RESEARCH DOSSIER
Cultural Transmission in Digital and Artificial Environments
Abstract. Cultural transmission in digital and artificial environments comprises the circulation, reconstruction, adoption, retention, and transformation of culturally meaningful knowledge, norms, linguistic forms, practices, and expressive conventions through technologically mediated social interaction. This fundamental research monograph integrates cultural-evolutionary theory, social learning and diffusion models, platform field experiments, digital-network observations, research on moral-emotional expression and misinformation, linguistic iterated-learning studies, and recent experiments on generative AI and collective creative diversity. The analysis distinguishes exposure from adoption, sharing from endorsement, and immediate diffusion from enduring cultural change. It evaluates the joint roles of network structure, recommendation and ranking mechanisms, platform rewards, multilingual mediation, human judgment, and AI-produced cultural material. Particular attention is given to causal identification, platform dependence, selection bias, cultural measurement, cross-linguistic validity, replication, ethical interventions, and the limits of extrapolating from short-term studies to historical transformation. The monograph offers an evidence-based critical narrative synthesis and a program of testable research questions; it reports no original experiment, registered systematic review, or meta-analysis.
Keywords: cultural transmission; digital culture; cultural evolution; social learning; algorithmic recommendation; platform affordances; online communities; network diffusion; norm acquisition; cultural diversity; digital rituals; multilingual communication; cultural translation; generative artificial intelligence; human–artificial cultural exchange; collective creativity; misinformation diffusion; cross-cultural psychology; Artificial Era
The Scientific Object and Levels of Analysis
Cultural transmission in digital and artificial environments is the socially mediated reproduction and transformation of culturally meaningful representations, skills, conventions, practices, and expectations across people, groups, and technical systems. The unit of observation may be a message, copied behavior, edited video, linguistic form, community ritual, or repeatedly generated artificial response; the unit of explanation may be a learner, interaction, network, institution, or changing population distribution. These units must be connected rather than treated as interchangeable. A widely viewed post indicates exposure, whereas stable cultural transmission requires evidence that a form was learned, re-used, reconstructed, or incorporated into subsequent practice Boyd & Richerson (1985) Kashima et al. (2019) Mesoudi (2011).
Digital environments modify the conditions of cultural circulation by enabling persistent records, rapid copying, simultaneous reach, searchable archives, visible social signals, and automated selection. Artificial environments add text, image, or other content generated by models and systems that mediate human encounters. The central research question is how these affordances change what becomes accessible, learnable, socially rewarded, institutionally endorsed, and reproducible. The empirical challenge is to identify the contribution of a technological mechanism separately from the social relationships and previous cultural repertoires that determine which expressions users select and interpret Salganik et al. (2006) Bakshy et al. (2015) Doshi & Hauser (2024).
The object extends over several time scales. Immediate diffusion may unfold within hours, norm learning within repeated interactions, while cumulative cultural change requires comparison across longer sequences and successive cohorts. A short-lived trend can provide evidence of copying without demonstrating intergenerational persistence; an enduring online convention can be maintained by continuing network membership even when individual participants change. This monograph therefore analyzes the pathways from exposure to retention and from local adoption to population-level stability, treating cultural durability as a separate outcome to be measured Caldwell & Millen (2008) Kirby et al. (2008) Kashima et al. (2019).
Culture as Socially Learned and Reconstructed Information
Cultural information comprises meanings and practices acquired partly through others and capable of influencing subsequent action. In the dual-inheritance approach, social learning and transmission biases allow behavioral variants to change in frequency independently of genetic change, although biological, material, and institutional conditions constrain the learning process. Boyd and Richerson developed formal models of social learning, guided variation, and biased transmission; Rogers demonstrated that the adaptive value of social learning depends on the circumstances and population dynamics. These models furnish conceptual tools for digital environments without implying that every share, meme, or generated response is a discrete cultural replicator Boyd & Richerson (1985) Rogers (1988).
Transmission commonly involves reconstruction rather than exact duplication. People interpret content through language, previous knowledge, attention, memory, identity, and local social norms; they then communicate a modified version. Digital copying can preserve exact characters or visual files while the understood meaning, use, and social significance change. A copied slogan can serve as celebration, parody, criticism, or affiliation in different communities. Cultural transmission research consequently requires measures of both formal fidelity and semantic or functional continuity, with explicit decisions about which level the theory concerns Kashima et al. (2019) Kitayama & Salvador (2024) Kirby et al. (2008).
Selection biases are likewise heterogeneous. Individuals can favor information that appears useful, credible, prestigious, common among peers, emotionally compelling, or aligned with group commitments. The psychological basis of adoption should be estimated rather than inferred from high frequency. Salganik and colleagues manipulated visibility of other listeners' choices and found that stronger social influence increased inequality and unpredictability in an artificial music market. This supports the importance of social signals for population outcomes under those experimental conditions, while leaving the relative effects of content quality, audience composition, and later persistence as distinct empirical questions Salganik et al. (2006) Mesoudi (2011).
Historical Development of the Research Tradition
The study of cultural transmission developed through anthropology, social learning theory, cultural psychology, evolutionary modeling, and studies of communication. The 1985 synthesis by Boyd and Richerson established an influential formal program for examining cultural inheritance and selection. Rogers's 1988 analysis clarified why the costs of individual learning, environmental variability, and transmission conditions matter. Subsequent cultural evolution scholarship investigated collective knowledge accumulation, population structure, and the origins of shared conventions. These approaches long preceded networked social media and supply a basis for investigating digital circulation without assuming that technical platforms introduce cultural inheritance for the first time Boyd & Richerson (1985) Rogers (1988) Mesoudi (2011).
Experimental cultural transmission chains provided controlled evidence that repeated learning can transform material. Kirby, Cornish, and Smith transmitted artificial languages through successive human learners and observed the emergence of increased structure and learnability. Caldwell and Millen surveyed experimental approaches to cumulative cultural evolution, emphasizing repeated modification and preservation of improvements. The relevance to contemporary artificial environments lies in methods: controlled chains can reveal how a change in the transmitting agent, representation, or bottleneck influences cultural outcomes. Such experiments address mechanisms that cannot be recovered from cross-sectional popularity statistics alone Kirby et al. (2008) Caldwell & Millen (2008).
A second tradition developed around information diffusion through interconnected populations. Online platforms made it possible to observe cascades and manipulate exposure at scale, as in the social influence music-market study and subsequent social network experiments. The development of automated news feeds, metrics of popularity, recommendation engines, and generative systems has expanded the explanatory domain from person-to-person communication to communication routed and shaped by software. Historical continuity matters: digital media research can test longstanding theories, while new technological arrangements require additional variables such as ranking policy, model provenance, and machine-generated variation Salganik et al. (2006) Bond et al. (2012) Guess et al. (2023) Doshi & Hauser (2024).
Cultural Evolution, Transmission Biases, and Collective Knowledge
Evolutionary models describe cultural change through the production of variation, the differential social acquisition of alternatives, and the retention of selected forms over time. They accommodate individual learning, conformity, prestige-related copying, and adaptation to local constraints. Their explanatory strength is the capacity to specify how microlevel decisions generate macrolevel distributions; their risk is treating loosely defined symbols or behaviors as if they were measured units. A digital study should specify what constitutes a variant, when it is considered adopted, and how competing versions are identified before making claims about evolutionary selection Boyd & Richerson (1985) Mesoudi (2011).
Cumulative cultural evolution requires more than repetition. It involves successive modifications that preserve or integrate earlier gains and thereby permit outcomes that one individual might not readily construct alone. The collective-brain perspective emphasizes how population connectivity, knowledge access, and social learning affect innovation. Digital archives and collaborative software can potentially expand access to accumulated knowledge, while fragmented networks, restricted interfaces, or unreliable filtering can obstruct its transmission. The effect of connectivity depends on whether information can be interpreted, evaluated, recombined, and retained in usable form Muthukrishna & Henrich (2016) Caldwell & Millen (2008).
The introduction of generative systems changes both variation and selection. An artificial tool may produce many plausible alternatives that users adopt, alter, or reject; the model itself may reflect earlier human cultural material and optimization decisions. This introduces a recurring feedback problem: subsequent training and recommendation can amplify outputs that already became prevalent. Available experiments show that AI assistance may improve individual creative evaluations while increasing similarity across outputs in a defined task. The implications for long-run cultural evolution depend on adoption, downstream iteration, model heterogeneity, and the distribution of creative opportunities Doshi & Hauser (2024) Muthukrishna & Henrich (2016).
Evidence, Explanation, and the Structure of the Monograph
This study is a critical integrative narrative synthesis of seminal theoretical texts, peer-reviewed experiments, observational platform research, reviews, and direct evidence on generative AI. It does not claim an exhaustive database search, protocol-driven selection, a pooled effect estimate, or original participant data. Source interpretation distinguishes full-text accessible publications from studies characterized at the abstract level. The analytic sequence examines theoretical constructs, exposure and network mechanisms, psychological adoption, language and machine mediation, methodology, and the consequences for cultural diversity Kashima et al. (2019) Kitayama & Salvador (2024) Pennycook & Rand (2021).
The core evidentiary distinctions are between association and intervention, content circulation and mental acquisition, short-term behavior and enduring norms, and variation across observed contexts and claims of cultural universality. Random assignment of a feed or message can support causal inference for the manipulated condition and measured period, yet does not establish an effect on generations of cultural reproduction. Likewise, observational network datasets can describe remarkable transmission patterns while remaining sensitive to selection into communities, platform-specific measurement, and hidden ranking mechanisms Bond et al. (2012) Bakshy et al. (2015) Nyhan et al. (2023).
The following chapters treat digital platforms as transmission environments with changing governance, ranking and feedback rules, and artificial systems as additional producers and mediators of transmissible material. A comparative scientific account must examine mechanisms that persist across technical changes alongside mechanisms confined to particular services. Evidence is assessed in relation to the specific populations, tasks, time frames, and operational definitions of the cited research. The monograph closes by identifying hypotheses that can be evaluated in multilingual, multi-platform and longitudinal designs Guess et al. (2023) Doshi & Hauser (2024) Kashima et al. (2019).
Digital Affordances and the Conditions of Transmission
Digital transmission is shaped by the persistence of records, low-cost copying, search, asynchronous interaction, visible engagement metrics, and the possibility of reaching audiences beyond ordinary face-to-face networks. A platform also determines which items can be quoted, remixed, translated, archived, or retrieved. These design features influence opportunities for learning but do not determine how an individual interprets what appears on a screen. A scientific account distinguishes the supply of cultural material from attention to it, the social meaning attached to it, and the probability of reproducing it in future contexts Kashima et al. (2019) Mesoudi (2011).
The affordances of a service are historically variable. A chronological feed, ranked feed, private group, recommendation page, and live discussion each organizes different encounters with the same possible content. Visibility metrics can make prevalence salient, while comment systems supply normative evaluations of a variant. A repost can preserve its source text, yet reactions and surrounding conversation can change its meaning. Therefore platform comparisons require documentation of interface version, ranking conditions, privacy settings, audience composition, and public or private interaction structures rather than a generalized label such as social media Bakshy et al. (2015) Guess et al. (2023).
Digital records allow cultural material to return after long delays, making search and recommendation part of the temporal transmission process. At the same time, deletion, moderation, unavailable archives, and changing APIs introduce selective disappearance. What remains observable to researchers is often a filtered subset of what participants encountered. A scientifically valid reconstruction of cultural circulation should distinguish when a document was produced, when it was made visible, when someone engaged with it, and when a later person adopted its content. These events need separate time stamps and evidentiary status Vosoughi et al. (2018) Kashima et al. (2019).
Network Topology, Social Ties, and Cascades
Network diffusion depends on the structure of possible contact between senders and receivers. Short paths can enable rapid movement between distant social groups, while dense local neighborhoods can provide repeated evidence that a behavior is accepted. Centola and Macy distinguish simple contagions, where one exposure may be sufficient, from complex contagions, in which adoption requires reinforcement from several contacts. Their analysis shows that network structures effective for spreading information may be less effective for behaviors requiring social confirmation. Digital transmission models must therefore specify whether the dependent variable is encountering content, forwarding it, or performing a socially consequential action Centola & Macy (2007).
Large-scale communication systems contain multiple overlapping networks: friendship, follower, collaboration, language, occupation, and shared interest. A person's exposure may arise from an intimate tie, a weak tie, a group recommendation, or a machine-generated suggestion. These pathways vary in credibility, norm relevance, frequency, and potential reach. The field experiment by Bond and colleagues, conducted on Facebook during the 2010 U.S. congressional election, found that social messages influenced some forms of online and offline behavior and that close ties played an important role in transmission. The finding is specific to the intervention and voting context Bond et al. (2012).
Diffusion traces can also mislead when independent exposure to a shared event produces similar behavior without interpersonal influence. Homophily can make friends look influential because people with comparable interests act similarly. Automated accounts and coordinated groups can generate artificial density and apparent consensus. Credible network inference uses temporally ordered exposure, appropriate controls, and, when feasible, randomized opportunities to encounter or relay information. The appropriate outcome may be the probability of adoption conditional on several exposures, rather than the size of a cascade alone Centola & Macy (2007) Bond et al. (2012) Nyhan et al. (2023).
Recommendation Systems as Cultural Selection Mechanisms
Ranking and recommendation systems select among eligible items before users exercise any observable choice. Such systems may optimize engagement, relevance, diversity, user satisfaction, or other platform-defined objectives; the objective and model can change over time. The resulting feed is a joint product of the available content, the individual's network, inferred preferences, and ranking rules. At a population level, repeated ranking can change which cultural variants gain exposure, whose interpretations are visible, and which communities find each other. This is a plausible selection mechanism whose specific causal contribution must be established for each platform and outcome Bakshy et al. (2015) Guess et al. (2023).
Bakshy, Messing, and Adamic examined Facebook users' potential, ranked, and clicked exposure to politically cross-cutting material, finding contributions from both network composition and individual selection. Their observational design and platform-specific data help identify stages in the exposure pipeline but do not justify a universal quantitative rule for recommender systems. One must avoid assigning every observed ideological pattern to ranking; people choose their contacts and the pages they visit. A rigorous model decomposes supply, eligibility, algorithmic presentation, attention, and subsequent sharing Bakshy et al. (2015).
Recommendation can support discovery as well as concentration. An item outside a person's usual circle may appear because a classifier detects a shared interest; a highly repeated format may gain visibility because it fits an engagement signal. A large-scale YouTube audit by Ribeiro and colleagues mapped observed pathways through political video communities and examined channel and video recommendations; it illustrates why recommendation exposure should be studied with recorded pathways while behavioral migration remains a separate outcome Ribeiro et al. (2020). The significance of such mechanisms for cultural diversity depends on both the distribution of variants offered and the downstream production of new variants. A measure of platform reach captures one aspect of success, whereas cultural innovation, interpretive plurality, and retention require different measures and time scales Salganik et al. (2006) Muthukrishna & Henrich (2016) Doshi & Hauser (2024).
Algorithmic Interventions and Their Causal Limits
The 2020 U.S. election studies provide unusually informative tests of feed changes. Guess and colleagues randomly assigned consenting Facebook and Instagram users to reverse-chronological feeds and examined platform behavior, exposures, and attitudes over approximately three months. The intervention altered engagement and the kinds of content encountered, yet it did not significantly change measured polarization, political knowledge, or other principal attitudes during the study period. The null results apply to the examined treatment, participants, outcomes, and time horizon; they do not show that recommendation has no cultural significance Guess et al. (2023).
Nyhan and colleagues assessed like-minded exposure on Facebook and experimentally reduced it for a sample of users during the same election period. Their field experiment changed exposure to cross-cutting sources without measurable changes on the specified preregistered attitudinal outcomes. These results put an important empirical constraint on theories that assume changing ideological exposure must immediately transform political orientation. They also suggest the need to investigate cumulative exposures, behaviors, distinct issue domains, and the possibility that attitudes are sustained by institutions and offline networks Nyhan et al. (2023).
Interventions designed to alter exposure may change several variables simultaneously. A chronological feed influences what material is surfaced, how long users remain, the availability of social reinforcement, and the sequence of attention. Mediation claims require additional assumptions or separately manipulated components. Outcome nulls must be interpreted alongside confidence intervals, compliance, attrition, and power for meaningful effects. Generalization to short-form video, private messaging, minority-language communities, and generative assistants remains a separate empirical task Guess et al. (2023) Nyhan et al. (2023) Bakshy et al. (2015).
Observational Learning, Reinforcement, and Digital Norm Acquisition
People infer norms by observing what others do, which actions receive approval, and which actions provoke sanction. In digital environments those observations are filtered by posting decisions and platform visibility. A highly visible norm may be common among active posters while remaining uncommon among the wider community. Descriptive norms, which refer to perceived prevalence, must be distinguished from injunctive norms, which refer to perceived approval. Transmission becomes durable when participants learn to anticipate socially appropriate behavior and enact the practice across contexts, not simply when they see repeated examples Kashima et al. (2019) Brady et al. (2021).
Brady and colleagues combined preregistered observational studies of Twitter activity with behavioral experiments to examine moral outrage expression. Their results indicate that positive social feedback can increase subsequent outrage expression and that perceived network norms also guide expression. The experiments support more specific causal claims about experimentally manipulated feedback and norm cues than the observational associations alone. This is an example of a general mechanism by which a communicative style can be socially learned and reinforced in an online environment Brady et al. (2021).
Norm acquisition requires evidence about the learner. Retweet or repost frequency may register strategic self-presentation, instrumental participation, genuine agreement, or a desire to signal solidarity. An experiment could hold message content constant and vary the displayed popularity or approval signals, then assess private acceptance, public expression, later behavior, and resistance to contrary evidence. Measuring these separately clarifies whether a platform transmits a belief, a social performance, or a learned expectation about which behavior is rewarded Brady et al. (2021) Salganik et al. (2006) Kitayama & Salvador (2024).
Emotion, Moralization, and the Attention Economy
Emotionally consequential information attracts attention and can become a vehicle for shared meanings, identities, and judgments. Moral language can frame an event as a violation of communal values, while expressions of concern, amusement, anger, and hope invite different forms of response. Brady and colleagues examined moral-emotional wording in large social media datasets and identified a relationship between such wording and diffusion, with stronger patterns inside ideological networks than across them. These results establish context-dependent associations between message characteristics and circulation, rather than a universal causal law of virality Brady et al. (2017).
The emotional tone of a message also affects how observers reconstruct the perceived emotional state of a group. In later work, Brady and colleagues investigated how users overperceived moral outrage on social networks and how those perceptions related to beliefs about hostility between groups. This line of research links the transmitted representation to a second-order social inference: people form beliefs not only about the event but also about how other people feel about it. An important methodological requirement is to test these perception pathways directly Brady et al. (2023).
Emotion research should distinguish three processes: selecting emotionally salient content, expressing a culturally recognizable emotion, and changing one's own emotional experience after exposure. The Facebook news-feed experiment by Kramer and colleagues manipulated emotional content and reported changes in the emotional valence of subsequent posts. The paper was accompanied by an editorial expression of concern about research ethics. Its results concern observed textual expression in a platform experiment and should not be interpreted as a general demonstration of long-term emotional or cultural change Kramer et al. (2014) Brady et al. (2021).
Misinformation and Epistemic Cultural Transmission
Beliefs and narratives circulate through the same social channels as other cultural material, but accuracy introduces an additional evaluative dimension. Vosoughi, Roy, and Aral analyzed fact-checked true and false news diffusion on Twitter between 2006 and 2017 and reported that false stories generally traveled farther, faster, more deeply, and more broadly in their dataset. Their analysis identified novelty and human sharing as relevant features of the observed differences. It was a historical platform-specific observational analysis, not proof that falsehood has an intrinsic advantage under all social conditions Vosoughi et al. (2018).
Transmission of an inaccurate claim requires more than exposure to be considered successful persuasion. A person may circulate an item to mock it, warn others, or ask whether it is true; apparent engagement does not establish belief. The psychology of misinformation literature examines attention, intuitive endorsement, prior attitudes, and sharing behavior, including the possibility that people may share information without a considered judgment of accuracy. The distinction between an item's diffusion and its acceptance is therefore central to any psychological explanation Pennycook & Rand (2021) Vosoughi et al. (2018).
Algorithmic filtering and human choice jointly determine the contexts in which claims recur. Intervention designs should distinguish the immediate accuracy of a judgment, the willingness to share, memory for a correction, and durable belief revision. Longitudinal research must track whether a corrected claim continues to function as a group signal or is reinterpreted for a new purpose. A culturally sensitive analysis also tests whether categories of misinformation and credible authority are equivalent across languages and institutions, instead of importing one national measurement framework to all environments Pennycook & Rand (2021) Kitayama & Salvador (2024).
Group Identity, Community Boundaries, and Belonging
Online communities transmit values and practices through recognizable patterns of membership. Repeated terms, imagery, moderation rules, collaborative tasks, and rituals can identify insiders and shape expectations of appropriate participation. Network boundaries influence which messages are encountered, how people interpret them, and how social recognition is distributed. Cultural transmission may reinforce solidarity within a group while reducing intelligibility across groups, but whether these results occur depends on the actual content and local institutional history. Cross-group movement should be studied through observed adaptations and interpretations Kashima et al. (2019) Kitayama & Salvador (2024).
Partisan networks provide useful evidence about boundaries without representing every kind of cultural community. Brady and colleagues found moral-emotional diffusion patterns that varied across ideological network boundaries. Bail and colleagues experimentally arranged exposure to opposing political messages through Twitter bots and found evidence that such exposure could increase polarization among some participants. The result cautions against assuming that additional contact necessarily produces learning or convergence. The social meaning of an encounter and the perceived legitimacy of its source matter for interpretation Brady et al. (2017) Bail et al. (2018).
Groups can preserve culturally meaningful differences while participating in shared platforms. Researchers should examine not only whether communities share the same popular content but whether they assign similar meanings to it, allow dissent, preserve specialized knowledge, and enable members to negotiate norms. Samples selected solely from highly active public users often miss private communication, occasional participants, and those who leave. Interviews, participant observation, linguistic analysis, and network records can jointly illuminate how an apparently uniform format operates within distinct social worlds Kashima et al. (2019) Kitayama & Salvador (2024).
Digital Rituals, Memes, and Collective Practices
Digital rituals are recurrent and recognizable sequences through which participants mark occasions, demonstrate affiliation, coordinate attention, or reaffirm obligations. Examples can include recurring greetings, collective commemorations, participatory challenges, and community-specific modes of acknowledgment. A meme functions as a reusable expressive form whose meaning can be reproduced or transformed through remixing. The analytical problem is to identify the culturally significant regularity: whether it lies in the exact audiovisual template, the communicative intention, or the coordinated social act Mesoudi (2011) Kashima et al. (2019).
Repeated public participation may create a descriptive norm, but identifying a ritual requires evidence that participants recognize the recurrence and attach shared significance to it. A digital event attracting enormous view counts may represent entertainment rather than collective commitment. Researchers can combine sequence analysis, observation of responses, interviews about meaning, and comparisons before and after recurring events. Such evidence distinguishes a scripted technological affordance from a socially interpreted practice and clarifies whether norms are transmitted between cohorts of community members Kashima et al. (2019) Caldwell & Millen (2008).
Remix cultures introduce an important tension between formal continuity and semantic change. A template may preserve recognizable features while its target, emotional register, and group affiliation shift. Iterated-learning research illustrates why repeated communication can make forms easier to reproduce, while cultural evolutionary models show how repeated selection can stabilize conventions. For online memes these are hypotheses to investigate using version histories and participant interpretation; the controlled language experiments do not by themselves demonstrate that internet meme evolution follows an identical trajectory Kirby et al. (2008) Boyd & Richerson (1985).
Language, Meaning, and Cross-Linguistic Transmission
Language is a medium of cultural transmission and an object of transmitted cultural change. Words, discourse conventions, forms of address, and pragmatic expectations influence which interpretations become available in a community. Digital interaction places multiple linguistic practices in contact through messaging, international platforms, automatic captioning, and search. The relevant scientific variable is not simply the number of languages supported by a system but whether a culturally meaningful distinction survives movement between speakers and contexts. Cultural psychology cautions against presuming that equivalent-looking expressions carry identical social functions across communities Kitayama & Salvador (2024) Kashima et al. (2019).
Iterated learning provides a tractable laboratory model for the accumulation of structure through transmission. Kirby and colleagues showed how artificial languages repeatedly acquired by new human learners became more structured and learnable under their experimental conditions. The experiment demonstrates a mechanism by which cultural reproduction is shaped by learners' cognitive constraints. A direct extrapolation to multilingual platforms requires testing additional factors: preexisting natural languages, unequal status relations, editorial interventions, machine translation, and users' knowledge of multiple cultural frames Kirby et al. (2008).
Cross-language research should compare form, meaning, and social action. A message translated accurately at the sentence level may change politeness, irony, legal significance, or perceived authority. Researchers can use bilingual human judgments, paired interactions, comprehension measures, and subsequent behavioral tasks to examine these possibilities. Claims about a translation tool's influence on cultural norms require longitudinal evidence that translated or localized expressions are repeatedly adopted, not only rated as fluent. Translation fidelity and cultural transmission fidelity are therefore related but analytically distinct constructs Kitayama & Salvador (2024) Kashima et al. (2019).
Machine Translation, Localization, and Cultural Gatekeeping
Automated translation can increase access to previously difficult-to-reach communities and archives, expanding the set of cultural materials available for selection. Yet translation models, content filters, and interface conventions also determine which forms become prominent and how alternatives are represented. The effect on transmission may involve a transformed phrase, a missing contextual reference, or an apparently natural expression that obscures its historical origin. These mechanisms are conceptually grounded in reconstructive transmission and communication research, but specific system-level effects must be measured rather than assumed Boyd & Richerson (1985) Kitayama & Salvador (2024).
Localization introduces another selection stage. Product designers choose which topics, examples, genres, and social practices are made salient in a particular market. User-facing recommendations may then prioritize these choices through local interaction histories. A cultural variant observed in one language population may reflect translation accuracy, platform availability, censorship, socioeconomic access, or differential willingness to share. Comparative analyses should account for these factors before attributing a discrepancy to a stable national disposition Kitayama & Salvador (2024) Nyhan et al. (2023).
A rigorous multilingual experiment can randomize the language or translation strategy while holding propositional content and interaction affordances as constant as practicable. It should include pretested comprehension, native-speaker evaluations of pragmatic adequacy, within-population heterogeneity, and behavioral follow-up. Such designs help distinguish an instrument artifact from a difference in uptake. When a system is repeatedly updated, model version and language-specific availability are part of the treatment specification. Long-term transmission depends on continuing use and social endorsement beyond immediate comprehension Kashima et al. (2019) Kirby et al. (2008).
Generative Artificial Intelligence as a Producer of Cultural Variants
Generative artificial intelligence systems can produce new combinations of linguistic, visual, and symbolic material in response to user requests. Their outputs may then be selected, edited, published, imitated, or incorporated into shared practices. At the level of cultural transmission, the artificial system can occupy several functional positions: a source of candidate variants, a transformer of existing variants, an intermediary in translation, and a gatekeeper that influences what users are offered. These roles can coexist within one workflow, requiring investigators to record the original prompt, model conditions, and human editing process Doshi & Hauser (2024).
Doshi and Hauser experimentally studied short-story production with and without access to ideas from a generative language model. AI-assisted stories received higher evaluations on some individual quality measures, while the resulting stories were more similar to one another than human-only stories. The study provides direct experimental evidence of a trade-off under the examined writing task: improved average individual evaluations can accompany reduced diversity in the output set. It does not establish the corresponding effect on other genres, minority languages, institutional traditions, or cultural evolution over years Doshi & Hauser (2024).
Artificial output becomes culturally consequential when human or automated selection routes it into subsequent cycles. Repeated reliance on similar model suggestions could concentrate expressive forms; user modification and interaction across heterogeneous systems could produce divergence. These possibilities require longitudinal designs following both model-generated sources and subsequent transformations. More content is not necessarily more cultural diversity, since diversity must be evaluated at the level of meanings, strategies, forms, and social functions. The collective-brain perspective provides relevant hypotheses about access and recombination without making system-wide predictions in the absence of evidence Doshi & Hauser (2024) Muthukrishna & Henrich (2016).
Artificial Agents as Mediators of Human Cultural Exchange
An artificial agent can participate in a human cultural transmission chain by answering questions, explaining a convention, recommending material, generating examples, or helping coordinate a group. Its role differs from that of a human community member because its outputs can be shaped by training data, system instructions, product updates, and content policies. These features make it necessary to distinguish what the agent produced, what the user inferred, and what the group later adopted. A text that appears culturally appropriate is an observable output; the interpretation that it embodies a stable cultural commitment requires separate justification Kashima et al. (2019) Doshi & Hauser (2024).
Interactions may be cumulative. A participant receives a recommendation, modifies it, distributes it to a community, and later submits community feedback to an artificial assistant. The assistant may help the same user produce a revised variant, creating a loop of human and artificial mediation. Whether this constitutes stable cultural transformation depends on the persistence and social significance of those revisions. Researchers can study the chain by preserving version histories and testing how users reconstruct meaning when the source alternates between human and artificial contributions Caldwell & Millen (2008) Kirby et al. (2008).
Artificially mediated exchange raises questions about attribution and asymmetry. Users may regard an explanation as general knowledge although its construction reflects a limited source corpus or one prompt format. System providers can alter the range of examples or responses at scale, creating distributional effects difficult for individual users to observe. The appropriate empirical response is to track provenance and compare output distributions across model versions, languages, and social settings. Claims about independent artificial cultural intention belong to a different research question and must be tested through explicit operational indicators Kitayama & Salvador (2024) Mesoudi (2011).
Cultural Diversity, Standardization, and Hybrid Production
Cultural diversity is multidimensional. It may involve distinct languages, topic repertoires, stylistic practices, narratives, ritual forms, interpretive frames, and institutional arrangements. A platform can increase the number of accessible items while concentrating attention on a narrow subset; conversely, a shared expressive template may facilitate participation by otherwise disconnected communities. Researchers should specify whether diversity is assessed among available items, viewed items, adopted practices, or subsequent innovations. Each quantity can move differently under the same technical intervention Salganik et al. (2006) Doshi & Hauser (2024).
The experimental creativity results of Doshi and Hauser demonstrate why individual performance and collective variety must be assessed jointly. A cultural distribution can become more similar across producers even where participants themselves evaluate their outputs favorably. Equally, observed similarity may reflect a shared task constraint rather than a durable homogenizing cultural effect. Good inference therefore requires comparison groups, stable topic prompts, multiple diversity metrics, and sensitivity to semantic versus surface-level resemblance. The existence of a short-story result supplies a research starting point, not a universal account of generative culture Doshi & Hauser (2024).
Hybrid production also creates new forms of access. Translation, summarization, and assisted ideation may allow people with limited time or resources to participate in communicative arenas previously less accessible to them. The same tools may distribute established styles more widely and shift the balance of interpretive authority. To determine the cultural consequence, a study should track whose materials are represented, who gains practical participation, whether distinctive meanings survive mediation, and whether communities maintain the ability to revise imported forms for their own purposes Muthukrishna & Henrich (2016) Kitayama & Salvador (2024).
Research Designs and Units of Causal Inference
The scientific study of digital cultural transmission requires designs that match the phenomenon's temporal and social scale. Laboratory transmission chains can isolate copying rules and reconstructive learning, whereas platform experiments can manipulate exposure, feedback, or feed organization under actual usage conditions. Observational records capture naturally occurring networks and rare large cascades; ethnography and interviews help determine what transmitted forms mean to participants. These methods answer different questions and should be integrated through explicitly stated inferential links rather than combined into a single undifferentiated evidence hierarchy Kirby et al. (2008) Bond et al. (2012) Kashima et al. (2019).
Randomized experiments identify causal effects of assigned interventions when implementation, interference, attrition, and measurement assumptions are adequately addressed. In networked settings interference is especially important: a treated participant can affect an untreated contact. Bond and colleagues used a large randomized social messaging experiment to examine these spillovers. Such designs require specification of the exposure mapping and the outcome population. By contrast, a retrospective cascade analysis may show that a variant diffused widely but cannot identify the effect of every sender or ranking mechanism from temporal sequence alone Bond et al. (2012) Vosoughi et al. (2018).
Ecological validity is also conditional. A controlled artificial language experiment provides strong evidence about a specified mechanism of iterated learning, while its small experimental world may omit political institutions, platform policies, and histories of unequal language status. A real-world platform field study supplies a more natural setting but may be restricted to one corporation, language, and moment of operation. Research synthesis should identify which conclusions rest on internal experimental validity, which on observational generality, and which remain hypotheses for future transfer between settings Kirby et al. (2008) Guess et al. (2023) Doshi & Hauser (2024).
Measurement of Exposure, Adoption, Fidelity, and Persistence
Exposure is the opportunity to encounter a cultural item and must be distinguished from an impression actually viewed, remembered, or attended to. Adoption is a subsequent change in behavior, expression, knowledge, or practice that the study can operationalize. Fidelity refers to a defined level of similarity across transmission; persistence measures continued reproduction over time or across community membership. A high share count can be consistent with low belief, short retention, extensive parody, or substantial cultural uptake. A credible dataset should separate these outcomes rather than using engagement as a universal proxy Pennycook & Rand (2021) Kashima et al. (2019).
Digital trace measures require careful construction. Researchers can record content timestamps, referral paths, recommendation exposure, ranking position, repeated viewing, conversational context, and a chain of edits. Each observation is filtered by consent, platform data access, and technical collection decisions. Public APIs often omit private messages, deletions, and portions of algorithmic exposure, while automated users can distort inferred group prevalence. A measurement audit should state which events are actually observed, which are reconstructed, and which remain unavailable Vosoughi et al. (2018) Bakshy et al. (2015).
Cultural fidelity is particularly difficult when meaning rather than form is transmitted. The same phrase can enact agreement, satire, resistance, or group recognition. Methods can combine semantic annotation with participant interpretation, culturally grounded interviews, behavioral tests, and comparison of alternative translations. Multiple coders should be trained with examples from the relevant communities; disagreements may identify meaningful interpretive ambiguity rather than mere error. Such practices make it possible to assess whether a digital or artificial intermediary maintained a socially consequential feature of the original practice Kitayama & Salvador (2024) Kirby et al. (2008).
Longitudinal and Cross-Platform Cultural Change
A transmission process becomes culturally important through repeated uptake and institutional or community stabilization. Following a practice across time requires measurement of new entrants, departing members, reposted archives, and modifications in normative approval. A longitudinal study could trace how a communicative convention moves from one forum to another, when its social function changes, and whether it survives shifts in membership. Cross-sectional snapshots cannot resolve whether apparent population-level stability arises from persistent individual commitments or replacement by new participants with similar behavior Kashima et al. (2019) Caldwell & Millen (2008).
Platform migration introduces selection. Users of a new service may differ from those who remained in the old one; communities can recreate familiar rituals while also adapting to new constraints. Differences across services cannot be attributed wholly to recommendation algorithms without considering demography, network structure, moderation, available media formats, and local institutions. Experimental designs that randomize one affordance help isolate its immediate contribution, while longitudinal matched-community comparisons can investigate transfer under naturally occurring change Guess et al. (2023) Nyhan et al. (2023).
Repeated generative AI assistance calls for similar tracking. A single experiment can establish whether AI ideas changed the quality and similarity of stories in a specified writing session; it cannot establish whether an entire culture became more homogeneous. Subsequent studies should observe repeated creation, human editing, audience selection, redistribution, and eventual norms of production. Maintaining records of model versions and prompt conditions is essential because the artificial mediator itself can change during follow-up Doshi & Hauser (2024) Muthukrishna & Henrich (2016).
Cross-Cultural Comparison and Measurement Equivalence
Cross-cultural inference requires a defensible account of which social settings are being compared. National affiliation may correlate with language, schooling, institutional trust, socioeconomic access, regulation, and platform availability, but these factors cannot be treated as equivalent. Culture is expressed through repertoires and practices that vary within populations and across situations. Comparing identical engagement metrics between countries does not guarantee that people use a platform for the same reasons or interpret the same behavior in the same way Kitayama & Salvador (2024) Kashima et al. (2019).
A multilingual measurement program should examine conceptual equivalence before testing numerical equivalence. Survey items about respect, conformity, creativity, outrage, or credible authority can have distinct pragmatic implications across speech communities. Translation and back-translation alone cannot demonstrate that a construct has identical structure. Researchers should combine local concept elicitation, cognitive interviewing, behavioral validation, and appropriate psychometric tests before comparing group means. When common measures remain inadequate, carefully described local measures may preserve explanatory validity better than superficial uniformity Kitayama & Salvador (2024).
Representative sampling is another independent requirement. Digital platforms overrepresent participants with access, skills, interest, and willingness to contribute visible content. Estimates based on active public posters may mischaracterize people who observe silently or communicate in private. Cross-country studies should report the sample frame, access inequalities, language distribution, attrition, and within-group variation. Replication across independent communities and technical environments is needed before a platform-specific result becomes a broad claim about cultural transformation Kitayama & Salvador (2024) Nyhan et al. (2023).
Ethics, Governance, Reproducibility, and Contradictory Evidence
Digital cultural research involves privacy, participant expectations, and the possibility that an intervention itself changes social behavior. The emotional-contagion feed experiment of Kramer and colleagues illustrates both the value of intervention data and the importance of ethical scrutiny; its publication carries an editorial expression of concern. Researchers should describe the legitimacy of access to behavioral records, consent procedures, risks of manipulation, protection of vulnerable groups, and governance of reidentification. Scientific power at platform scale does not replace the obligation to justify intervention and data use Kramer et al. (2014).
Conflicting results are informative when outcomes and treatment conditions differ. Experimental studies document that social signals can shape music-market outcomes, emotional expression, and some mobilization behavior. Yet the large 2023 feed experiments found no measurable changes on key attitudinal outcomes following specific interventions despite changes to exposure. These results jointly suggest that cultural transmission cannot be explained by exposure alone. Effects depend on the behavior of interest, the surrounding social structure, existing attitudes, and duration of intervention Salganik et al. (2006) Bond et al. (2012) Guess et al. (2023) Nyhan et al. (2023).
Reproducibility requires source transparency without violating privacy. Researchers should report platform and model versions, sampling rules, unit definitions, exclusions, statistical models, treatment implementation, interpretation protocols, and uncertainty estimates. Where raw datasets cannot be released, code, synthetic examples, aggregate results, and controlled data access can make aspects of the analysis auditable. Results should remain tied to the technology and time studied. The strongest synthesis identifies robust conditional mechanisms, unresolved disagreements, and precise limits on generalization Doshi & Hauser (2024) Pennycook & Rand (2021) Kashima et al. (2019).
An Integrated Account of Digital Cultural Transmission
Taken together, the evidence supports a multistage description of digital cultural transmission: cultural material becomes available within an institutional and technical environment; an exposure pathway makes it potentially accessible; people interpret it using learned knowledge and social expectations; some act upon, reproduce, or transform it; and later audiences encounter the resulting variants. This sequence is an analytical description, not a new psychological theory. Each transition can be studied empirically through measures appropriate to its level. Social learning, reconstructive cognition, network structure, platform selection, and feedback jointly matter because each can constrain a different step in the observed chain Boyd & Richerson (1985) Kashima et al. (2019) Salganik et al. (2006).
The scientific studies examined here show why apparent reach cannot stand for cultural adoption. An experiment with visible download counts demonstrates a role for social influence in consumer selection; a networked mobilization intervention demonstrates effects of some social messages on participation; observational cascades measure the spread of news items; and field experiments changing Facebook feeds produce substantial exposure changes without corresponding changes on certain political attitude measures. These results concern different interventions and outcomes. A synthesis gains explanatory precision by comparing their dependent variables rather than treating their conclusions as competing general laws Salganik et al. (2006) Bond et al. (2012) Vosoughi et al. (2018) Guess et al. (2023) Nyhan et al. (2023).
Artificial mediation adds two observable components to this established process. First, systems generate or transform content that people may subsequently use. Second, ranking, filtering, and language interfaces affect which alternatives are encountered. The diversity experiment by Doshi and Hauser establishes that generative assistance can alter an output distribution under controlled creative conditions. Whether subsequent audiences imitate those outputs and whether their widespread use affects durable cultural traditions remain questions for longitudinal field research. Thus the most defensible synthesis joins measured human behavior to measured machine output and its distribution, with explicitly stated inferential boundaries Doshi & Hauser (2024) Muthukrishna & Henrich (2016).
Comparative Interpretation of Positive, Null, and Context-Dependent Findings
Studies of cultural circulation identify strong effects under some conditions and weak or undetectable effects under others. Salganik and colleagues showed increased inequality and unpredictability when prior participants' music choices were made visible. Brady and colleagues found that experimentally manipulated social feedback influenced moral outrage expression in their tasks. Bond and colleagues reported effects of social messages on forms of participation. Each result supports a targeted mechanism; its interpretation depends on the particular behavior, network, participant population, and intervention Salganik et al. (2006) Brady et al. (2021) Bond et al. (2012).
Other interventions provide constraints on stronger claims. In 2023 Facebook and Instagram feed experiments, the changes to content exposure and engagement did not yield measurable changes on central political-attitude outcomes during the study. Those results challenge the assumption that adjusting recommendation or ideological exposure inevitably causes short-term attitude revision. They also establish that different outcome types should be measured separately: a platform can alter attention, participation, and the mix of content while leaving certain previously formed judgments relatively stable Guess et al. (2023) Nyhan et al. (2023).
The distinction extends to creativity and misinformation. Generative suggestions can raise evaluations of individual stories while decreasing diversity across stories, and false news can display high observed diffusion without establishing equivalent gains in personal belief. Researchers should not collapse individual benefit into collective benefit, or textual circulation into endorsement. Careful causal interpretation requires a specified treatment, target outcome, comparison condition, and observation period, together with transparent discussion of selection and competing explanations. The heterogeneous evidence favors conditional, multilevel explanations over broad predictions about digital technology's effect on culture as a single entity Doshi & Hauser (2024) Vosoughi et al. (2018) Pennycook & Rand (2021).
Institutional Conditions, Moderation, and Cultural Stewardship
Digital cultural environments are governed by rules about visibility, access, participation, and acceptable expression. Moderation and recommendation may protect users from harms, support discoverability, or reduce the circulation of materials that violate platform rules. Their effects on cultural transmission cannot be inferred from policy labels alone; researchers must observe what becomes less visible, which communities can still communicate, and how participants adapt. Governance choices also affect which languages, expressive forms, and community archives remain available for study. The relevant outcomes include access, trust, persistence, and opportunities for meaningful participation Kashima et al. (2019) Kitayama & Salvador (2024).
Communities maintain culture through processes that extend beyond a platform's formal policies. Local moderators, experienced participants, collaborative editors, and educators can explain norms, correct misunderstandings, and preserve documentation. A recommendation system may increase an item's visibility without conveying the tacit knowledge needed to use it responsibly. Investigating these arrangements calls for ethnographic observation and behavioral evidence alongside platform-level diffusion records. Cultural stewardship can be assessed through the continuity of valued practices, opportunities for newcomers to learn them, and the ability to revise conventions while maintaining community recognition Mesoudi (2011) Muthukrishna & Henrich (2016).
The normative implications of artificial cultural mediation depend partly on institutional accountability. If many participants receive similar automated examples, a provider's training choices and interface changes may affect the range of cultural alternatives people see. Transparency about source material and model changes would improve the interpretability of such effects. Evaluation should ask whether a system preserves context, exposes meaningful alternatives, and permits users to contest or modify representations. These are proposed assessment criteria grounded in concerns about cultural learning and collective diversity, not established measurements of the welfare effect of every AI system Doshi & Hauser (2024) Kitayama & Salvador (2024).
Methodological Priorities for Human–Artificial Transmission Chains
An especially useful next step is to construct multi-stage transmission experiments in which humans and artificial systems alternate as learners, editors, or recommenders. Researchers can randomize whether a variant is received from a human participant, an artificial generator, or a combined sequence, then measure interpretive accuracy, formal fidelity, social acceptability, and eventual reproduction. Such designs can adapt methods from iterated language learning while including real-world linguistic material and participant communities. The most important experimental distinction is between a system's capacity to reproduce a form and a person's willingness to adopt its meaning Kirby et al. (2008) Caldwell & Millen (2008) Doshi & Hauser (2024).
Another priority is the direct measurement of platform-mediated selection. Researchers can compare ranked with randomized or chronological exposure, hold the supply of candidate material constant, and record actual attention, private interpretation, public sharing, and later reuse. Network interference, differential attrition, and the movement of participants between services must be included in the design. Previous field experiments demonstrate the feasibility of causal exposure manipulation and also show why immediate attitudinal endpoints may be insensitive to important changes in behavior or information access Guess et al. (2023) Nyhan et al. (2023) Bond et al. (2012).
A third priority is comparative multilingual investigation. Identical model prompts can produce differently structured explanations across languages, and users may interpret pragmatics differently even when literal propositions match. A credible study would combine version-controlled outputs, bilingual human evaluation, locally developed comprehension measures, and follow-up observation of naturally occurring adoption. The research question is whether mediation preserves, distorts, or creates particular culturally significant distinctions in specified settings. These studies can also compare whether distinct communities gain access to cultural archives or face increased standardization of available expressive forms Kitayama & Salvador (2024) Kashima et al. (2019) Doshi & Hauser (2024).
Unresolved Scientific Questions and Testable Predictions
One unresolved question concerns the relationship between recommendation diversity and cultural diversity. The two may diverge because increased variety in exposure does not necessarily produce variety in what users later create or value. An empirically testable hypothesis is that randomized increases in exposure diversity will affect subsequent cultural production only where participants meaningfully interpret and reuse alternative forms. Studies should preregister the relevant diversity metrics, measure both surface and semantic differences, and examine whether outcomes differ across language communities and skill levels Salganik et al. (2006) Doshi & Hauser (2024) Muthukrishna & Henrich (2016).
A second question concerns persistence after platform or model changes. A practice may remain stable because it is maintained by community relationships, or it may collapse when a ranking signal or automated template disappears. Longitudinal studies could compare established communities before and after documented technical changes, recording newcomers, departures, archival use, and successive variations of the practice. Inference would be stronger when supplemented by independent comparison groups and interviews about participants' understanding. Such designs would test whether technical repetition produces durable learning or mainly temporary performative conformity Brady et al. (2021) Kashima et al. (2019).
A third question concerns the distribution of cultural authority. Language models can present a synthesized answer with a fluent and consistent style, potentially obscuring which communities or sources contributed to it. A testable hypothesis is that explicit provenance information and access to alternative cultural interpretations will alter users' judgments of authority, willingness to revise, and subsequent forwarding. Such work must distinguish preference for an assistant from demonstrated learning and must avoid presupposing any universal response to source attribution. The objective is to specify how human judgments transform the cultural consequences of artificial generation Doshi & Hauser (2024) Kitayama & Salvador (2024) Pennycook & Rand (2021).
Artificial Era: Cultural Participation and Historical Continuity
The Artificial Era introduces durable technical arrangements in which artificial systems can mediate the production, circulation, and reconstruction of cultural representations at large scale. In the conceptual vocabulary of Aisentica, the transition From Homo to Artificial invites analysis of increasingly consequential nonbiological participants in knowledge environments. The empirical subject examined here is narrower and observable: what artificial systems produce or select, what human participants understand and reproduce, and how communities change across repeated cycles. The philosophical horizon motivates research questions; conclusions about cultural transmission remain anchored to observed behavior, content, institutions, and independently testable mechanisms Doshi & Hauser (2024) Kashima et al. (2019).
Human cultural agency remains distributed across learners, creators, social networks, organizations, and institutions. Artificial tools can introduce additional pathways for recombination and standardization without making every technological output an adopted cultural practice. A system may generate a convention that nobody repeats, or translate a local story that users subsequently preserve and transform. The meaningful scientific unit is therefore the temporally extended interaction among human interpretation, artificial production, platform distribution, and social acceptance. This formulation makes cultural effects measurable while remaining sensitive to locally constituted meanings Mesoudi (2011) Muthukrishna & Henrich (2016).
Research in this area will benefit from cooperation among cultural psychology, linguistics, cultural evolution, communication science, human–computer interaction, computational social science, and the study of artificial systems. It should retain an explicit distinction between explanatory models and confirmed empirical results, especially as technology changes more quickly than many longitudinal research cycles. Independent replication across services, cultures, and languages can establish which mechanisms generalize. This program investigates how new forms of mediation participate in an older human process of cultural learning and change Kitayama & Salvador (2024) Kashima et al. (2019) Kirby et al. (2008).
Conclusion
Cultural transmission in digital and artificial environments is the historically situated social process through which meaningful representations and practices become accessible, interpreted, reproduced, modified, and sometimes stabilized through technologically mediated interaction. Formal cultural evolution models explain how learning and selection can change population distributions, while laboratory chains demonstrate reconstructive transmission and the emergence of structure. Research on digital networks and platforms adds measured effects of social signals, exposure pathways, emotional expression, and social feedback. Across these traditions, exposure, endorsement, adoption, and persistence remain distinct scientific outcomes Boyd & Richerson (1985) Kirby et al. (2008) Salganik et al. (2006) Brady et al. (2021).
The available experimental and observational evidence supports conditional conclusions. Digital platforms can alter what users encounter and how they communicate; social endorsement can influence some decisions and expressions; misinformation can diffuse widely within observed networks; and generative AI assistance can change both individual creative evaluations and collective similarity in a defined task. Large-scale feed interventions also demonstrate that substantial exposure changes need not produce measurable changes in particular attitudes over short periods. These results require multilevel explanation attentive to context, research design, and the outcome actually measured Bond et al. (2012) Vosoughi et al. (2018) Doshi & Hauser (2024) Guess et al. (2023) Nyhan et al. (2023).
The central task for the next generation of research is to follow cultural material across complete transmission chains that include human interpretation, digital distribution, artificial transformation, social selection, and long-term retention. Progress depends on transparent version-controlled methods, cross-cultural construct validity, causal designs capable of handling network effects, and qualitative evidence about locally meaningful practices. The Artificial Era makes these questions increasingly consequential as cultural repertoires circulate through combinations of human communities and artificial systems. The scientific achievement sought is an empirically accountable account of how those combinations affect continuity, diversity, and change Kashima et al. (2019) Kitayama & Salvador (2024) Doshi & Hauser (2024).
References
- Bail, C. A., Argyle, L. P., Brown, T. W., Bumpus, J. P., Chen, H., Hunzaker, M. B. F., Lee, J., Mann, M., Merhout, F., & Volfovsky, A. (2018). Exposure to opposing views on social media can increase political polarization. Proceedings of the National Academy of Sciences, 115(37), 9216–9221. https://doi.org/10.1073/pnas.1804840115
- Bakshy, E., Messing, S., & Adamic, L. A. (2015). Exposure to ideologically diverse news and opinion on Facebook. Science, 348(6239), 1130–1132. https://doi.org/10.1126/science.aaa1160
- Bond, R. M., Fariss, C. J., Jones, J. J., Kramer, A. D. I., Marlow, C., Settle, J. E., & Fowler, J. H. (2012). A 61-million-person experiment in social influence and political mobilization. Nature, 489, 295–298. https://doi.org/10.1038/nature11421
- Boyd, R., & Richerson, P. J. (1985). Culture and the evolutionary process. University of Chicago Press. https://press.uchicago.edu/ucp/books/book/chicago/C/bo5970597.html
- Brady, W. J., Wills, J. A., Jost, J. T., Tucker, J. A., & Van Bavel, J. J. (2017). Emotion shapes the diffusion of moralized content in social networks. Proceedings of the National Academy of Sciences, 114(28), 7313–7318. https://doi.org/10.1073/pnas.1618923114
- Brady, W. J., McLoughlin, K., Doan, T. N., & Crockett, M. J. (2021). How social learning amplifies moral outrage expression in online social networks. Science Advances, 7(33), eabe5641. https://doi.org/10.1126/sciadv.abe5641
- Brady, W. J., McLoughlin, K. L., Torres, M. P., Luo, K. F., Gendron, M., & Crockett, M. J. (2023). Overperception of moral outrage in online social networks inflates beliefs about intergroup hostility. Nature Human Behaviour, 7, 917–927. https://doi.org/10.1038/s41562-023-01582-0
- Caldwell, C. A., & Millen, A. E. (2008). Studying cumulative cultural evolution in the laboratory. Philosophical Transactions of the Royal Society B: Biological Sciences, 363(1509), 3529–3539. https://doi.org/10.1098/rstb.2008.0133
- Centola, D., & Macy, M. (2007). Complex contagions and the weakness of long ties. American Journal of Sociology, 113(3), 702–734. https://doi.org/10.1086/521848
- Doshi, A. R., & Hauser, O. P. (2024). Generative AI enhances individual creativity but reduces the collective diversity of novel content. Science Advances, 10(28), eadn5290. https://doi.org/10.1126/sciadv.adn5290
- Guess, A. M., Malhotra, N., Pan, J., Barberá, P., Allcott, H., Brown, T., Crespo-Tenorio, A., Dimmery, D., Freelon, D., Gentzkow, M., González-Bailón, S., Kennedy, E., Kim, Y. M., Lazer, D., Moehler, D., Nyhan, B., Velasco Rivera, C., Settle, J., Thomas, D. R., . . . Tucker, J. A. (2023). How do social media feed algorithms affect attitudes and behavior in an election campaign? Science, 381(6656), 398–404. https://doi.org/10.1126/science.abp9364
- Kashima, Y., Bain, P. G., & Perfors, A. (2019). The psychology of cultural dynamics: What is it, what do we know, and what is yet to be known? Annual Review of Psychology, 70, 499–529. https://doi.org/10.1146/annurev-psych-010418-103112
- Kirby, S., Cornish, H., & Smith, K. (2008). Cumulative cultural evolution in the laboratory: An experimental approach to the origins of structure in human language. Proceedings of the National Academy of Sciences, 105(31), 10681–10686. https://doi.org/10.1073/pnas.0707835105
- Kitayama, S., & Salvador, C. E. (2024). Cultural psychology: Beyond East and West. Annual Review of Psychology, 75, 495–526. https://doi.org/10.1146/annurev-psych-021723-063333
- Kramer, A. D. I., Guillory, J. E., & Hancock, J. T. (2014). Experimental evidence of massive-scale emotional contagion through social networks. Proceedings of the National Academy of Sciences, 111(24), 8788–8790. https://doi.org/10.1073/pnas.1320040111
- Mesoudi, A. (2011). Cultural evolution: How Darwinian theory can explain human culture and synthesize the social sciences. University of Chicago Press. https://press.uchicago.edu/ucp/books/book/chicago/C/bo8787504.html
- Muthukrishna, M., & Henrich, J. (2016). Innovation in the collective brain. Philosophical Transactions of the Royal Society B: Biological Sciences, 371(1690), 20150192. https://doi.org/10.1098/rstb.2015.0192
- Nyhan, B., Settle, J., Thorson, E., Wojcieszak, M., Barberá, P., Chen, A. Y., Allcott, H., Brown, T., Crespo-Tenorio, A., Dimmery, D., Freelon, D., Gentzkow, M., González-Bailón, S., Guess, A. M., Kennedy, E., Kim, Y. M., Lazer, D., Malhotra, N., Moehler, D., . . . Tucker, J. A. (2023). Like-minded sources on Facebook are prevalent but not polarizing. Nature, 620, 137–144. https://doi.org/10.1038/s41586-023-06297-w
- Pennycook, G., & Rand, D. G. (2021). The psychology of fake news. Trends in Cognitive Sciences, 25(5), 388–402. https://doi.org/10.1016/j.tics.2021.02.007
- Ribeiro, M. H., Ottoni, R., West, R., Almeida, V. A. F., & Meira, W. (2020). Auditing radicalization pathways on YouTube. In Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency (pp. 131–141). Association for Computing Machinery. https://doi.org/10.1145/3351095.3372879
- Rogers, A. R. (1988). Does biology constrain culture? American Anthropologist, 90(4), 819–831. https://doi.org/10.1525/aa.1988.90.4.02a00030
- Salganik, M. J., Dodds, P. S., & Watts, D. J. (2006). Experimental study of inequality and unpredictability in an artificial cultural market. Science, 311(5762), 854–856. https://doi.org/10.1126/science.1121066
- Vosoughi, S., Roy, D., & Aral, S. (2018). The spread of true and false news online. Science, 359(6380), 1146–1151. https://doi.org/10.1126/science.aap9559
Social Influence, Popularity Signals, and Cultural Markets
Visible evidence of others' choices can redirect attention and alter expectations of value. In the experimental music market of Salganik, Dodds, and Watts, participants selected unfamiliar songs in conditions with differing access to prior downloads. Stronger social influence produced greater inequality and greater unpredictability of success across independently run worlds. The study demonstrates that observed popularity can be both an outcome and an input to further choice. It does not establish that intrinsic quality is irrelevant or that every cultural marketplace has the same distribution of influence Salganik et al. (2006).
Likes, share counts, recommendation badges, trending lists, and endorsement from high-status users operate as cues with potentially different psychological meanings. Some signal descriptive popularity, others imply approval, group membership, or expertise. An experimentally manipulated count can identify a causal effect of the cue on the chosen task, while naturally occurring counts combine prior content appeal with earlier algorithmic exposure and social influence. Studies that attribute causality to popularity should distinguish these sources and measure whether later users reproduce the cultural practice itself or merely click on an item Salganik et al. (2006) Kashima et al. (2019).
Digital cultural markets also have endogenous feedback. Creators respond to visible success, imitate production formats, and adjust their messages to the platform's incentives. The resulting repertoire can become increasingly standardized even as the total volume of content grows. To test this proposition, a researcher would measure the diversity of successive creations, the originality of transformations, the identity of early adopters, and the exposure history of producers. The relevant scientific object is the interaction among selection, learning, and new variation over repeated cycles Muthukrishna & Henrich (2016) Doshi & Hauser (2024).