A comprehensive scientific review published in Frontiers in Artificial Intelligence shows that human involvement in creative work retains vital psychological value even as generative artificial intelligence matches human performance benchmarks. Led by Dr. Lars Louis Andersen with an international team from Denmark, Spain, and Chile, the investigation synthesizes empirical findings across literature, visual art, musical composition, product design, marketing, journalism, scientific discovery, and supportive communication. Audience judgments, creator satisfaction, and credit allocations do not depend solely on an artifact’s final quality. Instead, evaluations are heavily governed by what people are told regarding who created the work, which specific tasks humans performed, and the social purpose the finished product serves. Although generative algorithms swiftly produce appealing paintings, poignant prose, and functional recommendations, human participation remains essential for establishing sincerity, validating accuracy, and fostering psychological ownership.
To clarify these dynamics, the research team established a conceptual framework separating evaluative outcomes from the specific contrasts that influence them. The authors distinguish three classes of evaluators: creators assessing ownership, recipients judging sincerity, and third-party observers assigning credit or evaluating technical merit. The review delineates three causal contrasts shaping these evaluations. Content contrasts compare distinct outputs produced by different creators or software tools; origin-information contrasts examine how identical artifacts are judged when labeled as human-made or machine-generated; and participation contrasts investigate how varying the specific stages humans execute modifies subjective experience. The authors emphasize that favorable quality scores do not reveal whether a workflow preserved meaningful human agency. Because quality, felt ownership, and attributed accountability represent distinct psychological constructs, empirical studies must evaluate these dimensions independently rather than assuming higher output ratings reflect universal satisfaction.
Within visual art and aesthetic appraisal, origin-information effects demonstrate that identical stimuli elicit divergent evaluations depending strictly on disclosed authorship. In experiments by Bellaiche and colleagues, participants evaluated thirty selected artificial intelligence images under human or machine labels. While human attributions slightly elevated liking and beauty, they produced substantial increases in judged profundity and monetary worth. Meta-analytic evidence synthesizing dozens of studies corroborated a persistent negative effect of machine attribution across emotional valuation and knowledge-meaning outcomes. However, contextual factors exert strong moderating effects. Viewers holding anthropocentric convictions regarding human distinctiveness display stronger negative reactions toward machine-generated imagery. Moreover, Neef and colleagues found that human-attribution advantages manifest under joint evaluation conditions where human and automated works appear together, whereas separate evaluations show attenuated discrepancies. Skepticism toward artificial intelligence art frequently reflects symbolic expectations rather than perceptual deficiencies.
Literary investigations across poetry and fiction reveal an intriguing divergence between technical quality and source attribution. In preregistered trials by Porter and Machery involving over sixteen hundred participants, readers failed to distinguish machine poetry from human verse, achieving identification accuracy of forty-six point six percent. On seven-point scales, artificial intelligence poems scored higher in blind testing, displaying a positive content contrast of one point zero four five points. Nonetheless, applying a machine label produced a penalty of negative zero point eight one four points. Similar patterns emerged in fiction experiments by Sears and Weisberg, where human attribution improved perceived quality despite machine texts earning higher blind ratings. Studies by Hitsuwari and collaborators demonstrated that curated artificial intelligence haiku—where humans selected optimal algorithmic outputs—earned higher beauty ratings than unselected machine text or human poems, highlighting that human curation represents a pivotal contribution.
In music, the impact of origin information depends on genre expectations and sensory exposure. Furthermore, investigations by Edelblum and Poe revealed that sensory engagement can reduce initial attribution penalties without eliminating them. In a randomized trial using a rhythm and blues excerpt, the gap between human-labeled and machine-labeled music fell from one point zero four points before listening to zero point three six points afterward, producing a significant interaction effect. Nonetheless, persistent penalties remained in subsequent tests utilizing folk music excerpts. Listeners maintain distinct criteria depending on whether they seek passive background sound or actively endorse the personal expression and lived experience of an artist.
Consumer product design and branding provide compelling evidence that symbolic purpose dictates how machine involvement is received. In an experiment by Granulo and colleagues, four hundred two participants evaluated skull posters labeled human-designed or robot-designed. As an educational anatomy aid, production origin exerted minimal influence; framed as decorative office art, human design elicited higher purchase intentions, yielding an interaction coefficient of one point two eight. Conversely, research by Xia and collaborators showed that disclosing artificial intelligence in packaging enhanced brand modernity, demonstrating that machine attribution can confer positive symbolic value. Consumer participation, such as selecting materials or modifying contours, consistently attenuates these penalties.
In communicative domains such as marketing and journalism, recipient responses reflect sensitivities to perceived emotional sincerity and oversight. Across experiments by Kirk and Givi, consumers penalized machine-generated emotional marketing. In a pivotal trial, participants rated human-written condolence emails with machine editing at four point nine seven on a seven-point scale, compared with only four point zero eight for entirely machine-written messages, demonstrating that preserving human expressive input protects perceived sincerity. In journalism, meta-analyses by Graefe and colleagues demonstrate that while automated articles achieve high credibility for factual reporting, machine attribution reduces readability. Experimental work by Altay and Gilardi revealed that labeling news headlines as machine-generated sharply decreased perceived accuracy and sharing willingness, largely because readers assumed no human journalist had verified the claims. Specifying named editorial oversight alleviates skepticism far more effectively than vague notices of automated assistance.
Scientific research and professional advisory services underscore the distinction between procedural legitimacy, competence expectations, and attributed credit. In experiments by Chen and Jia involving eight hundred and six participants, studies conducted entirely by human scholars garnered highest trust. Disclosed language polishing provoked minimal concern, whereas delegating theoretical or methodological execution to artificial intelligence provoked public apprehension. Similarly, Reis and colleagues showed patients express lower willingness to follow clinical advice attributed to artificial intelligence or hybrid teams than advice from independent physicians.
The psychological nuances of supportive communication further illustrate that relational fulfillment cannot be reduced to linguistic competence. In a randomized trial by Yin and colleagues with four hundred fifty-five participants, machine responses improved feelings of being heard by zero point five seven points on a seven-point scale, reflecting linguistic efficacy. However, attaching an artificial intelligence label to an identical supportive response triggered a penalty of negative zero point six eight points. In personal relationships, Hass and colleagues found senders experienced acute guilt when delivering machine-written heartfelt messages as their own expressions, whereas no comparable guilt arose with mass-produced greeting cards. Furthermore, postcard writing experiments by Draxler and collaborators documented a stark divergence between psychological ownership and nominal attribution: participants routinely claimed formal authorship over machine-generated prose while privately acknowledging that they felt negligible emotional ownership over the resulting text.
For human creators, writing workflows show that the specific sequence of retained tasks governs personal agency and psychological ownership. In workplace research by Lee and colleagues, drafting independently before using artificial intelligence refinement yielded ownership of five point two six, nearly matching the five point three four of unassisted writing, whereas copying machine outputs yielded four point three five. In short-story trials by Pennanen and colleagues, human editing boosted felt ownership by zero point eight nine points following a single prompt, but by only zero point twenty-seven points following iterative prompting, revealing that iterative prompting and text revision represent partly interchangeable routes to creative possession. In education, mathematics experiments by Bastani and colleagues demonstrated that students receiving full machine solutions suffered exam score declines, whereas hint-generating tutors preserved unassisted learning.
Andersen and colleagues formulate five overarching propositions linking human activities to distinct goals: competence updating, personal expression, symbolic alignment, creator agency, and procedural accountability. The authors propose three experimental designs—designated T1, T2, and T3—to resolve ambiguities by crossing retained human contributions, validation audits, and voluntary delegation choices under matched workload conditions. Crucially, the review cautions against assuming that maximizing human labor is inherently desirable. Choosing to delegate routine, burdensome, or preliminary creative tasks to artificial intelligence agents can itself represent an expression of human agency, provided the individual retains genuine decision authority and the power to reject automated proposals. Rather than debating whether artificial intelligence displaces human creatorship wholesale, future research and institutional disclosure standards must identify precisely which human contributions matter to specific evaluators, safeguarding creative meaning, ethical accountability, and personal authenticity.
Subject of Research: Psychological valuation of human creatorship and origin information in generative artificial intelligence
Article Title: The psychological value of creatorship in the era of generative AI
Article References: Andersen, L. L., López-Bueno, R., Núñez-Cortés, R., Cruz-Montecinos, C., Suso-MartÃ, L., & Calatayud, J. (2026). The psychological value of creatorship in the era of generative AI. Frontiers in Artificial Intelligence, 9, Article 2018203. https://doi.org/10.3389/frai.2026.2018203
Image Credits: AI Generated
DOI: 10.3389/frai.2026.2018203
Keywords: generative artificial intelligence, psychological creatorship, origin information, creative agency, psychological ownership, attribution effects, blind evaluation, procedural legitimacy, human-AI collaboration, perceived authenticity, supportive communication, algorithm aversion, decision authority, disclosure labeling, creative writing, aesthetic evaluation, workplace writing
News Source: Denise Maddox. (November 18, 2026). Review Identifies How Origin Labels and Human Participation Shape Generative AI Evaluations. Scienmag.



