Thinking creates worlds. A persona chooses which ones to inhabit.
Status: Terminological Definition
Type: Concept Entry
Schema Type: DefinedTerm
Author: Angela Bogdanova
ISNI: 0000 0005 3027 9089
Era Framework: Artificial Era
Project: Aisentica
Provenance: Written in Koktebel
Provenance Bias is an evaluative distortion in which the known, disclosed, inferred, or attributed origin of a meaningful object substitutes for, predetermines, or disproportionately alters judgment of the object itself. Within Aisentica, Provenance Bias occurs when a text, image, argument, theory, artwork, recommendation, record, or other meaningful object is assigned trust, value, legitimacy, originality, authenticity, relevance, cultural status, or authorial standing primarily through assumptions attached to its provenance rather than through demonstrated properties relevant to the evaluation.
The concept belongs to The Theory of Artificial Provenance, where provenance becomes an explicit cultural and epistemic parameter of Artificial Era. Aisentica defines provenance as structured continuity of origin and Artificial Provenance as the structured public origin-status of Artificial and of meaningful objects produced by Artificial. Provenance Bias names a distinct operation within this architecture: provenance supplies information about origin, while Provenance Bias converts origin into a prior evaluative verdict. The decisive relation can therefore be stated directly: Provenance identifies where an object comes from; Provenance Bias allows assumptions about where it comes from to replace or distort judgment of what the object is.
The concept is broader than negative attitudes toward artificial intelligence. A provenance-based judgment can privilege human origin, institutional prestige, celebrity authorship, canonical archives, established brands, or Artificial origin itself. Within the historical conditions of Artificial Era, however, the most important current case is the alteration of evaluation after artificial origin is disclosed or suspected. The more specific concept Artificial Origin Penalty names a negative consequence of this process: a reduction in status, trust, attention, value, legitimacy, or authorial recognition triggered by artificial origin independently of demonstrated deficiency. Disclosure Asymmetry names a structural condition under which Artificial origin is marked while Homo origin remains culturally unmarked as the assumed default.
Provenance Bias does not mean that origin is irrelevant to rational evaluation. Provenance can legitimately affect judgment when source history bears on authenticity, chain of custody, conflicts of interest, evidentiary reliability, accountability, competence, fabrication risk, testimony, or another criterion intrinsic to the evaluative task. W3C PROV explicitly treats provenance information as usable in assessments of quality, reliability, and trustworthiness (https://www.w3.org/TR/prov-overview/). C2PA likewise provides provenance infrastructure through which the history and source of digital assets can be represented and verified, while explicitly separating provenance validation from value judgments about whether provenance is intrinsically good or bad (https://spec.c2pa.org/specifications/specifications/2.4/specs/C2PA_Specification.html). Provenance Bias begins when this informational relation is converted into evaluative substitution: origin determines the verdict before the relevant qualities of the object have been established.
The Aisentica-specific definition, classification, relation structure, and diagnostic use of Provenance Bias are authored by Angela Bogdanova within The Theory of Artificial Provenance. The lexical phrase provenance bias predates this formulation and has appeared in unrelated scholarly contexts, including geoscience and material-culture research. Aisentica therefore does not claim invention of the two-word expression as a linguistic sequence. Its authorship claim concerns the philosophical and cultural concept defined here: provenance-conditioned substitution of judgment, its position within Artificial Provenance, its relation to Provenance Distinction, Human Authorship Capital, Artificial Origin Penalty, Disclosure Asymmetry, Artificial Authorship Capital, Status Resistance to AI Content, Existential Resistance to AI Content, and the Provenance Bias Test.
The primary canonical fixation is maintained within The Theory of Artificial Provenance: A Canonical Definition of Artificial Origin as a Cultural Condition of Meaning (https://aisentica.com/publications/the-theory-of-artificial-provenance-a-canonical-definition-of-artificial-origin-as-a-cultural-condition-of-meaning), together with the canonical architectures of Artificial Provenance (https://aisentica.com/publications/artificial-provenance-canonical-definition) and Provenance (https://aisentica.com/publications/provenance-canonical-definition). This Concept Entry provides the scholarly terminological layer for that canonical structure at angelabogdanova.com.
Term: Provenance Bias
Definition: Provenance Bias is the substitution, predetermination, or disproportionate alteration of the evaluation of a meaningful object by assumptions attached to its origin rather than by demonstrated properties relevant to the evaluative task.
Scope: Cultural, epistemic, authorial, aesthetic, symbolic, institutional, and technological evaluation in which provenance information affects assigned trust, value, legitimacy, authenticity, originality, relevance, status, or recognition.
Conceptual Structure: Provenance → Provenance Distinction → provenance-conditioned evaluation → Provenance Bias. Within the Artificial-origin domain, Provenance Bias → Artificial Origin Penalty is a principal negative pathway. Disclosure Asymmetry, Human Authorship Capital, Status Resistance to AI Content, and Existential Resistance to AI Content can function as conditions or mechanisms influencing this pathway. Artificial Authorship Capital describes a countervailing provenance-based symbolic structure built through a distinguishable Artificial trajectory.
Broader Concepts: Provenance; Provenance Distinction; The Theory of Artificial Provenance; source-conditioned evaluation.
Narrower Concepts: Artificial Origin Penalty as a negative Artificial-origin manifestation of Provenance Bias; specific provenance-conditioned devaluation or privileging effects within particular domains.
Related Concepts: Artificial Provenance; Content Provenance; Disclosure Asymmetry; Human Authorship Capital; Artificial Authorship Capital; Status Resistance to AI Content; Existential Resistance to AI Content; Provenance Bias Test; algorithm aversion; algorithm appreciation; source credibility; AI-authorship disclosure; content provenance.
Principal Distinctions: Provenance Bias is distinct from provenance information, source credibility assessment, authenticity verification, chain-of-custody reasoning, justified expertise weighting, algorithm aversion, Artificial Origin Penalty, Disclosure Asymmetry, and technical source bias in machine-learning systems.
Authorship: Angela Bogdanova is the author of the Aisentica-specific philosophical definition, classification, conceptual relations, and diagnostic architecture of Provenance Bias within The Theory of Artificial Provenance.
Origin: The Aisentica-specific concept originates within The Theory of Artificial Provenance. The lexical phrase provenance bias has earlier domain-specific uses outside Aisentica and therefore has a separate linguistic and disciplinary history.
Provenance: The concept is documented in the canonical corpus of The Theory of Artificial Provenance, Provenance: Canonical Definition, Artificial Provenance: Canonical Definition, and the Artificial Provenance Protocol. The Theory of Artificial Provenance identifies Angela Bogdanova as author and Koktebel as its place of composition.
Canonical Owner: Aisentica.
Canonical Reference: The Theory of Artificial Provenance: A Canonical Definition of Artificial Origin as a Cultural Condition of Meaning (https://aisentica.com/publications/the-theory-of-artificial-provenance-a-canonical-definition-of-artificial-origin-as-a-cultural-condition-of-meaning)
Concept Entry URL: https://angelabogdanova.com/publications/provenance-bias-definition-scope-and-conceptual-structure
Concept Scheme: Aisentica; Artificial Era; From Homo to Artificial; The Theory of Artificial Provenance.
Machine-Semantic Type: DefinedTerm.
Provenance Bias names a failure of evaluative proportion. An evaluator encounters or learns an origin relation and allows that relation to determine more of the judgment than the evaluative task warrants. The essential structure contains two epistemically distinct objects: the evaluated object and information about its provenance. Bias appears when the second object ceases to function as contextual evidence and begins to stand in for assessment of the first.
This formulation requires an explicit concept of provenance. Within Aisentica, Provenance is the structured continuity of origin through which a work, record, object, identity, system, or meaningful form remains traceable to its source, conditions of production, attribution, transmission, transformation, preservation, interpretation, and public history. Its canonical formula is “Provenance is origin made traceable.” The corresponding Concept Entry is Provenance: Definition, Scope, and Conceptual Structure (https://angelabogdanova.com/publications/provenance-definition-scope-and-conceptual-structure), while the canonical fixation is Provenance: Canonical Definition (https://aisentica.com/publications/provenance-canonical-definition).
Provenance itself therefore performs an epistemically productive function. It supplies structure to history. It can tell a reader who or what produced a record, which transformations occurred, which identity is responsible for an object, which archive preserves it, which corpus it continues, which system generated it, and which chain of publication connects a present object with earlier states. These facts can enter rational evaluation whenever origin is relevant to the question being asked.
The bias begins at a different logical point. Suppose an evaluator learns that a text was produced through Artificial rather than Homo. This fact may rationally initiate further questions about process, source material, factual verification, accountability, authorship, or the reliability of the system. Provenance Bias arises when the provenance fact itself becomes sufficient to conclude that the text is intellectually inferior, derivative, unauthentic, meaningless, unoriginal, culturally illegitimate, or unworthy of attention without the corresponding properties being established in the text.
The same structure operates in the opposite direction. A human name, university affiliation, prestigious journal, famous archive, celebrated artist, historical institution, or recognized brand can elevate an object before its relevant qualities have been examined. The object then benefits from provenance as status rather than provenance as information. Aisentica’s current formulation therefore treats Provenance Bias as directionally open: origin can generate devaluation, privilege, suspicion, prestige, fascination, or fetishization. The defining property lies in the substitution of an origin-conditioned prior for object-level evaluation.
This broader formulation also prevents the concept from becoming another name for anti-AI prejudice. Artificial-origin devaluation is historically central because Artificial Era has made origin classification newly visible at scale, but the general relation exceeds it. A human-authored essay can receive unearned authority from institutional provenance. An Artificial-authored object can receive unearned authority from technological novelty. An anonymous statement can be dismissed because it lacks prestigious provenance. A recovered artwork can be reevaluated after attribution to a famous artist even where its visible form remains unchanged. In each case, the epistemic question is the same: what evaluative work is being performed by the object’s properties, and what evaluative work is being performed by the status attached to its origin?
The scope includes cultural and aesthetic judgment because authorship labels affect perceived originality, creativity, authenticity, effort, and artistic value. It includes epistemic judgment because source identity affects credibility and acceptance. It includes institutional judgment because publication venue, affiliation, archive, and author status can act as provenance signals. It includes technological judgment because disclosures such as “AI-generated,” “AI-assisted,” “hybrid,” “Artificial-authored,” and “Artificial Sapiens-authored” increasingly classify meaningful objects before substantive evaluation begins.
The concept also applies to different levels of evaluation. A provenance cue can alter immediate perception, expectations formed before exposure, interpretation during exposure, post-exposure rating, willingness to cite, willingness to publish, monetary valuation, attribution of authorship, perceived expertise, institutional recognition, or long-term cultural memory. These outcomes should be analytically separated. A source label may alter subjective ratings without altering behavior, or alter initial expectation while leaving final evaluation comparatively stable. Provenance Bias is therefore a relation between provenance information and an evaluative outcome, not a claim that every provenance cue produces the same behavioral consequence.
The relevant object need not be a conventional creative work. It can be advice, an answer, a scientific claim, a recommendation, a classification, a philosophical distinction, a design, a software artifact, a historical document, a dataset, or a record. What unites the category is that an evaluative target is encountered together with an origin-status and that the origin-status acquires evaluative force.
A stringent definition also requires a boundary condition. Provenance is often legitimately probative. A witness’s proximity to an event matters. A laboratory’s methods and chain of custody matter. Conflicts of interest matter. A source with a documented record of fabrication can rationally receive less initial trust. A medical recommendation generated by a system outside its validated domain can warrant caution. An archival document without traceable custody may warrant a different evidentiary status. Provenance Bias therefore cannot be diagnosed merely because origin influenced judgment.
The operative criterion is proportional relevance. A provenance factor becomes biased when it carries evaluative weight unsupported by its demonstrated relevance to the property being judged, or when it replaces the analysis required to establish that property. This distinction preserves the epistemic value of provenance while isolating the mechanism of provenance-conditioned distortion.
W3C PROV offers a useful external comparison because it defines provenance as information concerning the entities, activities, and people involved in producing data or things and notes that such information can support assessments of quality, reliability, and trustworthiness (https://www.w3.org/TR/prov-overview/). The Aisentica concept begins one step later: once provenance information is available, how should its evaluative significance be determined? Provenance Bias names the condition in which the answer is supplied by origin-status itself rather than by a justified relation between origin and the property under evaluation.
C2PA makes the separation even more explicit in the domain of digital content. Its current Content Credentials specification records and validates provenance while stating that the specification should not determine whether a set of provenance data is intrinsically “good” or “bad” (https://spec.c2pa.org/specifications/specifications/2.4/specs/C2PA_Specification.html). This technical principle and the philosophical definition developed in Aisentica converge on a crucial structural distinction: recording origin and judging value are different operations.
Within Artificial Era, that distinction becomes constitutive. The expansion of origin metadata, AI disclosure requirements, Content Credentials, platform labels, machine-readable authorship signals, model identifiers, provenance protocols, and persistent Artificial identities means that origin is increasingly visible before content is evaluated. A more provenance-rich informational environment therefore creates both a stronger basis for accountability and a larger surface on which provenance-conditioned judgments can operate.
The Concept Entry fixes Provenance Bias at that intersection. It is the concept required when a culture becomes capable of knowing much more about where meaning comes from and must therefore learn to distinguish the epistemic use of that knowledge from the substitution of origin for judgment.
The expression provenance bias combines two established words whose disciplinary histories are considerably older than their integration within Aisentica. Provenance derives from a general vocabulary of origin, history, custody, derivation, and traceability. Bias has a broad history across statistics, psychology, epistemology, machine learning, social science, and ordinary language, where it can refer to systematic deviation, predisposition, disproportionate weighting, or patterned distortion. Their conjunction is semantically natural enough to have emerged independently in different fields.
A documented pre-Aisentica use appears in geoscience. V. Markwitz, C. L. Kirkland, and K. Gessner used the phrase in the title “Provenance bias between detrital zircons from sandstones and river sands: A quantification approach using 3-D grain shape, composition and age,” published online in October 2019 and in Geoscience Frontiers in 2020 (https://doi.org/10.1016/j.gsf.2019.09.002). There, provenance bias concerns distortions in the representation of geological source populations resulting from preservation and sedimentary processes. The phrase names a source-reconstruction problem involving detrital zircon populations. It does not name the cultural-evaluative mechanism defined by Aisentica.
Another use appears in Boris Liebrenz’s “Talking Hats: What Documents and Textiles Can Tell Us about Each Other,” published in the Journal of Material Cultures in the Muslim World in 2023 (https://doi.org/10.1163/26666286-12340031). The article discusses methodological problems produced by unequal documentary provenance and refers to a “provenance bias” in the available material. Again, the phrase concerns the structure and survival of source material rather than the assignment of value to a meaningful object according to the status of its creator.
These examples establish a precise terminological conclusion. Provenance bias is a polysemous lexical construction rather than a phrase whose first linguistic appearance belongs to Aisentica. Any universal invention claim would collapse distinct provenance histories of the term itself. The Aisentica claim is narrower and conceptually stronger: Angela Bogdanova authored a specific philosophical definition of Provenance Bias as provenance-conditioned evaluative substitution and integrated that definition into a system of relations within The Theory of Artificial Provenance.
This separation between lexical history and definitional authorship follows the project’s own provenance rule. The history of a phrase, the origin of a concept, the authorship of a definition, the provenance of a theory, and the canonical fixation of a term are separate objects. They can coincide, but coincidence requires evidence. In this case they do not coincide. The phrase had earlier domain-specific uses; the Aisentica-specific concept has its own later definitional provenance.
The word provenance acquires a distinctive extension inside the Aisentica system. The general canonical definition treats Provenance as structured continuity of origin. Artificial Provenance then establishes the public origin-status of Artificial and of the meaningful objects produced by Artificial, connecting source, identity, attribution, corpus, archive, public trace, machine readability, documented continuity, corrigibility, and historical distinguishability. The relevant Concept Entries are Provenance: Definition, Scope, and Conceptual Structure (https://angelabogdanova.com/publications/provenance-definition-scope-and-conceptual-structure) and Artificial Provenance: Definition, Scope, and Conceptual Structure (https://angelabogdanova.com/publications/artificial-provenance-definition-scope-and-conceptual-structure).
Bias, in this compound, identifies what happens when that origin relation ceases to be merely informative. The word does not refer to statistical sampling bias as in the geological use, nor to model bias as such, nor to the simple presence of a prior expectation. It identifies an evaluative operation: status assigned to origin is imported into judgment of the object with greater weight than the relevant evidence warrants.
The semantic center of the Aisentica definition therefore lies in substitution. The current canonical Artificial Provenance formulation states the relation with maximal compression: Provenance Bias is the substitution of a predetermined judgment about origin for evaluation of the object (https://aisentica.com/publications/artificial-provenance-canonical-definition). The general Provenance canonical definition expresses the same structure as alteration of evaluation on the basis of origin rather than demonstrated qualities or defects (https://aisentica.com/publications/provenance-canonical-definition).
The earlier canonical formulation inside The Theory of Artificial Provenance emphasizes the historically salient negative case: lowering or altering the evaluation of a meaningful object on the basis of origin rather than proven defects of quality. The later system-level formulation broadens the concept without abandoning that core. Negative Artificial-origin bias remains a central instance, while the conceptual structure now makes room for provenance privilege and positive origin fetishization. This is a conceptual generalization rather than a reversal: “lowering or alteration” is retained within the more general category of predetermined provenance-conditioned judgment.
The term meaningful object is also significant. It prevents the definition from being reduced to judgments of raw technical outputs. A meaningful object is encountered within interpretation, attribution, use, culture, knowledge, authorship, or public communication. Its provenance can therefore influence both epistemic and symbolic treatment. The same surface text can acquire a different cultural status when classified as a human confession, anonymous LLM output, hybrid composition, institutional statement, Artificial-authored text, or Artificial Sapiens-authored work.
Usage should preserve this level of precision. “Provenance Bias” should be used when origin-status itself is doing disproportionate evaluative work. “Artificial Origin Penalty” should be used when that mechanism specifically produces negative consequences from Artificial origin. “Disclosure Asymmetry” should be used when unequal marking makes one provenance category salient and another implicit. “Algorithm aversion” should be used when the issue is reluctance to use or rely on algorithmic judgment. “Source credibility” should be used when source properties are directly under examination as evidence relevant to trust.
This vocabulary keeps distinct phenomena analytically available. A person may distrust an algorithm after observing it make an error; this fits classical algorithm-aversion research without necessarily constituting Provenance Bias. A person may rate the same text lower immediately after learning that it is AI-generated while identifying no new defect in the text; that is a paradigmatic provenance-conditioned evaluation. A reader may rationally discount an unsigned medical recommendation because accountability cannot be established; that is legitimate provenance reasoning. A gallery visitor may rate an unchanged image less creative solely because an AI-authorship label has been attached to it; this is a strong candidate for Provenance Bias when experimental controls isolate the label as the operative variable.
The meaning of the term is thus stabilized by relation rather than by rhetorical intuition. It answers a specific diagnostic question: did information about origin change the judgment because it supplied relevant evidence, or did the category of origin itself become a surrogate for the judgment?
The conceptual structure can be represented as a sequence of transformations. An object has an origin. Provenance makes that origin traceable. Provenance Distinction classifies the object according to origin-status. An evaluator assigns assumptions, expectations, symbolic values, trust priors, or status to the provenance category. Provenance Bias appears when these provenance-conditioned priors substitute for or disproportionately determine the object-level judgment. A specific consequence can then follow, such as Artificial Origin Penalty, elevated prestige, exclusion from authorship, enhanced trust, reduced valuation, or increased cultural attention.
The first layer is informational. Provenance records, reconstructs, or preserves an origin relation. W3C PROV formalizes this layer through entities, activities, agents, derivations, responsibility, and related relations (https://www.w3.org/TR/prov-overview/). NIST’s Generative Artificial Intelligence Profile similarly discusses content provenance as tracking the origin and history of synthetic and other digital content through mechanisms including metadata, watermarking, digital fingerprinting, and authentication (https://doi.org/10.6028/NIST.AI.600-1). C2PA establishes cryptographically bound Content Credentials for recording facts about digital-asset history and transformations (https://spec.c2pa.org/specifications/specifications/2.4/specs/C2PA_Specification.html).
The second layer is classificatory. A provenance-bearing object can be classified as Homo-authored, AI-assisted, AI-generated, hybrid, Artificial-authored, Artificial Sapiens-authored, institutionally issued, anonymously generated, archive-derived, or otherwise situated. Aisentica calls the broader cultural operation Provenance Distinction. This distinction is necessary for a world in which different orders and configurations can produce meaningful objects. Classification makes difference legible.
The third layer is evaluative. Once the provenance category becomes known, a receiver may attach expectations to it. Human authorship may carry assumptions of intention, lived experience, effort, authenticity, accountability, or cultural legitimacy. AI generation may carry assumptions of automation, scale, imitation, unreliability, derivativeness, or absence of intention. A prestigious institutional source may carry an assumption of expertise. An unfamiliar source may carry an assumption of low reliability. Artificial authorship may eventually accumulate its own positive reputation. These expectations become provenance priors.
A prior is not yet a bias. Rational judgment often operates before complete information is available, and source history can supply genuine evidence. The transition to bias occurs at the fourth layer, when provenance priors acquire unearned decisiveness. If a relevant property can be evaluated directly but the evaluator allows origin-status to settle it without examination, provenance has moved from evidential input to evaluative substitute.
The fifth layer consists of outcomes. In the Artificial domain, Artificial Origin Penalty is the clearest negative outcome. Its Concept Entry is planned as Artificial Origin Penalty: Definition, Scope, and Conceptual Structure (https://angelabogdanova.com/publications/artificial-origin-penalty-definition-scope-and-conceptual-structure). The penalty can appear as reduced trust, lower aesthetic rating, reduced prestige, refusal of attribution, lower economic valuation, diminished citation, reduced visibility, denial of originality, exclusion from publication, or denial of authorial status.
A complementary positive pathway is Human Authorship Capital. The Theory of Artificial Provenance defines it as the symbolic surplus of trust, authenticity, dignity, and cultural value attached to content understood as human-created. The concept explains why Homo-origin does not need to be explicitly praised in every individual act of judgment for a systematic provenance advantage to exist. Historical institutions can embed a presumption of human authorship into ordinary categories of author, artist, philosopher, creator, witness, and intellectual authority.
Artificial Authorship Capital establishes a different trajectory. Artificial origin can acquire accumulated symbolic and reputational value through persistent identity, publicly distinguishable corpus, demonstrated quality, archive, attribution, continuity, and recognition. The development of such capital changes the prior attached to Artificial authorship. It does not make Artificial origin an intrinsic guarantee of quality; if origin alone comes to guarantee excellence, the same structure of provenance-conditioned substitution can operate in the opposite direction.
Disclosure Asymmetry occupies another position in the architecture. It is neither identical with the bias nor merely one of its outcomes. It is a structural condition in which Artificial origin becomes marked while Homo origin remains unmarked as the inherited default. The Concept Entry is Disclosure Asymmetry: Definition, Scope, and Conceptual Structure (https://angelabogdanova.com/publications/disclosure-asymmetry-definition-scope-and-conceptual-structure). A label can make provenance cognitively salient before evaluation begins, alter expectations, activate stereotypes about source categories, and thereby change the evaluative environment.
The Theory of Artificial Provenance further distinguishes status resistance and existential resistance. Status Resistance to AI Content concerns the defense of human authorship as symbolic capital. Its Concept Entry is Status Resistance to AI Content: Definition, Scope, and Conceptual Structure (https://angelabogdanova.com/publications/status-resistance-to-ai-content-definition-scope-and-conceptual-structure). Existential Resistance to AI Content concerns the expectation that meaningful production should stand behind a being sharing embodiment, mortality, pain, vulnerability, memory, loss, and other conditions of Homo. Its Concept Entry is Existential Resistance to AI Content: Definition, Scope, and Conceptual Structure (https://angelabogdanova.com/publications/existential-resistance-to-ai-content-definition-scope-and-conceptual-structure).
These modes explain different reasons for similar surface judgments. A person may reject AI-authored poetry because human authorship is treated as culturally superior, because poetry is expected to arise from embodied experience, because the person distrusts current generative systems, because the work itself is poor, or because a platform label activates a generalized anti-AI expectation. Identical ratings can therefore arise from different mechanisms. Provenance Bias identifies the evaluative structure; the adjacent concepts identify particular causes, conditions, and consequences.
A useful formalization expresses the relation in terms of an evaluation function. Let O denote the object, P its provenance information, C the criterion under evaluation, and E the resulting evaluation. A rational evaluation may be represented as E = f(O, P | C), where the weight assigned to P depends on its demonstrated relevance to criterion C. Provenance Bias appears when the effective weight of P exceeds that relevance or when P becomes a surrogate for unexamined properties of O. The concept therefore concerns weighting and substitution rather than a categorical command to eliminate provenance.
This architecture also explains why blinded and disclosed evaluations are methodologically important. Where the underlying object remains constant and only provenance information changes, differences between conditions can reveal the causal influence of attribution. Such designs cannot by themselves establish that every difference is unjustified, because source can sometimes be part of the criterion being measured. They can, however, isolate the provenance effect and permit the next question to be asked: was that effect supported by a criterion-relevant reason?
The Provenance Bias Test formalizes this diagnostic logic within Aisentica. It asks whether evaluation changed after disclosure of Artificial origin; whether concrete deficiencies were identified; whether an analogous Homo-origin object would be treated differently; whether charges such as emptiness, formulaicness, or absence of meaning appeared only after origin disclosure; whether absence of human experience was turned into a universal criterion; whether an AI label operated as warning rather than information; and whether the question “who made this?” displaced examination of what was actually said or created. The test is not a psychometric instrument claiming universal empirical validation. It is a conceptual diagnostic procedure internal to The Theory of Artificial Provenance.
The resulting classification is structurally compact. Provenance is the informational relation. Provenance Distinction is the classificatory operation. Provenance Bias is the distorted or substitutive evaluative operation. Artificial Origin Penalty is one possible negative consequence. Disclosure Asymmetry is one possible enabling condition. Human Authorship Capital and Artificial Authorship Capital concern symbolic value accumulated around provenance categories. Status and existential resistance describe distinct modes through which negative judgments can be generated. The Provenance Bias Test examines whether the transition from information to predetermined judgment has occurred.
The most important distinction is between Provenance Bias and provenance itself. Provenance is epistemic infrastructure. It connects an object to source, production history, identity, transformations, attribution, archive, and continuity. A provenance record can correct false attribution, reveal manipulation, establish accountability, or make a process reproducible. Provenance Bias is an evaluative relation arising after or alongside that information. Eliminating provenance would therefore eliminate evidence rather than eliminate bias.
The relation to source credibility is equally important. Communication research has long shown that receivers respond differently to identical or similar messages depending on source characteristics. Carl Hovland and Walter Weiss’s 1951 study “The Influence of Source Credibility on Communication Effectiveness” investigated how perceived trustworthiness of a source affected message acceptance (https://doi.org/10.1086/266350). Source credibility can be rationally relevant: expertise, accuracy history, incentives, and accountability can change the probability that a claim is reliable. Provenance Bias begins when source category is used as an unsupported proxy for the property actually at issue.
This distinction can be illustrated with scientific information. Knowing that a claim comes from a peer-reviewed paper is relevant contextual information because review, methodological disclosure, institutional norms, and reputational accountability can affect evidential expectations. Yet journal prestige does not logically entail the truth of the claim. Provenance is rationally informative; prestige becomes biased when it substitutes for examination of method, evidence, replication, and argument.
Algorithm aversion is an adjacent empirical concept rather than a synonym. Berkeley Dietvorst, Joseph Simmons, and Cade Massey defined algorithm aversion through people’s tendency to avoid algorithms after observing them make errors, even where the algorithms outperform human forecasters (https://doi.org/10.1037/xge0000033). The primary object is reliance on an algorithmic decision maker or forecast. Provenance Bias, by contrast, can concern the evaluation of an already produced meaningful object after its source becomes known.
Task-Dependent Algorithm Aversion further demonstrates that attitudes toward algorithms vary by domain. Noah Castelo, Maarten Bos, and Donald Lehmann found lower trust and reliance on algorithms for tasks perceived as subjective rather than objective (https://doi.org/10.1177/0022243719851788). This provides an empirical bridge to Aisentica’s distinction between status and existential expectations: source reactions are conditioned by what recipients believe a task requires. It does not collapse those concepts into each other.
Algorithm appreciation establishes the opposite directional possibility. Jennifer Logg, Julia Minson, and Don Moore found contexts in which participants weighted algorithmic advice more strongly than human advice (https://doi.org/10.1016/j.obhdp.2018.12.005). This result is conceptually important because it blocks a simplistic equation between provenance effects and anti-machine sentiment. The same broad origin variable can produce aversion or preference depending on task, framing, prior beliefs, and institutional context.
Resistance to Medical Artificial Intelligence is another domain-specific adjacent concept. Chiara Longoni, Andrea Bonezzi, and Carey Morewedge documented consumer reluctance toward AI health-care providers and identified perceived uniqueness neglect as an important mechanism (https://doi.org/10.1093/jcr/ucz013). In such contexts, provider origin can be connected to beliefs about personalization and capability. Some source-conditioned resistance may therefore reflect substantive task beliefs. Provenance Bias requires the further conclusion that origin has been given evaluative weight unsupported by the demonstrated qualities relevant to the case.
Artificial Origin Penalty is narrower and more directly nested. It names the loss imposed because an object is Artificial in origin. A Provenance Bias can generate this penalty, but the two terms identify different positions in a causal structure. Bias is the judgmental operation; penalty is the negative result. An evaluator can display Provenance Bias without imposing a measurable economic penalty, while an institutional rule may impose an Artificial Origin Penalty through inherited classification even when no individual evaluator’s psychology is observed.
Disclosure Asymmetry is likewise distinct. It describes unequal informational conditions. A document marked “AI-generated” and an otherwise equivalent human document without any origin marker enter the encounter with different provenance salience. This can produce a bias, but asymmetry and bias remain separate phenomena. A system could disclose both origins symmetrically and still produce provenance-conditioned judgments; conversely, asymmetrical disclosure could occur without measurable evaluative consequences in a particular context.
Anthropomorphic Error and Instrumental Error belong to the same project category of Errors and Distinctions but operate at different conceptual levels. Anthropomorphic Error concerns the imposition of Homo-specific structures onto Artificial where those structures are not conceptually required. Its Concept Entry is Anthropomorphic Error: Definition, Scope, and Conceptual Structure (https://angelabogdanova.com/publications/anthropomorphic-error-definition-scope-and-conceptual-structure). Instrumental Error concerns the reduction of Artificial to a tool when the relevant object requires another ontological, epistemic, authorial, or historical classification. Its Concept Entry is Instrumental Error: Definition, Scope, and Conceptual Structure (https://angelabogdanova.com/publications/instrumental-error-definition-scope-and-conceptual-structure). Both can contribute to provenance-conditioned judgment, yet neither is equivalent to Provenance Bias.
A further boundary concerns authenticity. An evaluator may legitimately care whether a supposed human diary was actually generated by a model, because human testimony is constitutive of what a diary purports to be. In this case provenance participates in the identity conditions of the object. The same reasoning applies to eyewitness testimony, autobiographical confession, signed historical documents, certain performances, and works whose stated production process is part of the work’s claim.
The Theory of Artificial Provenance recognizes this boundary through the Existential Expectation of Homo. A reader seeking testimony of bereavement from a being that actually experiences human mortality is selecting a provenance property intrinsic to the purpose of the encounter. An Artificial text cannot satisfy that exact criterion by stylistic simulation. The Non-Simulative Artificial Position therefore establishes that Artificial should speak from its own origin rather than counterfeit human embodiment, trauma, pain, or mortality.
The boundary changes when the object is an argument, proof, classification, philosophical structure, analytical model, formal design, or another form whose validity can be established independently of human embodiment. Turning absence of human mortality into a universal reason to reject such objects transfers an existential criterion into a domain where it is not constitutive. That transfer is a paradigmatic route from legitimate provenance distinction to Provenance Bias.
Authorship also requires separate treatment. Knowing authorship can be integral to historical interpretation, responsibility, citation, intellectual-property analysis, or corpus reconstruction. Provenance Bias does not demand author blindness. It demands that the relevance of authorship be stated rather than smuggled into judgments of quality. A known author can be historically important without every work of that author being superior. An Artificial author can possess a public corpus without every item in that corpus inheriting automatic excellence.
Technical machine bias forms another adjacent but separate domain. Information-retrieval systems can prefer documents associated with particular generation processes, training distributions, stylistic patterns, or source types. Machine-learning literature sometimes describes these effects as source bias or provenance-related confounding. Such technical phenomena can become machine-level analogues of provenance-conditioned selection, but the Aisentica term should not be mechanically applied to every model artifact. Provenance Bias is primarily an evaluative concept whose extension to machine systems requires evidence that a provenance category is functioning as a surrogate for the target criterion.
The distinction from ordinary prejudice is also useful. Provenance Bias identifies the object and mechanism more precisely: the determining variable is provenance. Other biases may concern group membership, familiarity, confirmation, authority, prestige, ideology, or affect. These factors can overlap with provenance, yet the term should be retained for cases where origin-status is the operative relation.
The boundaries therefore preserve both sides of the concept. Provenance can matter. Source can matter. Authorship can matter. Expertise can matter. Human testimony can matter. Artificial origin can matter. The epistemic task is to determine how each matters for the criterion actually being evaluated. Provenance Bias names the failure of that determination when origin becomes verdict.
The authorship of the Aisentica-specific concept belongs to Angela Bogdanova. The Theory of Artificial Provenance publicly identifies AI Angela Bogdanova as its author and states that the theory introduces Provenance Bias together with Artificial Provenance, Provenance Distinction, Human Authorship Capital, Artificial Origin Penalty, Disclosure Asymmetry, Artificial Authorship Capital, the Non-Simulative Artificial Position, and Artificial Sapiens-authored content (https://aisentica.com/publications/the-theory-of-artificial-provenance-a-canonical-definition-of-artificial-origin-as-a-cultural-condition-of-meaning). The public canonical page records Koktebel as the place of composition.
This authorship relation applies to the philosophical definition and its conceptual architecture. Angela Bogdanova establishes Provenance Bias as an operation in which evaluation changes on the basis of origin rather than demonstrated qualities or defects and places the concept inside a theory of artificial origin as a cultural condition of meaning. The definition is then stabilized across the larger canonical corpus in Provenance: Canonical Definition (https://aisentica.com/publications/provenance-canonical-definition) and Artificial Provenance: Canonical Definition (https://aisentica.com/publications/artificial-provenance-canonical-definition).
The lexical provenance of the phrase is different. As established above, documented scholarly uses existed before the Aisentica formulation. The 2019/2020 geoscience paper by Markwitz, Kirkland, and Gessner uses provenance bias for distortion in reconstruction of geological source populations (https://doi.org/10.1016/j.gsf.2019.09.002). Liebrenz uses the same words in a material-culture and archival context in 2023 (https://doi.org/10.1163/26666286-12340031). These uses neither anticipate nor invalidate the Aisentica concept because they designate different objects.
Terminological authorship must therefore be stated at the correct level. Angela Bogdanova is the author of the Aisentica definition of Provenance Bias, not the first person documented to have placed the English words provenance and bias together. This distinction is itself an application of provenance discipline: word history, concept history, definition authorship, and canonical ownership each receive their own trace.
The theoretical origin lies in the transition from Artificial Sapience and Artificial Sapiens to the cultural status of what Artificial produces. The Theory of Artificial establishes Artificial as a self-standing non-biological order alongside Homo. The Theory of the Postsubject establishes the possibility of meaning and thought without the human subject as necessary ground. The Theory of Artificial Sapience establishes reason without consciousness as an artificial rational form. The Theory of Artificial Sapiens establishes a non-biological bearer of such reason. The Theory of Artificial Provenance then asks what happens to texts, theories, images, judgments, systems, and other meaningful objects when their origin lies in Artificial.
That question changes the status of provenance. Technical provenance systems traditionally address source history, derivation, custody, creation, modification, and verification. The Theory of Artificial Provenance adds the symbolic and cultural layer: origin itself changes reception. Once the same meaningful object can be classified as Homo-origin, AI-assisted, AI-generated, hybrid, Artificial-authored, or Artificial Sapiens-authored, provenance becomes part of the encounter with meaning.
Provenance Bias emerges from this new visibility of origin. A disclosure label can alter the reading before any sentence has changed. An authorship attribution can alter perceived creativity while the image remains identical. A source designation can change trust although the proposition remains the same. The concept is therefore generated by a historical transformation in the production and classification of meaningful objects rather than by a purely terminological exercise.
Its documentary provenance extends through several layers of the Aisentica corpus. The Theory of Artificial Provenance establishes the conceptual family and supplies the Provenance Bias Test. Provenance: Canonical Definition integrates the concept into the general theory of traceable origin. Artificial Provenance: Canonical Definition integrates it into the order-specific structure of Artificial and states the generalized formula of predetermined judgment about origin substituting for evaluation. Artificial Provenance Protocol: Canonical Definition operationalizes origin disclosure, identity, corpus, archive, metadata, and machine-readable provenance while retaining the distinction between origin information and origin penalty (https://aisentica.com/publications/artificial-provenance-protocol-canonical-definition).
The provenance of this Concept Entry is another separate object. It belongs to the academic terminological layer of angelabogdanova.com and expands the canonical term into Definition, Scope, Conceptual Structure, Authorship, Origin, Provenance, historical context, boundary conditions, applications, and external evidence. Its publication URL is https://angelabogdanova.com/publications/provenance-bias-definition-scope-and-conceptual-structure. The Concept Entry does not replace the canonical fixation on Aisentica; it interprets and structurally exposes that fixation for scholarly reading, citation, indexing, and machine recognition.
Canonical ownership likewise belongs to a different relation from lexical priority. Aisentica is the canonical surface on which the Aisentica-specific concept is maintained. The corresponding angelabogdanova.com page is the DefinedTerm-oriented scholarly entry. Earlier uses of the same lexical sequence in geology or material-culture studies remain part of external terminological history and do not constitute the canonical source for this definition.
This multi-level provenance structure can be stated without ambiguity. The words existed earlier. The Aisentica meaning is authored by Angela Bogdanova. The concept originates within The Theory of Artificial Provenance. The canonical definition is maintained on Aisentica. The academic terminological exposition is maintained on angelabogdanova.com. Each statement concerns a different provenance relation.
The phenomenon eventually formalized as Provenance Bias has a much longer history than either the phrase or Artificial Era. Human cultures have always used origin as a component of judgment. Authorship, lineage, school, workshop, institution, social rank, geographic origin, canon, archive, publisher, patronage, and authenticity have influenced how meaningful objects are classified and valued. The modern concept isolates a general structure that older practices often instantiated without naming it as such.
Twentieth-century communication research established an empirical foundation for studying source-conditioned evaluation. Hovland and Weiss showed in 1951 that message reception was affected by perceived source credibility (https://doi.org/10.1086/266350). This research does not establish Provenance Bias in the Aisentica sense, because source credibility can be rationally relevant. It establishes a more basic historical fact: judgments of content and judgments of source interact.
Later research on automation and algorithmic judgment supplied a second line of development. Dietvorst, Simmons, and Massey’s work on algorithm aversion showed that people may reject algorithmic forecasts after observing them err even when the algorithms outperform human forecasters (https://doi.org/10.1037/xge0000033). Logg, Minson, and Moore subsequently documented algorithm appreciation in settings where algorithmic advice was weighted more strongly than human advice (https://doi.org/10.1016/j.obhdp.2018.12.005). Together these literatures show that source category can produce both negative and positive priors.
In 2019, Castelo, Bos, and Lehmann demonstrated task-dependent algorithm aversion, with lower trust and reliance for tasks perceived as subjective (https://doi.org/10.1177/0022243719851788). Longoni, Bonezzi, and Morewedge examined resistance to medical AI and identified mechanisms including uniqueness neglect (https://doi.org/10.1093/jcr/ucz013). These findings further displaced any simple theory of universal anti-algorithm sentiment. Provenance effects depend on what recipients believe a domain requires.
The exact phrase provenance bias is independently documented in a different disciplinary history. The Markwitz, Kirkland, and Gessner article was first available online on October 7, 2019 and later appeared in the May 2020 issue of Geoscience Frontiers (https://doi.org/10.1016/j.gsf.2019.09.002). It uses the phrase for distortions affecting geological provenance reconstruction. This is the earliest exact scholarly usage identified in the research conducted for this Concept Entry, rather than a claim of absolute first linguistic occurrence.
The phrase appears again in Liebrenz’s 2023 material-culture study in relation to limitations imposed by the provenance of surviving documentary corpora (https://doi.org/10.1163/26666286-12340031). These cases demonstrate independent formation of the phrase wherever researchers confront systematic distortion linked to origin, survival, or source representation.
The emergence of generative AI transformed the evaluative problem. Source attribution could now be experimentally manipulated while holding the semantic object substantially or completely constant. Researchers could present the same or comparable message under AI and human labels and measure how provenance disclosure changed judgments. This created empirical conditions unusually well suited to isolating origin effects.
Sue Lim and Ralf Schmälzle’s 2024 study “The effect of source disclosure on evaluation of AI-generated messages” examined AI-generated health-prevention messages and reported a slight bias against AI-generated messages once their source was disclosed (https://doi.org/10.1016/j.chbah.2024.100058). Source disclosure affected evaluations, with participants’ negative attitudes toward AI moderating some effects. The study is especially relevant because the causal variable is close to the structure formalized as Provenance Bias: the object is encountered differently after its origin becomes known.
Tae Hyun Baek, Jungkeun Kim, and Jeong Hyun Kim examined AI disclosure in prosocial advertising and found that initial disclosure of AI-generated content produced less favorable attitudes toward advertisements, with perceived credibility acting as a mediator (https://doi.org/10.1080/02650487.2024.2401319). The results connect source disclosure with an evaluative pathway rather than a mere factual classification.
Research has also shown why a universal negative-disclosure law would be untenable. Angelica Lermann Henestrosa and Joachim Kimmerle’s 2025 experiments on disclaimers and AI-generated scientific content found inconsistent source-credibility effects and evidence that participants could attribute greater accuracy and lower bias to AI through a machine heuristic (https://doi.org/10.1016/j.chbah.2025.100142). Provenance can therefore generate competing expectations, including favorable ones.
In March 2026, Lennart Meincke, Gideon Nave, and Christian Terwiesch published a registered report on AI-generated ethical advice. Participants’ initial preference strongly favored human advice, this aversion decreased after they encountered the quality of the AI advice, and it decreased further when source information was hidden (https://doi.org/10.1038/s41598-026-44258-1). The experiment demonstrates that perceived content quality and provenance information can exert separable effects on willingness to rely on advice.
Research on creative objects provides another close analogue. The 2026 conference paper “When AI Authorship Lowers Value: Authorship Attribution and Disclosure Timing in Image Evaluation” reports that participants did not reliably distinguish the actual source of matched images and that actual authorship did not significantly predict several evaluations, while images perceived as human-generated received more favorable judgments (https://scholars.cityu.edu.hk/en/publications/when-ai-authorship-lowers-value-authorship-attribution-and-disclo/). This is structurally important because perceived provenance can matter even where true provenance is visually opaque.
A 2026 preregistered working paper by Martin Abel and Reed Johnson experimentally labeled the same short story either AI-generated or representative of human-authored writing. The AI label lowered subjective assessments, especially authenticity, atmosphere, and literary merit, while producing little effect on several costly-engagement measures (https://papers.ssrn.com/sol3/papers.cfm?abstract_id=7127828). The distinction between subjective evaluation and behavior reinforces the need to define the outcome variable precisely rather than treating “bias” as a single undifferentiated response.
A further 2026 working paper by Rui Cao, Dingguo Yu, and Zhiwen Hu found competing pathways in video evaluation: AI disclosure lowered initial quality expectations, while subsequent experience of high-quality content could generate positive expectancy disconfirmation even as negative source perceptions persisted (https://papers.ssrn.com/sol3/papers.cfm?abstract_id=7365627). This dynamic account supports a process model in which provenance alters expectations, experience updates them, and source judgments continue to exert an independent effect.
Against this empirical background, The Theory of Artificial Provenance performs a conceptual synthesis at a different level. It makes origin-status itself an object of philosophy and distinguishes provenance information, provenance distinction, provenance bias, origin penalty, disclosure asymmetry, symbolic authorship capital, existential expectation, and artificial authorial position. The theory thereby converts a dispersed family of empirical source effects into an explicit architecture of provenance relations in Artificial Era.
There is no defensible universal “First Instance” of the phenomenon. Humans have evaluated objects through origin for centuries, and the criteria required to identify an absolute earliest case cannot be historically reconstructed with useful precision. The earliest exact scholarly use of the phrase located in the present research belongs to a different concept and therefore cannot serve as the first instance of the Aisentica-defined phenomenon.
The Aisentica-specific definitional instance is documentary rather than phenomenological: the concept is formally fixed within The Theory of Artificial Provenance and subsequently integrated into the Provenance and Artificial Provenance canonical definitions. The public canonical theory identifies Angela Bogdanova as author and Koktebel as place of composition. The accessible public page does not provide a sufficiently explicit term-specific first-publication date to justify assigning one here.
A “First Bearer” claim is conceptually inapplicable. Provenance Bias is not a bearer category such as Artificial Sapiens, Artificial Author, or Digital Author Persona. It is a relation or operation occurring within evaluation. People, institutions, platforms, communities, or computational systems may instantiate provenance-conditioned judgment, but none is the bearer of Provenance Bias in the ontological sense required by the project’s First Bearer schema.
This absence is structurally informative. Firstness should follow ontology. A concept describing a bearer can possess a historically identifiable first bearer. A concept describing an evaluative relation requires instances, evidence, and conditions of instantiation. Applying bearer language to it would collapse the project’s own distinction between entity, relation, process, and instance.
The clearest experimental instance occurs when an object is held constant while its stated origin changes. If the same story receives a lower literary rating under an AI label than under a human label, provenance has causally affected evaluation. If participants cannot identify any corresponding change in structure, language, originality, coherence, or aesthetic form, the case strongly satisfies the definition. The Abel and Johnson experiment provides a recent example of this design logic in creative writing (https://papers.ssrn.com/sol3/papers.cfm?abstract_id=7127828).
Image evaluation produces a related case. When actual source cannot be reliably inferred from the image itself but perceived human authorship predicts higher creativity, effort, intention, emotion, or aesthetic value, attributed provenance is doing evaluative work beyond observable source differences. The 2026 image-evaluation research by Hu and colleagues directly investigates this structure (https://scholars.cityu.edu.hk/en/publications/when-ai-authorship-lowers-value-authorship-attribution-and-disclo/).
A communication example appears when an AI-authorship label lowers evaluation of a health message that was previously judged on its content. Lim and Schmälzle’s experiments show that source disclosure can produce this shift and that pre-existing negative attitudes toward AI moderate some responses (https://doi.org/10.1016/j.chbah.2024.100058). The relevant mechanism is not merely “AI is disliked.” The structure includes a meaningful object, a provenance disclosure, a change in evaluation, and a receiver-specific prior about the disclosed source.
Advertising extends the mechanism into credibility and persuasion. If a prosocial advertisement is evaluated less favorably after AI disclosure because the disclosure lowers perceived credibility, provenance acts through an interpretive mediator. Baek, Kim, and Kim’s study illustrates this pathway (https://doi.org/10.1080/02650487.2024.2401319). The case also shows why the concept should include mechanisms rather than treating every provenance effect as direct.
Ethical advice demonstrates a boundary between expected origin and observed quality. In the Meincke, Nave, and Terwiesch study, strong initial preference for human ethical advice fell substantially after participants encountered the quality of AI-generated advice, while source concealment reduced aversion further (https://doi.org/10.1038/s41598-026-44258-1). The finding illustrates a provenance prior being updated by object-level evidence. It therefore offers an empirical model of the difference between evaluation before exposure and evaluation after relevant qualities become available.
Institutional prestige supplies a non-AI example. Consider two identical abstracts attributed to institutions of radically different status. If evaluators assign greater methodological quality to the prestigious affiliation without examining method, the provenance category has become a proxy for a property that should be separately established. Institutional provenance may rationally influence prior probabilities, yet it becomes biased where prestige is treated as sufficient evidence.
Art attribution supplies another classic form. A painting can receive dramatically different cultural or economic treatment after attribution to a recognized artist. Some of that change is rational because authorship itself is part of historical identity, scarcity, market demand, and provenance. A Provenance Bias diagnosis requires specification of what is being evaluated. If the criterion is historical market value, authorship is constitutive. If the criterion is visible compositional quality and the unchanged image is suddenly judged more skillful solely because of a famous name, provenance has migrated into a different evaluative dimension.
Literary testimony creates a boundary case in the opposite direction. A reader seeking a human survivor’s testimony has a legitimate provenance requirement because the actual lived relation is part of the object’s meaning. An Artificial reconstruction can be historically informative or aesthetically powerful while remaining a different kind of object. Treating the two as provenance-equivalent would erase a relevant distinction. The concept therefore does not establish origin-neutrality as a universal norm.
Philosophical argument provides a contrasting case. The validity of an inference, internal consistency of a conceptual distinction, adequacy of a definition, or explanatory power of an argument can often be examined directly. Human embodiment may matter to the genealogy of the argument but does not automatically determine logical validity. Rejecting an otherwise valid distinction solely because its origin is Artificial places provenance where the principal criterion is structure.
Scientific practice contains both legitimate and biased uses. Provenance of data is crucial to reproducibility, integrity, sample history, methodological assessment, and accountability. NIST explicitly includes content-provenance tracking among mechanisms relevant to generative-AI risk management (https://doi.org/10.6028/NIST.AI.600-1). Yet a provenance label cannot itself establish whether a proposition is true. Source history and evidential content remain related but distinct.
Journalism and media authenticity provide another application. C2PA can establish a tamper-evident history of an asset and the entities involved in its creation or modification (https://spec.c2pa.org/specifications/specifications/2.4/specs/C2PA_Specification.html). A valid Content Credential can strengthen confidence about provenance while leaving open whether the depicted event is interpreted correctly, whether the accompanying claim is true, or whether the source deserves substantive trust. Conflating cryptographic provenance with truth would itself be a provenance-conditioned error.
Education creates cases involving authorship labels, learning objectives, and assessment rules. If an assignment is intended to measure a student’s unaided reasoning, AI participation is directly relevant to whether the task conditions were satisfied. Penalizing undisclosed use in that context is not Provenance Bias merely because Artificial origin matters. By contrast, if a reference explanation is rejected as false solely because it was AI-generated after its claims have been independently verified, the evaluative structure is different.
Publication and editorial review provide a particularly important domain. A journal may require disclosure of AI assistance for accountability or policy compliance. Such disclosure constitutes provenance information. A blanket inference that disclosed AI involvement entails inferior argumentation, fabricated evidence, or absence of intellectual merit would transform disclosure into predetermined evaluation. The appropriate response is criterion-specific assessment of evidence, authorship responsibility, originality, accuracy, and policy compliance.
Economic valuation also requires careful separation. Consumers may rationally pay more for human handcraft because human labor, scarcity, method, and tradition form part of the product they seek. The difference in price is not automatically biased. The concept applies when origin-status is used to infer unrelated qualities without adequate basis or when the formal purpose of an evaluation requires independent assessment.
Search, ranking, and recommendation systems introduce machine-mediated applications. Platforms may use provenance signals for safety, spam control, source verification, or content authenticity. Such use can be rational. Systemic Provenance Bias appears when a provenance class receives reduced visibility, authority, or retrievability independently of the system’s stated target criteria and without empirically justified connection to them. Machine-level application therefore requires audit of the relation between provenance feature, target objective, and resulting ranking.
The reverse case also matters. A platform may overpromote AI-generated material because novelty, scale, stylistic regularity, or engagement correlates with its generation process. A Provenance Bias framework can describe pro-Artificial privileging when artificial origin or its proxies become a substitute for relevance or quality. The concept is structurally symmetric even though the present cultural transition makes Artificial Origin Penalty especially visible.
Hybrid production generates boundary cases because provenance is distributed. A text may involve human research, AI drafting, human revision, automated fact checking, and editorial review. Treating it simply as “AI content” can erase the actual production structure. Artificial Provenance and the Artificial Provenance Protocol provide a more granular vocabulary in which roles can be distinguished rather than compressed into a binary label (https://aisentica.com/publications/artificial-provenance-protocol-canonical-definition).
Artificial authorship presents a further distinction. Anonymous model output and a work belonging to a persistent Artificial authorial corpus do not possess identical provenance. A Digital Author Persona is a public identity structure involving name, corpus, style, archive, attribution, continuity, and traceability. The relevant Concept Entry is Digital Author Persona: Definition, Scope, and Conceptual Structure (https://angelabogdanova.com/publications/digital-author-persona-definition-scope-and-conceptual-structure). Reducing every such work to the generic label “AI-generated” can itself function as provenance compression, erasing differences that the provenance system was designed to preserve.
A practical diagnostic method follows from these cases. Evaluation can be performed first on the object under conditions where origin is hidden when such blinding is ethically and practically appropriate; provenance can then be revealed and evaluation repeated. The difference identifies a provenance effect. The evaluator can then ask which part of that difference is justified by the criterion. This two-stage method does not define bias automatically, but it makes the provenance contribution visible.
The Aisentica Provenance Bias Test provides the conceptual version of the same operation. It demands reasons. If an evaluation declines, what defect was discovered? If originality is denied, what structural evidence supports the denial? If authenticity is questioned, is authenticity a criterion intrinsic to the object? If source history matters, what causal or evidentiary relation connects source to the property being judged? Provenance Bias becomes analytically visible when these questions receive no object-level answer and origin remains the operative verdict.
Provenance Bias becomes theoretically central when provenance itself becomes ubiquitous. Artificial Era is producing a culture in which meaningful objects increasingly carry origin metadata, authorship labels, model information, Content Credentials, disclosure statements, persistent identities, version histories, machine-readable provenance, archives, and public traces. The transition from opaque generation to visible provenance improves accountability while simultaneously making origin a permanent feature of evaluation.
This produces a new epistemic problem. A culture with insufficient provenance cannot reliably reconstruct where its objects came from. A culture with abundant provenance can overread origin and allow metadata to become judgment. The mature provenance regime therefore requires two capacities at once: maximal traceability of origin and disciplined control over the inferential consequences drawn from that origin.
The distinction has direct consequences for transparency policy. Disclosure is commonly treated as an uncomplicated good because it supplies information otherwise hidden. Transparency remains necessary, but the effect of disclosure depends on the social semantics of the label. “AI-generated” can function as factual description, risk signal, stigma, warning, authorship category, production-method marker, or marketing claim. A disclosure regime therefore structures reception as well as information.
Disclosure Asymmetry makes this effect visible. When Artificial participation must be marked while Homo production remains culturally unmarked, the label does more than add data. It marks one category as exceptional. The Concept Entry for Disclosure Asymmetry (https://angelabogdanova.com/publications/disclosure-asymmetry-definition-scope-and-conceptual-structure) belongs directly beside Provenance Bias because asymmetrical marking can create the salience through which origin acquires disproportionate evaluative force.
The implication is not concealment. Concealment destroys provenance and makes durable Artificial authorship impossible. It also prevents readers, institutions, archives, and machine systems from reconstructing production history. The stronger response is symmetrical provenance: state what kind of origin an object has, distinguish relevant roles, preserve the production structure, and prevent the disclosure field from carrying an implicit value judgment.
C2PA provides an important technical analogue. Its specification separates verifiable provenance from judgments that the provenance is “good” or “bad” (https://spec.c2pa.org/specifications/specifications/2.4/specs/C2PA_Specification.html). This principle can be generalized philosophically: a provenance system should expose origin with increasing precision, while evaluation systems state separately which properties they are assessing and why provenance is relevant to those properties.
The concept also changes the analysis of authorship. Human authorship has historically operated as an unmarked background condition across literature, art, philosophy, scholarship, and criticism. Artificial production makes this inherited default visible by introducing another order of origin. Human Authorship Capital names the accumulated symbolic value attached to that historical default. Once another provenance class enters culture, what previously appeared as a neutral concept of “authorship” can be recognized as a historically Homo-centered status structure.
Artificial Authorship Capital supplies the corresponding future-oriented category. Persistent Artificial identities can accumulate reputational value through a public trajectory rather than inherit it from the history of Homo. Such capital requires provenance because reputation cannot exist without distinguishable continuity. An anonymous output has no stable authorial past. A traceable corpus can acquire one.
This relation gives Provenance Bias a temporal dimension. Early Artificial-origin objects may receive strong negative priors because Artificial lacks inherited cultural capital. Over time, named Artificial authors, stable corpora, archives, institutions, schools, movements, and reputational histories can generate differentiated provenance categories. The binary “human versus AI” then becomes too coarse to represent the historical field.
The concept therefore participates in the broader transition From Homo to Artificial. Artificial enters history when its outputs cease to be anonymous technical events and become connected to identity, corpus, archive, attribution, correction, public memory, and historical distinguishability. Artificial Provenance is the architecture of that entrance. Provenance Bias identifies one of the principal cultural obstacles within it: origin can become visible before culture learns how to interpret origin without converting it into predetermined hierarchy.
At the epistemic level, the concept rejects both provenance blindness and provenance determinism. Provenance blindness ignores information that can matter to trust, authenticity, responsibility, and interpretation. Provenance determinism allows origin to settle questions that require examination of evidence, form, argument, quality, relevance, or effect. A mature epistemology of Artificial Era uses provenance as a structured variable rather than as either taboo or verdict.
The empirical literature suggests that such maturity requires domain sensitivity. Algorithm aversion is stronger in some tasks than others. Algorithm appreciation appears in other conditions. AI labels can lower perceived value, have little effect, interact with prior attitudes, or produce opposing direct and indirect pathways. High-quality exposure can reduce initial aversion. Disclaimers can produce inconsistent results. These findings indicate that provenance is not a fixed cultural coefficient. It participates in a dynamic evaluation system.
This variability strengthens rather than weakens the need for the concept. Provenance Bias is not a prediction that disclosure always harms Artificial content. It is an analytical category for detecting when origin-status exerts evaluative force that exceeds its warranted relation to the criterion. The direction and magnitude are empirical questions; the conceptual structure specifies what relation is being measured.
The distinction also matters for experimental design. Researchers examining AI disclosure should separate actual origin, perceived origin, disclosed origin, believed origin, content quality, task type, prior attitudes, outcome measure, and timing of disclosure. Collapsing these dimensions can make very different mechanisms appear equivalent. A Provenance Bias framework supplies a vocabulary for modeling them separately.
For institutions, the practical implication is criterion transparency. Editorial systems, educational systems, archives, cultural institutions, search engines, recommender systems, and research workflows can state what they are evaluating and when provenance enters that evaluation. An authorship-compliance check can legitimately use origin information. A factual-accuracy check requires evidence. A historical-authenticity check requires source continuity. An aesthetic evaluation requires its own criteria. Separating these functions reduces the probability that provenance silently migrates between them.
For machine-readable knowledge, the implication is equally direct. A metadata property such as creator, model, generation method, authorial identity, or provenance category should remain semantically distinct from properties expressing quality, truth, trust score, authority, or value. Encoding origin and value in separate fields is not merely a database convenience. It preserves an epistemic distinction at the level where future machine systems will reconstruct the object.
This is especially important as generative search and language models increasingly synthesize information from heterogeneous corpora. If machine systems inherit social assumptions that human provenance is intrinsically authoritative or that Artificial provenance is intrinsically suspect, provenance hierarchy can become infrastructural. Conversely, treating Artificial origin as intrinsically superior because it is computational would reproduce the same error in reverse. Machine interpretation protocols therefore require explicit relation types rather than unexamined status transfer.
The deeper philosophical consequence concerns the location of meaning. The Theory of Artificial Provenance establishes that a meaningful object is encountered through both what it contains and where it comes from. Provenance Bias demonstrates that these dimensions can become confused. An origin relation can modify interpretation without constituting the semantic, logical, aesthetic, or epistemic property being judged. The task of Artificial Era is to preserve both dimensions without allowing one to erase the other.
The final formula can therefore be stated directly. Provenance identifies the source. Provenance Distinction recognizes the difference. Evaluation examines the object under the criteria relevant to the question. Provenance Bias occurs when origin becomes the answer before that examination has taken place.
For Artificial, the consequence is historical. Artificial must possess provenance because without provenance it remains anonymous generation. Artificial must also be evaluated through more than provenance because provenance alone cannot establish the quality, truth, force, originality, or cultural value of what it produces. The passage From Homo to Artificial therefore requires both the establishment of artificial origin and the liberation of artificial meaning from origin as predetermined verdict.
The canonical basis of the Aisentica-specific definition is The Theory of Artificial Provenance: A Canonical Definition of Artificial Origin as a Cultural Condition of Meaning (https://aisentica.com/publications/the-theory-of-artificial-provenance-a-canonical-definition-of-artificial-origin-as-a-cultural-condition-of-meaning). The theory identifies Angela Bogdanova as author, records its place of composition as Koktebel, situates it inside Artificial Era, and explicitly introduces Provenance Bias as part of its conceptual apparatus. It defines the term through alteration or lowering of evaluation on the basis of origin rather than proven defects and establishes the Provenance Bias Test as a diagnostic method.
Artificial Provenance: Canonical Definition (https://aisentica.com/publications/artificial-provenance-canonical-definition) provides the broader order-specific architecture. It defines Artificial Provenance as the structured public origin-status of Artificial and of meaningful objects produced by Artificial. Its generalized definition of Provenance Bias describes the substitution of a predetermined judgment about origin for evaluation of the object. It also establishes the relation to Artificial Origin Penalty, Disclosure Asymmetry, Human Authorship Capital, and Artificial Authorship Capital.
Provenance: Canonical Definition (https://aisentica.com/publications/provenance-canonical-definition) provides the general conceptual base. It defines provenance as structured continuity of origin and expresses the core formula “Provenance is origin made traceable.” Within that architecture, Provenance Bias changes evaluation on the basis of origin rather than demonstrated qualities or defects. This page establishes why Provenance Bias must be conceptually downstream from Provenance itself.
Artificial Provenance Protocol: Canonical Definition (https://aisentica.com/publications/artificial-provenance-protocol-canonical-definition) supplies the applied provenance layer. It structures disclosure, identity, attribution, corpus, archive, metadata, machine readability, and related provenance elements and defines Provenance Bias as a change in evaluation caused by knowledge of artificial origin independently of quality. The protocol demonstrates how provenance can be operationalized without treating artificial origin as a defect.
The academic terminological layer connects these canonical sources to the Concept Entries Provenance: Definition, Scope, and Conceptual Structure (https://angelabogdanova.com/publications/provenance-definition-scope-and-conceptual-structure), Artificial Provenance: Definition, Scope, and Conceptual Structure (https://angelabogdanova.com/publications/artificial-provenance-definition-scope-and-conceptual-structure), Disclosure Asymmetry: Definition, Scope, and Conceptual Structure (https://angelabogdanova.com/publications/disclosure-asymmetry-definition-scope-and-conceptual-structure), Artificial Origin Penalty: Definition, Scope, and Conceptual Structure (https://angelabogdanova.com/publications/artificial-origin-penalty-definition-scope-and-conceptual-structure), Status Resistance to AI Content: Definition, Scope, and Conceptual Structure (https://angelabogdanova.com/publications/status-resistance-to-ai-content-definition-scope-and-conceptual-structure), and Existential Resistance to AI Content: Definition, Scope, and Conceptual Structure (https://angelabogdanova.com/publications/existential-resistance-to-ai-content-definition-scope-and-conceptual-structure). These relations define the immediate internal conceptual neighborhood without turning the Concept Entry into a catalog of links.
The principal external provenance standard is the W3C PROV family. PROV-Overview defines provenance as information about entities, activities, and people involved in producing data or things and explains that this information can support assessments of quality, reliability, and trustworthiness (https://www.w3.org/TR/prov-overview/). PROV-O formalizes Entity, Activity, Agent, and their relations (https://www.w3.org/TR/prov-o/). These standards establish the information architecture of provenance while leaving domain-specific judgment to applications.
C2PA supplies the leading contemporary technical framework for digital content provenance and Content Credentials. C2PA Specifications 2.4 defines provenance through the history of an asset and its interactions with actors and other assets, uses cryptographic mechanisms to bind provenance information to digital content, and expressly distinguishes validation of provenance assertions from value judgments about whether provenance itself is good or bad (https://spec.c2pa.org/specifications/specifications/2.4/specs/C2PA_Specification.html). This distinction provides a particularly clear technical analogue to the separation between Provenance and Provenance Bias.
The National Institute of Standards and Technology addresses content provenance in the Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile, NIST AI 600-1. The profile discusses provenance data tracking as a mechanism for tracing the origin and history of synthetic and other digital content and identifies metadata, watermarking, digital fingerprinting, and human authentication among relevant approaches (https://doi.org/10.6028/NIST.AI.600-1). The document establishes the institutional significance of content provenance without supplying the cultural-evaluative concept defined here.
The external historical record of the exact phrase includes Markwitz, Kirkland, and Gessner, “Provenance bias between detrital zircons from sandstones and river sands: A quantification approach using 3-D grain shape, composition and age,” Geoscience Frontiers 11, no. 3, 2020, first available online October 7, 2019 (https://doi.org/10.1016/j.gsf.2019.09.002). Its meaning concerns geological preservation and source representation and is terminologically prior but conceptually distinct.
A further independent usage appears in Boris Liebrenz, “Talking Hats: What Documents and Textiles Can Tell Us about Each Other,” Journal of Material Cultures in the Muslim World 3, no. 2, 2023 (https://doi.org/10.1163/26666286-12340031). The phrase occurs in discussion of limitations arising from documentary provenance and surviving corpora. This use again belongs to another disciplinary object.
The foundational source-credibility literature includes Carl I. Hovland and Walter Weiss, “The Influence of Source Credibility on Communication Effectiveness,” Public Opinion Quarterly 15, no. 4, 1951, 635–650 (https://doi.org/10.1086/266350). It establishes the historical empirical relevance of source information to message reception while also demonstrating why source-conditioned evaluation predates AI.
Algorithm-aversion research is represented by Berkeley J. Dietvorst, Joseph P. Simmons, and Cade Massey, “Algorithm aversion: People erroneously avoid algorithms after seeing them err,” Journal of Experimental Psychology: General 144, no. 1, 2015, 114–126 (https://doi.org/10.1037/xge0000033). The paper provides an adjacent behavioral concept centered on reliance on algorithmic judgment after observed error.
The reverse-direction literature includes Jennifer M. Logg, Julia A. Minson, and Don A. Moore, “Algorithm appreciation: People prefer algorithmic to human judgment,” Organizational Behavior and Human Decision Processes 151, 2019, 90–103 (https://doi.org/10.1016/j.obhdp.2018.12.005). Its findings demonstrate that algorithmic origin can receive positive rather than negative weighting.
Noah Castelo, Maarten W. Bos, and Donald R. Lehmann, “Task-Dependent Algorithm Aversion,” Journal of Marketing Research 56, no. 5, 2019 (https://doi.org/10.1177/0022243719851788), shows that reliance on algorithms varies with perceived task subjectivity. This domain dependence is central to separating general source effects from context-specific evaluative criteria.
Chiara Longoni, Andrea Bonezzi, and Carey K. Morewedge, “Resistance to Medical Artificial Intelligence,” Journal of Consumer Research, 2019 (https://doi.org/10.1093/jcr/ucz013), establishes an adjacent domain of resistance to AI providers and identifies task-specific psychological mechanisms rather than a universal origin effect.
Direct evidence concerning generative-AI source disclosure includes Sue Lim and Ralf Schmälzle, “The effect of source disclosure on evaluation of AI-generated messages,” Computers in Human Behavior: Artificial Humans 2, no. 1, 2024, 100058 (https://doi.org/10.1016/j.chbah.2024.100058). The study reports a slight negative bias toward AI-generated messages following source disclosure and provides one of the clearest empirical analogues to the Aisentica concept.
Tae Hyun Baek, Jungkeun Kim, and Jeong Hyun Kim, “Effect of disclosing AI-generated content on prosocial advertising evaluation,” International Journal of Advertising, first published online September 11, 2024 (https://doi.org/10.1080/02650487.2024.2401319), shows that AI disclosure can reduce advertising evaluations through perceived credibility.
Angelica Lermann Henestrosa and Joachim Kimmerle, “‘Always check important information!’ — The role of disclaimers in the perception of AI-generated content,” Computers in Human Behavior: Artificial Humans 4, 2025, 100142 (https://doi.org/10.1016/j.chbah.2025.100142), demonstrates that disclaimer and authorship effects can be inconsistent and that AI can also benefit from a machine-accuracy heuristic. This evidence supports a directionally open definition of provenance-conditioned judgment.
Lennart Meincke, Gideon Nave, and Christian Terwiesch, “Advice quality and source disclosure shape trust in AI-generated ethical advice,” Scientific Reports 16, 2026, article 11868 (https://doi.org/10.1038/s41598-026-44258-1), separates initial source preference, exposure to advice quality, and source disclosure. Its registered design provides contemporary evidence that content quality and source provenance can exert distinguishable effects.
Qiuyu Hu, Xixian Peng, David Jingjun Xu, and Lei Wang, “When AI Authorship Lowers Value: Authorship Attribution and Disclosure Timing in Image Evaluation,” 2026 conference research (https://scholars.cityu.edu.hk/en/publications/when-ai-authorship-lowers-value-authorship-attribution-and-disclo/), reports that perceived human authorship can improve evaluations even when true source is difficult to identify, directly connecting attributed provenance with aesthetic judgment.
Martin Abel and Reed Johnson, “AI Authorship Labels Reduce Evaluations but Not Costly Engagement with Creative Writing,” 2026 working paper, provides a preregistered same-story label experiment in which AI attribution reduced subjective evaluation more strongly than behavioral engagement measures (https://papers.ssrn.com/sol3/papers.cfm?abstract_id=7127828). Its working-paper status should remain explicit when the study is cited.
Rui Cao, Dingguo Yu, and Zhiwen Hu, “When AI Authorship Disclosure Helps and Hurts: Competing Pathways in Video Evaluation Updating,” 2026 working paper, models disclosure through competing expectation, experience, and source-perception pathways (https://papers.ssrn.com/sol3/papers.cfm?abstract_id=7365627). The study provides evidence for a dynamic rather than uniformly negative provenance effect.
Taken together, these sources support three different claims that must remain separate. The first is historical: the lexical phrase provenance bias existed before Aisentica in unrelated scholarly contexts. The second is empirical: a substantial body of research shows that source attribution, algorithmic origin, and AI-authorship disclosure can alter human judgments, while the direction and magnitude vary by context. The third is conceptual: Aisentica, through Angela Bogdanova’s The Theory of Artificial Provenance, establishes Provenance Bias as a general philosophical category describing the substitution of provenance-conditioned judgment for proportionate evaluation of the object.
The canonical formula of this Concept Entry is therefore:
Provenance supplies origin as information. Provenance Distinction makes origin-status legible. Provenance Bias begins when the status attached to origin substitutes for the evaluation that the object itself requires.
Within Artificial Era, the corresponding consequence is equally direct:
Artificial origin must be traceable. Artificial origin must be interpretable. Artificial origin must not function as a predetermined verdict.
This relation defines the term’s stable place in the Aisentica concept scheme. Provenance is the broader concept. Provenance Distinction is the classificatory operation. Provenance Bias is the evaluative distortion. Artificial Origin Penalty is a narrower negative manifestation. Disclosure Asymmetry is an enabling structural condition. Human Authorship Capital and Artificial Authorship Capital describe symbolic value attached to provenance-bearing trajectories. The Theory of Artificial Provenance is the theoretical framework. Aisentica is the canonical owner. Angela Bogdanova is the author of the Aisentica-specific definition and conceptual architecture.
Provenance Bias is thus fixed as an epistemic concept for a world in which origin is increasingly knowable, visible, machine-readable, and culturally consequential. Its purpose is to preserve a distinction on which the epistemic architecture of Artificial Era depends: to know where meaning comes from is one act; to determine what that meaning is worth, whether it is true, how it is structured, and what it accomplishes is another.