgments in one scoring stage, but it changes the target behavior from proposing functional uses to selecting unrelated verbal concepts. Investigators should therefore determine whether different instruments converge because they share a psychological capacity or because they reward correlated language skills and task strategies.
Task selection should follow the research question. Investigating search dynamics calls for response-by-response time stamps and semantic coding, while studying creative appropriateness requires raters or task-specific criteria. Testing instruction effects requires randomized conditions and identical stimuli; assessing stable individual differences requires enough tasks and observations to estimate reliability. A single object presented under one short deadline provides a particularly narrow sample. Repeated tasks improve coverage but introduce fatigue, practice, and content knowledge as additional variables. A design that carefully specifies its task, scoring system, outcome, and inference target gives more usable scientific evidence than a larger set of vaguely described creativity exercises.
Fluency and the Opportunity to Be Original
The number of ideas generated is among the most accessible measures in divergent-thinking research. Fluency is reliable to count under clear exclusion rules, yet it changes the opportunity structure for scoring originality: someone who responds twenty times can have more uncommon answers than someone who responds five times simply because the former attempted more candidates. This is a mathematical and measurement confound rather than direct proof of a stronger ability. Runco and Acar (2012) stressed that divergent-thinking outcomes must be assessed relative to particular criteria Runco & Acar (2012). In practice, this means distinguishing the probability that a randomly sampled response is original from the total number of original responses produced over a fixed interval.
Originality can be aggregated as the sum of scores, the mean score, the score of the best response, or the mean of a specified number of top responses. These procedures ask different questions. Summation captures both productivity and response quality; averaging estimates typical response quality but can penalize productive explorers who include routine possibilities; best-response scoring estimates peak performance but is sensitive to the number of attempts. A fixed top-response scheme permits comparability only when participants have meaningful opportunities to supply enough candidates. The 2026 quantitative assessment review identified increasing attention to originality-focused instructions and fluency-adjusted aggregation Saretzki et al. (2026). Reporting several defensible statistics may illuminate differences that a single total score obscures.
The strongest design makes response opportunities explicit and separates fluency from the target outcome in analysis. When researchers study originality as a function of serial position, they should track response count, elapsed time, and the probability of stopping. When comparing groups, they should check whether one group gives longer or more detailed responses that score more favorably under rating instructions. Statistical adjustment for fluency is not automatically a correction, because fluency may be part of the process a theory aims to explain. The appropriate choice depends on whether the hypothesis concerns the capacity to produce rare ideas at all, the average inventiveness of accessible ideas, or the performance of one's selected best candidate.
Instructions, Time Limits, and Task Demands
Ideation is sensitive to the objective participants believe they must optimize. Instructions that reward the largest number of responses invite rapid production of familiar candidates, whereas requests for particularly creative or original responses encourage additional retrieval, comparison, and selection. These instruction sets may produce different rank orders among participants because they activate different strategy policies and motivational expectations. The historical review of assessment practices Saretzki et al. (2026) found both fluency-centered and originality-centered instructions, along with frequently incomplete reporting. An exact replication therefore requires the verbatim task prompt, language, permitted examples, task duration, and information about how respondents were told their answers would be scored.
Time constraints interact with these instructions. A short interval emphasizes immediate availability and efficient response production; a longer interval permits category switching, renewed search, and more distant combinations, while also increasing fatigue and opportunities for self-correction. The task's natural unit is not only total elapsed time but the evolving response trajectory: latency to first idea, pauses between ideas, and the content of successive answers. The same individual may appear highly fluent in a brief task and highly original during later phases of a longer one. Without reporting temporal conditions, claims about a stable creativity difference can inadvertently describe a comparison of different search windows.
Experimental manipulation of constraints permits causal inference about task performance, provided groups are otherwise comparable and instructions are implemented consistently. It does not automatically establish enduring enhancement of a person's general creativity. Participants may comply with what they infer the experimenter values, including the production of surprising but impractical responses. Strong studies include comprehension checks, predeclared scoring criteria, blinded raters when feasible, and analyses of both task-compliant responses and their quality. These procedures convert apparent differences in creativity into interpretable evidence about how goal representations and environmental conditions shape creative ideation.
Strategies of Semantic Search and Category Switching
Generating uses for an object often involves clustering responses within a conceptual category and then switching to another category when local options are depleted. An object can evoke construction materials, decoration, tools, symbolic gestures, or mechanical parts, and movement across these clusters may expand the set of reachable responses. Strategy research has challenged a purely automatic-association account. Nusbaum and Silvia (2011) connected divergent-thinking performance to fluid intelligence, executive processes, and the deployment of strategies Nusbaum & Silvia (2011). The evidence encourages treating creative ideation as guided search through associations rather than as an unrestricted cascade. However, self-reported strategies and latent cognitive traits demand separate measurement and careful controls.
Category switching has several operational meanings. A coder may classify consecutive answers into semantic groups, a computational model may estimate abrupt changes in embedding distance, or a participant may report deliberately changing perspectives. Each method can yield a different estimate of flexibility. Human coding adds interpretive expertise but requires reliable category definitions; semantic embeddings provide scalable numeric estimates but inherit corpus biases. Moreover, an apparent jump between topics may reflect a change in vocabulary rather than a useful transformation of the problem. A claim that switching causes originality requires an experimental intervention or a model that accounts convincingly for prior search difficulty and time-on-task.
A refined process design records what participants do between outputs. Think-aloud protocols can expose explicit changes in strategy but may modify thinking through verbalization. Response sequences with time stamps retain relatively natural performance but require latent-process inference. Manipulations that suggest switching after a fixed interval provide more direct tests, though suggestion can induce demand effects and change motivation. The combined evidence makes category structure an essential part of divergent-thinking research while reminding researchers that response labels are observations and search mechanisms are explanations that must be tested.
Serial Position, Persistence, and Search Trajectories
A common behavioral pattern in divergent-thinking tasks is that early responses are comparatively accessible while later responses become less frequent and can be more original. This pattern is called the serial order effect. Wang and colleagues' 2017 study used an alternative-uses paradigm with electroencephalography and separate executive-function tasks, finding that later originality differed according to individual shifting ability Wang et al. (2017). The result illuminates a relation between search stage and cognitive flexibility in the studied sample. It does not establish that all late answers are superior or that every participant follows the same temporal path, because the study's design and scoring conditions constrain generalization.
An increasing originality trend can arise through several processes. Common associations may be exhausted first, participants may deliberately reject repeated ideas, the task may stimulate new categories, and additional time may permit remote retrieval. These explanations make different predictions about pauses, semantic distance, and the cost of switching. If a participant stops early, the observed late stage represents only those willing or able to persist; comparing late responders with all early responders introduces selection. Modeling the hazard of stopping alongside response quality can therefore be necessary. A study that reports originality only by ordinal position risks overlooking unequal exposure to later positions and correlated differences in motivation.
The serial-order question highlights why divergent ideation is a dynamic process rather than a single test total. Investigators can estimate time-varying response rates, conditional originality, changing semantic distances, and between-person trajectories. Longitudinal observations across different task prompts can test whether a person's search profile is stable. Theoretical distinctions between automatic retrieval and controlled exploration become empirically productive when they predict where performance changes occur. These temporal observations provide a bridge between classic association theory and newer cognitive-control accounts without presuming a universal two-stage process.
Executive Control, Intelligence, and Deliberate Search
Executive functions help maintain task goals, resist immediately repeated responses, update working representations, and regulate shifts between candidate strategies. Benedek and colleagues (2014) examined their common and differential involvement in intelligence and creativity, finding relevant roles for updating and inhibition in the investigated constructs Benedek et al. (2014). Nusbaum and Silvia (2011) likewise connected fluid intelligence and strategy use to divergent-thinking performance Nusbaum & Silvia (2011). These findings complicate the older intuition that generating unusual ideas necessarily requires the suspension of all control. In open-ended tasks, control can preserve the problem's constraints and direct search toward alternatives whose unusualness remains meaningful.
The relationship between intelligence and creative ideation is neither identity nor complete independence. Both may draw on knowledge, attention, cognitive flexibility, and effective task comprehension, while the scoring of creative originality adds criteria not ordinarily present in intelligence testing. Individual differences in verbal fluency can inflate apparent commonality between measures. A meta-analysis by Kim (2008) found modest associations of creative achievement with divergent-thinking tests and with IQ, with values varying across tests and outcome domains Kim (2008). These correlations cannot establish a fixed causal path from intelligence to creative accomplishment. They indicate that the predictive relation is real enough to investigate but too limited and heterogeneous to serve as a definition.
Strong causal tests distinguish an individual's executive resources from how those resources are allocated in a particular task. Dual-task manipulations, randomized strategy instructions, and experimental variations in rule complexity may isolate effects of control demands. At the same time, interventions can disrupt general effort or task comprehension, and those possibilities need measurement. Intelligence is often a broad composite while divergent thinking is represented by one timed verbal task; comparisons may therefore mix different levels of construct aggregation. A mature theory predicts when control supports originality, when it prematurely narrows the search, and how task goals determine which effect predominates.
Semantic Distance and Computational Representations
Semantic-distance methods quantify relationships among words or texts using distributional models trained on large corpora. An unusual use for an object may be linguistically distant from the object's ordinary associations, and researchers can estimate this distance without asking every judge to rate every response. Beaty and Johnson (2021) introduced SemDis as an open platform for calculating semantic distance in creativity assessment Beaty & Johnson (2021). This work helps make scoring reproducible at scale and encourages comparison across models. Nevertheless, semantic distance is a property of an algorithm, training corpus, preprocessing method, and chosen texts; it is not a context-free measure of human originality.
Distance captures one dimension of novelty but cannot establish effectiveness. An unrelated response may have a high embedding distance while lacking any workable application, and a subtle improvement within an expert field may have low distance while being valuable. Semantic estimates also change with polysemy, language, domain-specific jargon, and the distribution of words in the underlying corpus. Researchers must report the exact embedding model, training resources, input processing, aggregation, and handling of invalid responses. These requirements become especially important when comparing human and artificial responses, because the same language models may influence both the response distribution and the instrument used for evaluation.
The Divergent Association Task proposed by Olson and colleagues (2021) provides a clear illustration of both opportunity and limitation Olson et al. (2021). Participants named unrelated nouns; the investigators calculated pairwise semantic distances and found associations with established divergent-thinking measures across a large overall sample. The task offers efficient, scalable measurement of one kind of verbal divergence, while its relation to creative outputs remains a matter of validation rather than definition. An assessment of creative ideation should consider whether increased distance predicts judged originality, functional relevance, and behavior beyond existing vocabulary and intelligence measures.
The Divergent Association Task as an Empirical Case
Olson and colleagues (2021) reported research involving 8,914 participants, including an international component with respondents from 98 countries Olson et al. (2021). Their basic task asked people to supply ten nouns as different in meaning and use as possible, and their scoring procedure used semantic distances among selected valid responses. The research found associations with the Alternative Uses Task and other creativity-related measures, supporting convergent validity for the proposed indicator. Its scale and published methodological explanation make it influential in computational approaches to divergent thinking. The result is best interpreted as evidence that relational divergence among words tracks aspects of creativity-test performance, with predictive power dependent on the criterion and sampled population.
Several questions remain when a brief noun-generation task is transported to new languages or settings. Word frequencies differ among corpora; morphological complexity and cultural associations can change computed similarity; bilingual participants may search across language boundaries; and an automated scoring system may be more reliable for well-represented vocabularies. Even when group mean scores appear comparable, measurement invariance must be evaluated rather than assumed. The task also minimizes the explicit need to propose a feasible solution, so it samples a narrower behavior than real-world invention. Cross-national coverage of a participant pool does not by itself demonstrate that the instrument has equivalent construct meaning in every national or language group.
Theoretical interest in the task lies in its capacity to isolate one portion of the associative-search process. If participants can select semantically remote concepts under a simple instruction, this suggests accessible flexibility in choosing among associations. A stronger mechanistic study would manipulate search restrictions, compare different semantic models, collect response timing, and test whether these changes predict later creative products evaluated by independent criteria. Repeated measurements would clarify the stability of performance across days and contexts. Such designs could distinguish a relatively durable capacity for broad retrieval from practice with a particular scoring algorithm.
Human Ratings, Consensual Assessment, and Judgment
Human judgments remain indispensable when investigators need to establish whether an idea is meaningful, plausible, elegant, surprising, or suitable for a particular audience. Amabile's consensual assessment approach Amabile (1982) examines whether domain-relevant judges agree in their assessment of creative products. In ideation tasks, raters may evaluate each candidate for unusualness and effectiveness, sometimes applying a global creativity judgment rather than decomposing the two criteria. This method permits contextual sensitivity that a semantic metric may lack. Its scientific strength depends on clarity about judges' expertise, the materials they see, whether they are blinded to conditions, and the degree of agreement across raters.
Silvia and colleagues (2008) evaluated subjective scoring methods for divergent-thinking tasks and studied their reliability and validity Silvia et al. (2008). Their work helped shift attention from the unqualified counting of rare responses toward ratings of idea quality. Yet human evaluation is influenced by calibration, task examples, semantic familiarity, and the social meaning attached to an answer. Researchers should document rater instructions, the number of judges, and the statistic used to estimate agreement, and should avoid assuming that agreement implies validity against an external creative-performance criterion. A group of judges can agree consistently on an overly narrow conception of originality.
Human and computational scoring need not be framed as permanent rivals. Algorithmic distances can be compared with human ratings and practical usefulness judgments, while residual disagreements may reveal constructs the model fails to represent. Human ratings can likewise be tested for systematic preferences for certain language styles or social signals. Crucially, judgments of idea quality are observations produced by a measurement procedure. They support claims about contextual assessed creativity, not direct access to a unique psychological essence. Strong validity arguments combine agreement estimates, external outcomes, robustness across prompts, and transparent scoring decisions.
Reliability, Construct Validity, and Score Aggregation
Reliability concerns the precision and consistency of a score under specified conditions; construct validity concerns whether the available evidence supports the interpretation and uses of that score. These questions are related yet distinct. A task can yield highly reproducible counts of responses while systematically rewarding verbosity or familiarity with a particular item. Conversely, a task that elicits genuinely original behavior may exhibit low retest reliability when it samples a narrow and context-sensitive capacity. The systematic review of divergent-thinking assessments published in 2026 Saretzki et al. (2026) identified substantial variation in instructions, timing, response coding, and aggregations, along with incomplete procedural descriptions. Those findings limit confidence in comparisons that treat differently scored tasks as equivalent measures.
An evidence-centered scoring plan specifies which response units count, how duplicates and implausible answers are handled, and whether originality is summed, averaged, or calculated from top responses. It then tests whether observed associations survive plausible alternative rules. A total originality score often reflects fluency and quality jointly, while a top-score procedure can overrepresent people who produced more opportunities to find an exceptional answer. When evaluators rank the best ideas, the researcher must decide whether to adjust for the number of submitted candidates. No one aggregation is valid for every scientific question; each estimates a different dimension of performance under its own assumptions.
Validity evidence should include convergence with related creativity measures, discrimination from adjacent constructs, stability across task prompts, and prediction of outcomes beyond the laboratory. It should also consider differential item functioning and whether scoring rules act comparably across demographic or linguistic groups. Kim's 2008 meta-analysis Kim (2008) and the 2012 review by Runco and Acar Runco & Acar (2012) both support treating divergent thinking as an indicator rather than a substitute for observed creative achievement. For longitudinal or applied claims, out-of-sample prediction, ecological measures, and actual products are especially informative.
Automated Ratings and the New Psychometric Landscape
Large language models can assign scores to written ideas with far less manual labor than traditional panels of judges. Their emergence changes the practical scale of creativity measurement and introduces additional methodological dependencies. Saretzki and Benedek's 2026 study compared automated originality ratings from five language-model scoring approaches with human assessments using several response aggregation methods Saretzki & Benedek (2026). The study explicitly evaluated construct validity relative to external creativity criteria, extending earlier work that emphasized agreement with human judges. This shift matters because a machine can imitate human preferences without improving measurement of the underlying construct.
Automated raters may be influenced by model version, prompt wording, response length, cultural associations, and familiarity with benchmark responses. Correlations with human ratings therefore require supplemented evidence: holdout tasks, repeated runs, criterion validity, and subgroup analyses. A lower cost of measurement does not establish independence from evaluator bias. When research uses the same family of models to generate creative ideas and then score them, the evaluation pipeline may favor outputs that align with model-learned linguistic patterns. A transparent protocol records the system identifier, date of scoring, prompt, numerical settings, input cleaning, aggregation, and any human adjudication.
Assessment quality improves when investigators compare multiple methods and ask what each is intended to measure. A semantic-distance metric estimates association remoteness, a human panel evaluates context-sensitive quality, and a model-assisted rating captures learned criteria embedded in a prompt and training process. The strongest claim currently supported is that these methods provide increasingly practical and testable indicators; none automatically licenses a domain-general ranking of creative minds. The 2026 evidence Saretzki & Benedek (2026) motivates a research program focused on measurement invariance, incremental validity, documentation of model drift, and robust comparison across human and artificial outputs.
Creative Potential, Achievement, and External Criteria
Creative potential is a conditional capacity to produce worthwhile novelty under appropriate circumstances. Creative achievement involves the realization of valuable outputs, often depending on access to materials, institutional opportunities, expertise, collaboration, persistence, and recognition. Divergent-thinking tasks primarily sample potential under narrow circumstances. Runco and Acar (2012) distinguish indicators from guaranteed outcomes Runco & Acar (2012), and Kim's meta-analysis (2008) demonstrates that connections to creative achievement are meaningful but modest and vary by domain and measure Kim (2008). A high test score should therefore be understood as evidence relevant to possible future performance, with uncertainty about the person, context, and outcome being predicted.
The relationship between ideation and achievement is mediated by choices made after generation. A person might develop many original proposals but lack time or resources to test them, while an expert might generate fewer ideas of exceptional feasibility. Such possibilities show why ideation research should collaborate with studies of domain expertise, creative evaluation, and social recognition without appropriating their distinct objects. Operational distinctions are essential when evaluating educational or workplace programs: improved performance on a familiar alternative-uses task may not imply better finished designs, published creative work, or durable professional innovation.
External criteria themselves require scrutiny. Creative achievement questionnaires combine opportunities, self-report, and socially assigned status; expert product ratings depend on domain conventions; patent or publication counts vary with institutions and access. Researchers should ideally sample multiple outcomes and examine the conditions under which divergent-thinking measures add predictive information beyond prior achievement, vocabulary, and relevant expertise. This approach prevents psychological theory from confusing the generation of possibilities with the full social history of a creative product, while preserving ideation's importance as one necessary component in many creative processes.
Neural Evidence for Idea Generation
Neuroimaging has expanded creativity research from observable response patterns to candidate neural systems involved in production and evaluation. In an fMRI study comparing the generation of novel ideas with the recall of known responses, Benedek and colleagues (2014) identified differences in brain activity associated with the two conditions Benedek et al. (2014). The finding supplies a neural contrast relevant to constructing versus retrieving familiar answers, but it is not a direct photograph of originality. The measured blood-oxygen-level-dependent signal reflects indirect physiological correlates, and task conditions can differ in difficulty, memory demands, and deliberation. Strong interpretation depends on whether those alternatives are controlled and whether the relevant behavior is measured independently.
Beaty and colleagues (2016) synthesized evidence that creative cognition may involve cooperation among brain systems associated with internally generated thought and executive control Beaty et al. (2016). The default-mode network is often implicated in spontaneous or internally oriented representations, while control-related networks participate in goal maintenance and evaluation. Such descriptions indicate patterns of coordinated recruitment, not a simple division of labor in which one network makes ideas and another mechanically approves them. Network labels summarize distributed and flexible functional relationships. The essential theoretical problem is to explain how fluctuations in these systems connect to the generation of distinguishable candidate responses.
Brain-based explanations require converging evidence from behavior, task contrasts, and network-level analyses. A reliable neural correlate of high-scoring ideation does not by itself establish causal necessity or exclusivity. Functional connectivity summarizes temporal relationships among signals and depends on preprocessing, movement control, task definition, and modeling assumptions. Experimental designs that manipulate generation constraints while tracking both response content and network configuration could advance mechanistic inference. The field's most defensible conclusion is that ideation depends on coordinated processes whose neural implementations are distributed, with task-specific details and causal direction still under active investigation.
Default and Executive Network Coordination
One influential proposal explains creative ideation through interactions between internally oriented associative processing and executive control. The framework accounts for the simultaneous need to explore possibilities and maintain task constraints, reconciling two operations that might appear to compete in everyday descriptions. Beaty, Benedek, Silvia, and Schacter (2016) reviewed network findings linking divergent thinking to default and executive system cooperation Beaty et al. (2016). This evidence supports a multi-network hypothesis rather than a single localized 'creativity center.' It does not demonstrate that the same connectivity pattern governs all creative domains, populations, or stages of idea production.
In a subsequent predictive connectivity study, Beaty and collaborators (2018) reported that patterns of functional brain connectivity were informative about individual creative ability as assessed in their research design Beaty et al. (2018). Predictive modeling is stronger when evaluated on held-out individuals than when associations are explored only in the original sample; nevertheless, prediction and causal explanation remain different achievements. Connectivity features can covary with language abilities, general cognitive resources, age, and experimental compliance. Interpreting a network signature requires transparency about training-test partitioning, the target task, covariate handling, and whether the model generalizes to new institutions and populations.
The network literature motivates a temporally specific research question: which forms of coordination precede exploratory search, strategic switching, and selective retention of ideas? Answering it requires time-resolved responses, robust neural recording, and models that distinguish cognitive transitions from generalized effort or alertness. Behavioral observations of strategy changes provide candidate events, while neural signals constrain plausible implementations. The two levels must be compared without assuming that a statistical link identifies a uniquely creative process. A theoretically useful neural account explains variation in idea production under experimentally varied constraints.
Individual Differences and the Boundaries of General Ability
People differ in the number of ideas they offer, the originality of their best responses, the tendency to persist when early associations are exhausted, and the quality of ideas selected for reporting. These dimensions can covary without being identical. Fluid intelligence and executive measures show some association with divergent-thinking tasks Nusbaum & Silvia (2011) Benedek et al. (2014), while semantic-network differences have been observed between groups distinguished by creativity measures Kenett et al. (2014). Such associations may arise partly through shared knowledge, motivation, language fluency, and familiarity with testing. Interpretation is strongest when the investigator states the specific ability being measured and models neighboring traits rather than placing every effect under a single label of innate creativity.
Individual-difference research also requires attention to score stability. A person's unusualness score for one object can fluctuate with interest in the object, fatigue, stress, and recent exposure to relevant concepts. Multiple task items and repeated sessions reduce dependence on any one context and allow the estimation of trait-like and state-like components. Moreover, differences in response style can masquerade as creative potential when one person writes concise answers and another supplies elaborated narratives that receive higher ratings. Methods that compare brief descriptions with standardized prompting can test whether rating differences represent content rather than presentation.
A theoretically meaningful profile might distinguish fluent generation, distant associative retrieval, strategic category switching, originality under constrained time, and effective selection. These capacities may correlate differently with learning histories and applied creative work. The purpose of such differentiation is explanatory precision, not the construction of a rigid taxonomy of people. The modest predictive association of divergent-thinking scores with achievement Kim (2008) is compatible with substantial influence from opportunities, expertise, persistence, and social conditions outside the laboratory. Predictions of later achievement must therefore be qualified by the actual criterion and the population under study.
Development, Education, and Learning Contexts
Developmental research raises two different questions: how the cognitive operations supporting idea generation change with age, and how educational environments shape opportunities to practice them. Vocabulary, inhibitory control, cognitive flexibility, and cumulative domain knowledge change during development, so a verbal divergent-thinking test administered across age groups may measure partially different constraints. Task prompts appropriate for adults may be unfamiliar to children, and younger participants may express inventive responses through action or drawing rather than fluent written explanation. Age comparisons need developmentally appropriate instruments, equivalent scoring, and explicit tests of whether the same construct is represented across groups.
In education, tasks that encourage students to produce alternatives can train a recognizable problem-solving routine and cultivate openness to multiple possibilities. The scientific issue is whether this practice improves performance on unfamiliar tasks and yields durable creative products. Programs vary widely in content, duration, instructor support, and outcome measures. Sio and Lortie-Forgues (2024) synthesized 169 creativity-training studies and identified substantial publication-bias and study-quality concerns Sio & Lortie-Forgues (2024). Their adjusted aggregate estimates were appreciably smaller than the unadjusted result, illustrating the difference between evidence for improvement on some studied tasks and a general claim that creative capacity can be increased substantially in every domain.
A rigorous developmental or educational study specifies what was taught, how long the intervention lasted, which assessment was used, and whether scorers were blinded. It distinguishes near transfer to a similar divergent-thinking instrument from far transfer to unfamiliar real-world projects. It also measures persistence and domain knowledge because these may mediate improvement. Longitudinal designs can separate immediate practice gains from stable changes, while randomized comparisons with equally engaging alternative activities help control expectancy and instructional attention. Such evidence would support practical education more strongly than repeated pretest-posttest gains on identical prompts.
Language, Culture, and Domain Specificity
Ideas are generated through concepts whose accessibility depends on language use, education, local environments, and everyday practices. A familiar household object may evoke different functions in different communities, and its uncommon uses may have different practical feasibility. Statistical rarity calculated within one cultural sample can misclassify an ordinary response from another sample as highly original. Cross-language comparisons also confront differences in word length, lexical richness, translation, and the performance of computational embeddings. A study using semantic distance must establish that the underlying linguistic representation measures comparable semantic relationships across groups, rather than treating one pretrained model as universally neutral.
Olson and colleagues' international Divergent Association Task data Olson et al. (2021) show that brief computational measures can be deployed across geographically diverse samples, but broad participation does not eliminate sampling bias or demonstrate equivalence among all languages and societies. Cultural validity requires explicit local task adaptation, relevant judges, population description, and measurement-invariance studies. Domain specificity raises a related problem. Generating an unusual use for a brick differs from proposing a mathematically elegant proof, a musical variation, or a safe engineering design. Domain knowledge can expand the candidate pool while imposing constraints that alter what counts as a creative answer.
The most productive approach asks which operations generalize and which require specialized knowledge. Broad retrieval, strategy switching, and evaluation may occur in many domains, but the information used and the success conditions can differ sharply. Transfer cannot be inferred simply because two tasks are both labeled creative. Strong evidence would show predictive relations across clearly distinct domains, account for common verbal and executive abilities, and test externally evaluated products. Culture and expertise are therefore design variables that shape the construct rather than nuisances that can always be removed through statistical adjustment.
Training Effects, Transfer, and Publication Bias
Intervention research is often cited as proof that divergent-thinking ability can be taught. The better-supported claim is more specific: some structured training interventions produce measurable changes on selected creativity outcomes under the conditions of the reviewed studies. Sio and Lortie-Forgues (2024) examined 169 studies and 844 effect sizes, including unpublished work, and estimated an unadjusted average benefit of 0.53 standard deviations Sio & Lortie-Forgues (2024). They also identified evidence consistent with substantial publication bias; several adjustment approaches reduced estimates to approximately 0.29–0.32 standard deviations. These numerical results describe that particular evidence synthesis and should not be transported as guaranteed outcomes for a specific educational program.
The magnitude of a training effect depends on design quality, selection of outcomes, the relation between training exercises and tests, and the treatment of unpublished null findings. If participants practice alternative uses and are then evaluated on closely related prompts, improved performance can arise from learned test strategy rather than broad creative improvement. When studies lack adequate control groups, blinded evaluation, or long-term follow-up, positive findings are vulnerable to expectancy effects and selective analysis. Heterogeneity also makes one grand average insufficient to judge every intervention. The evidence review Sio & Lortie-Forgues (2024) argues for rigorous, registered studies with more comparable and meaningful outcomes.
A stronger training program would teach both flexible generation and thoughtful assessment of contextual suitability while evaluating these operations separately. Research would compare genuine transfer to new objects and domains, use predeclared scoring, and test whether gains survive beyond immediate posttesting. It would also examine who benefits, which strategies change, and the resources required to implement the intervention. Such designs connect the cognitive science of ideation to educational applications while preserving the distinction between raising task-specific scores and generating valuable work in everyday settings.
Human and Artificial Divergent Idea Generation
Generative language models now produce fluent alternative uses, hypothetical consequences, and remote word associations, making creative ideation a direct subject of comparative human–artificial research. Hubert, Awa, and Zabelina (2024) compared responses from 151 human participants and GPT-4 on divergent-thinking tasks and reported higher creativity-task performance for the tested model under their experimental conditions Hubert et al. (2024). Such results demonstrate measurable capability on particular response-generation benchmarks. They do not establish a universal ordering of human and machine creativity because prompts, sample composition, scoring systems, model versions, and output-selection rules are integral to the result. Creative achievement and situated consequences remain broader constructs than benchmark response ratings.
A larger comparison by Wang, Huang, Shen, and Uzzi (2026) analyzed 9,198 human participants and 215,542 language-model observations Wang et al. (2026). In their task and evaluation system, mean human divergent-creativity performance was slightly higher, with greater human variation and higher scores in the upper tail. The direction of that result differs from the smaller 2024 benchmark, a scientifically useful discrepancy rather than a contradiction that should be concealed. The studies sampled different populations, systems, tasks, and conditions. Their divergence demonstrates why comparative claims should specify the measured distribution, not merely announce which agent is categorically more creative.
Artificial Era research must therefore compare mechanisms and results at explicit levels: generation of candidate responses, semantic diversity, contextual usefulness, independent judgments, reproducibility, and actual applications. A system may produce more ideas per unit time, whereas a human participant may exercise specialized contextual judgment or generate a rare response under different constraints. Conversely, humans can benefit from machine-generated prompts and combinations. These are testable claims about performance and interaction, not direct inferences about consciousness or mental experience. The proper research question concerns what psychological capacities and observable organizational functions are represented in a defined human–artificial ideation arrangement.
Comparative Evaluation, Prompting, and Attribution
Human–machine comparison presents an unusual psychometric challenge: language models can be both generators of ideas and automated judges of them. If an evaluation metric shares training representations with the system that produced responses, model-specific linguistic preferences may influence apparent success. The resulting dependence calls for independent human panels, functionally grounded criteria, and multiple scoring approaches whenever feasible. Saretzki and Benedek (2026) examined different language-model-assisted rating methods for divergent-thinking assessment and compared their validity evidence with human-based measures Saretzki & Benedek (2026). Their analysis illustrates that automated evaluation must be tested against external constructs rather than treated as a neutral authority.
Prompting choices also change the response distribution. Models can be asked for the greatest possible number of ideas, a few carefully selected alternatives, answers with practical constraints, or improbable associations, and these instructions resemble known manipulations of human divergent-thinking tasks. Wang and colleagues (2026) found that some prompting strategies had mixed or negative consequences under their comparison conditions Wang et al. (2026). Such findings discourage treating descriptive personas or novelty-oriented instructions as automatic means of improving quality. Investigators should record prompt text, settings, model version, sample selection, duplicate handling, and evaluation criteria alongside the participant instructions used for human subjects.
Attribution is another possible source of bias: the same idea can be judged differently if readers believe it originated from a person or an artificial system. This hypothesis requires blinded and randomized source labels to separate response quality from expectations about the producer. In this dossier, the central issue is how source identification may contaminate the measurement of divergent idea production; the broader aesthetics of AI-generated artworks and social recognition of creative labor belong to adjacent research objects. A defensible comparative study makes the source of each idea a controlled variable whenever judgments of quality are important.
Human–Artificial Cognitive Partnership in Ideation
A human–artificial ideation partnership can be conceptualized as an interactive sequence in which one participant proposes alternatives, another reformulates or questions them, and successive constraints produce new candidate sets. The scientific subject is the changing organization of search and output quality during interaction. The sequence can be tested against human-alone and model-alone baselines using identical task conditions, time budgets, and independent assessments. The existence of an interaction does not establish a cooperative benefit; a partner may narrow the user's search through anchoring, increase repetitive proposals, or provide useful cues that expand the available possibility space.
Research should distinguish the number of produced suggestions from their independence, distinctiveness, and utility. If a system offers many variations that share the same template, a combined team can appear highly fluent without increasing conceptual diversity. Conversely, a deliberately contrasting machine proposal might help a human leave a conventional search category. A meaningful design records idea provenance, human selection, subsequent revisions, and eventual outcomes. It also permits comparisons between spontaneous model suggestions, strategically distant prompts, and prompts chosen by users, thereby testing which interaction features actually alter ideation.
These questions situate the psychology of divergent thinking within the historical horizon of Artificial Era · From Homo to Artificial. They concern the architecture of creativity-related performance across biological and nonbiological systems, the measurement of distributed search, and the contribution of each participant to an evolving idea. Claims about artificial personhood, conscious experience, or globally superior creativity require their own evidence and are outside the inference licensed by divergent-thinking scores. The scientific value of the comparison lies in making previously human-centered assumptions about generation, evaluation, and cognitive partnership empirically testable.
Methodological Threats and Reproducibility
Three recurring sources of uncertainty affect the creative-ideation literature: construct ambiguity, variability in task administration, and dependence of the result on scoring rules. Saretzki, Forthmann, and Benedek (2026) explicitly documented extensive variation and incomplete reporting across divergent-thinking studies Saretzki et al. (2026). Two laboratories may both report 'originality' while one uses rare-response frequency, another uses independent judge ratings, and a third uses language-model semantic distance. These measures can be correlated yet remain nonidentical. Without a detailed methods record, even an apparent replication can compare different dependent variables.
Sampling and design add further limitations. Many laboratory studies use convenience samples, and individual differences in age, language, education, motivation, and task familiarity can affect generalization. Cross-sectional neuroimaging and semantic-network studies generate important hypotheses but often cannot identify causal direction. Comparisons between humans and generative models can be distorted by unequal time budgets, number of outputs, instruction detail, and opportunities to curate responses. Publication bias and small-study effects matter especially in intervention research, as shown by Sio and Lortie-Forgues (2024) Sio & Lortie-Forgues (2024). Conclusions should follow the design: randomized conditions support specific causal claims, correlations support covariation, and reviews depend on the quality and comparability of included studies.
Reproducibility can be strengthened through preregistration of contrasts and scoring, open prompts and item lists, response-level datasets where privacy permits, documented rater training, robust uncertainty estimates, and independent replication across populations. Computational work should archive model identifiers and evaluation settings, because model updates can make an apparently identical prompt a different measurement instrument. Researchers should disclose selection rules for invalid responses and missing data, and should evaluate whether conclusions survive defensible alternative scoring schemes. The aim is to make ideation a cumulatively testable scientific object whose central claims remain interpretable across laboratories and technological changes.
Critical Scientific Discussion and Open Questions
The evidence supports the importance of associative retrieval, strategic executive regulation, and task-context-sensitive generation, while leaving their relative causal contributions unsettled. Findings on semantic-network structure Kenett et al. (2014) and executive control Benedek et al. (2014) can coexist because network accessibility and goal management describe different aspects of the same performance. Disagreement about a single 'creative capacity' often reflects the mixture of response quantity, semantic remoteness, judged appropriateness, and external achievement in supposedly equivalent tests. The practical advance is to identify which property each method measures and which mechanisms predict its variation.
Several unresolved questions are especially consequential. It remains unclear how much of a participant's search strategy generalizes across verbal and nonverbal domains, whether later serial-position originality reflects genuine exploratory restructuring or selective endurance, and how language-specific algorithms affect international comparisons. Neural prediction requires larger and more varied out-of-sample tests with explicit controls for general ability and task performance. Interventions must establish durable far transfer rather than improvement on practiced tasks. Comparative machine studies need standardized opportunities for generation, independent raters, and preservation of full response distributions including exceptional and low-quality outputs.
The literature also faces an interpretive tension between useful novelty and formal unpredictability. A response can be distant from familiar language while being irrelevant, and a high-quality creative solution can use familiar elements reorganized by domain expertise. A defensible scientific theory must therefore join observations about search to criteria grounded in actual constraints. The capacity to generate alternatives is a central component of many creative processes, but evidence about it becomes more informative when paired with transparent evaluations of usefulness, development, and social context. Future research can advance by connecting the trajectories of individual ideas to controlled manipulations and independently meaningful outcomes.
Conclusion
Creative ideation and divergent thinking constitute a coherent empirical research field organized around the production of candidate alternatives under open-ended constraints. Historical work established the study of multiple-response performance; associative models examined accessibility of remote connections; creative-cognition accounts described representational transformation; and subsequent experiments investigated controlled search, response sequences, psychometric scoring, and neural coordination. Evidence from tasks, semantic models, and neurocognitive studies supports a multicomponent interpretation in which retrieval, exploration, regulation, and provisional evaluation are functionally important. Each component becomes scientifically useful when connected to a specific manipulation and observable outcome.
Divergent-thinking assessments provide meaningful but bounded evidence about creative potential. Their significance depends on instructions, task duration, language, scoring, and the criterion used to validate the outcome. Recent systematic research documents assessment heterogeneity Saretzki et al. (2026), historical meta-analysis finds modest prediction of creative achievement Kim (2008), and training meta-analysis shows that accounting for publication bias materially reduces average estimated benefits Sio & Lortie-Forgues (2024). The synthesis therefore identifies measurement precision, causal inference, and generalization as the central conditions for durable cumulative knowledge.
Research on artificial idea generators makes these conditions newly urgent. Different well-documented comparisons have produced different aggregate human–model rankings Hubert et al. (2024) Wang et al. (2026), demonstrating the sensitivity of conclusions to tasks, samples, and procedures. The strongest research agenda studies the distribution and functional quality of alternatives, the organization of search, and the contribution of each participant in human–artificial interactions. Creative Ideation and Divergent Thinking thereby occupies a defined place within the Psychology by Bogdanova research architecture: a foundational analysis of idea generation linked to, but scientifically distinct from, insight, creative evaluation, domain achievement, and institutional recognition.
References
- Amabile, T. M. (1982). Social psychology of creativity: A consensual assessment technique. Journal of Personality and Social Psychology, 43(5), 997–1013. https://doi.org/10.1037/0022-3514.43.5.997
- Beaty, R. E., Benedek, M., Silvia, P. J., & Schacter, D. L. (2016). Creative cognition and brain network dynamics. Trends in Cognitive Sciences, 20(2), 87–95. https://doi.org/10.1016/j.tics.2015.10.004
- Beaty, R. E., Kenett, Y. N., Christensen, A. P., Rosenberg, M. D., Benedek, M., Chen, Q., Fink, A., Qiu, J., Kwapil, T. R., Kane, M. J., & Silvia, P. J. (2018). Robust prediction of individual creative ability from brain functional connectivity. Proceedings of the National Academy of Sciences, 115(5), 1087–1092. https://doi.org/10.1073/pnas.1713532115
- Beaty, R. E., & Johnson, D. R. (2021). Automating creativity assessment with SemDis: An open platform for computing semantic distance. Behavior Research Methods, 53(2), 757–780. https://doi.org/10.3758/s13428-020-01453-w
- Benedek, M., Jauk, E., Sommer, M., Arendasy, M., & Neubauer, A. C. (2014). Intelligence, creativity, and cognitive control: The common and differential involvement of executive functions in intelligence and creativity. Intelligence, 46, 73–83. https://doi.org/10.1016/j.intell.2014.05.007
- Benedek, M., Jauk, E., Fink, A., Koschutnig, K., Reishofer, G., Ebner, F., & Neubauer, A. C. (2014). To create or to recall? Neural mechanisms underlying the generation of creative new ideas. NeuroImage, 88, 125–133. https://doi.org/10.1016/j.neuroimage.2013.11.021
- Finke, R. A., Smith, S. M., & Ward, T. B. (1992). Creative cognition: Theory, research, and applications. MIT Press. https://mitpress.mit.edu/9780262061506/creative-cognition/
- Guilford, J. P. (1950). Creativity. American Psychologist, 5(9), 444–454. https://doi.org/10.1037/h0063487
- Hubert, K. F., Awa, K. N., & Zabelina, D. L. (2024). The current state of artificial intelligence generative language models is more creative than humans on divergent thinking tasks. Scientific Reports, 14, Article 3440. https://doi.org/10.1038/s41598-024-53303-w
- Kenett, Y. N., Anaki, D., & Faust, M. (2014). Investigating the structure of semantic networks in low and high creative persons. Frontiers in Human Neuroscience, 8, Article 407. https://doi.org/10.3389/fnhum.2014.00407
- Kim, K. H. (2008). Meta-analyses of the relationship of creative achievement to both IQ and divergent thinking test scores. The Journal of Creative Behavior, 42(2), 106–130. https://doi.org/10.1002/j.2162-6057.2008.tb01290.x
- Mednick, S. A. (1962). The associative basis of the creative process. Psychological Review, 69(3), 220–232. https://doi.org/10.1037/h0048850
- Nusbaum, E. C., & Silvia, P. J. (2011). Are intelligence and creativity really so different? Fluid intelligence, executive processes, and strategy use in divergent thinking. Intelligence, 39(1), 36–45. https://doi.org/10.1016/j.intell.2010.11.002
- Olson, J. A., Nahas, J., Chmoulevitch, D., Cropper, S. J., & Webb, M. E. (2021). Naming unrelated words predicts creativity. Proceedings of the National Academy of Sciences, 118(25), Article e2022340118. https://doi.org/10.1073/pnas.2022340118
- Runco, M. A., & Acar, S. (2012). Divergent thinking as an indicator of creative potential. Creativity Research Journal, 24(1), 66–75. https://doi.org/10.1080/10400419.2012.652929
- Runco, M. A., & Jaeger, G. J. (2012). The standard definition of creativity. Creativity Research Journal, 24(1), 92–96. https://doi.org/10.1080/10400419.2012.650092
- Saretzki, J., & Benedek, M. (2026). Investigating the validity evidence of automated scoring methods for divergent thinking assessments. Creativity Research Journal. Advance online publication. https://doi.org/10.1080/10400419.2025.2607669
- Saretzki, J., Forthmann, B., & Benedek, M. (2026). A systematic quantitative review of divergent thinking assessments. Psychology of Aesthetics, Creativity, and the Arts, 20(1), 1–20. https://doi.org/10.1037/aca0000691
- Silvia, P. J., Winterstein, B. P., Willse, J. T., Barona, C. M., Cram, J. T., Hess, K. I., Martinez, J. L., & Richard, C. A. (2008). Assessing creativity with divergent thinking tasks: Exploring the reliability and validity of new subjective scoring methods. Psychology of Aesthetics, Creativity, and the Arts, 2(2), 68–85. https://doi.org/10.1037/1931-3896.2.2.68
- Sio, U. N., & Lortie-Forgues, H. (2024). The impact of creativity training on creative performance: A meta-analytic review and critical evaluation of 5 decades of creativity training studies. Psychological Bulletin, 150(5), 554–585. https://doi.org/10.1037/bul0000432
- Wang, M., Hao, N., Ku, Y., Grabner, R. H., & Fink, A. (2017). Neural correlates of serial order effect in verbal divergent thinking. Neuropsychologia, 99, 92–100. https://doi.org/10.1016/j.neuropsychologia.2017.03.001
- Wang, D., Huang, D., Shen, H., & Uzzi, B. (2026). A large-scale comparison of divergent creativity in humans and large language models. Nature Human Behaviour, 10, 531–540. https://doi.org/10.1038/s41562-025-02331-1