Cultural Transmission Experiment

Updated 22 July 2025

Cultural transmission experiments are empirical and computational studies that examine how behaviors, knowledge, and symbols spread via imitation, innovation, and selection.
They employ diverse methodologies—from minimal computational models to human micro-societies and agent-based simulations—to quantify cognitive biases and transmission fidelity.
Findings show that network structures, conformity biases, and innovation-to-imitation balances crucially shape the evolution, stabilization, and diversification of cultural traits.

A cultural transmission experiment is an empirical or computational paper designed to investigate the mechanisms by which cultural information—behaviors, knowledge, artifacts, strategies, or symbolic representations—is passed from one individual or group to another across time or social structure. These experiments typically seek to isolate and quantify the roles of cognitive biases, social structure, transmission fidelity, and environmental influences on the evolution, stabilization, and diversification of cultural traits. In the following sections, a range of methodologies, models, and empirical results are summarized, drawing on prominent work that spans both human and artificial systems.

1. Foundational Models and Experimental Paradigms

Cultural transmission experiments take multiple forms, from minimal computational frameworks that abstract the evolutionary process to large-scale human participant studies and agent-based artificial intelligence simulations.

Minimal Computational Models

The "Meme and Variations" (MAV) model (Gabora, 2013) is a seminal example, modeling the evolution of ideas as composed of discrete loci, subject to mutation (innovation), imitation, selection via a fitness function, and cognitive learning mechanisms. In MAV, each agent generates or adopts ideas that control simulated behaviors and these are selected based on a structured fitness landscape with biological analogues like over-dominance and epistasis.

Human Experimental Micro-Societies

Human-based experiments have employed micro-societies for systematic paper of transmission. For instance, in the graphical-communication experiments (Fay et al., 2014), participants iteratively created and adopted new iconic representations for given concepts, with the spread of variants tracked and modeled using parameters for memory, conformity (behavioral alignment), and content bias (intrinsic usefulness or learnability).

Agent-Based and Iterated Learning Paradigms

Agent-based models such as EVOC (Gabora et al., 2013) and the Semantic Axelrod Model (Madsen et al., 2014) simulate populations where agents invent and imitate actions or transmit complex, structured knowledge with hierarchical dependencies. Iterated learning paradigms with human participants, like the video game knowledge experiments (Tessler et al., 2021), specifically probe how cultural knowledge accumulates through passing of natural language advice over multiple generations.

2. Mechanisms and Biases in Cultural Transmission

A central goal of cultural transmission experiments is to identify and quantify the forces that shape the adoption, modification, and persistence of cultural variants.

Imitation and Innovation: Agents or participants balance innovation (creating new variants via guided or random mutation) and imitation (copying perceived high-fitness or desirable behaviors). Empirical finding in MAV suggests an optimal innovation-to-imitation ratio for maintaining both adaptability and diversity.
Conformity and Content Biases: Models and experiments distinguish between conformity-biased transmission (tendency to adopt popular variants as in music sampling studies (Youngblood, 2019)) and content biases (preference for variants with superior functional properties as shown in communication micro-societies (Fay et al., 2014)).
Prestige and Success Biases: Analysis of chess move transmission (Lappo et al., 2023) demonstrates that players favor moves played by elites (prestige bias) or those shown to be successful (success bias), beyond simple frequency effects.
Anti-Conformity and Novelty Bias: Negative frequency-dependent transmission (anti-conformity), where rare traits are favored, has been observed in certain domains (chess openings (Lappo et al., 2023)). Conversely, anti-novelty bias, as inferred from baby name datasets (O'Dwyer et al., 2017), suppresses the propagation of unusual new variants.

Network topology is pivotal in shaping transmission dynamics. Layered ego-centric networks (with nested social proximity, inspired by the social brain hypothesis (Palchykov et al., 2014)) predict faster or slower propagation depending on community structure and tie strength. Large-scale experiments with explicit non-linear social network topologies demonstrate that melodies and other artifacts evolve more favorably and preserve greater diversity than in chain or linear structures (Marjieh et al., 18 Feb 2025).

3. Experimental Designs and Methodological Innovations

Cultural transmission experiments are conducted with a range of methodological approaches:

Paradigm	Agents/Subjects	Mode of Transmission
MAV/EVOC	Computational agents	Mutation, imitation, simulation
Micro-society	Human participants	Direct interaction, signaling
Iterated learning	Sequential human generations	Successive advice or behavior
Agent-based LLMs	Artificial neural agents	Prompt-driven text transmission
Reward network games	Humans + AI systems	Social learning, demonstration

Measurement and Analysis

Fitness Functions: Many models rely on explicit fitness landscapes to evaluate the adaptive value of cultural variants, as in MAV and EVOC.
Statistical Transmission Models: Dirichlet-multinomial models and agent-based simulations are employed to disentangle the effects of various biases on the temporal evolution of variant frequencies (e.g., chess move choice (Lappo et al., 2023)).
Similarity Metrics: Conceptual network analyses (Veloz et al., 2013) and cosine similarity matrices in LLM-based frameworks (Perez et al., 13 Mar 2024) are used to quantify convergence, divergence, and the influence of network topology on transmitted information.
Parameter Manipulation: Strategies such as ablation (removing selection or reproduction) (Marjieh et al., 18 Feb 2025), manipulation of social network properties, and the introduction of structured prerequisites (as in the Semantic Axelrod Model (Madsen et al., 2014)) are central for causal inference.

4. Theoretical Insights and Model Extensions

Cultural transmission experiments substantiate and refine key theories in cultural evolution:

Cumulative Culture and Ratcheting: Stepwise improvements and retention (“ratcheting”) of behaviors or knowledge over successive generations can emerge even with minimal cognitive architecture. Social proximity and route memory alone are sufficient to generate cumulative cultural improvement without sophisticated thought (Dalmaijer, 2022).
Cultural Selection vs. Drift: Empirical evidence and modeling indicate that drift (random copying) alone is rarely sufficient to explain real-world data, with selection—both conformity-based and content-based—playing a dominant role (Fay et al., 2014, Youngblood, 2019).
Emergence of Subcultures: Models demonstrate that assortative interaction (preference for self-similar role models) and inheritance where offspring variance is proportional to parental distance can lead to distinct, sympatric subcultural clustering without necessitating explicit spatial separation (Tureček et al., 2022).
Network Effects: The structure of experience—e.g., clustered networks—can preserve diversity, facilitate the rise of local traditions (modular clusters in melodies (Marjieh et al., 18 Feb 2025)), or accelerate homogenization under different conditions.

5. Artificial Intelligence and Machine-Informed Cultural Transmission

Recent work investigates the bidirectional relationship between human and artificial cultural transmission:

AI as Cultural Innovator: In reward network experiments, AI agents discovered superior, non-trivial strategies that humans independently failed to invent. Once socially transmitted, these “machine discoveries” were internalized, rationalized, and preserved by human groups, shifting the cultural baseline (Brinkmann et al., 21 Jun 2025). Essential conditions for machine innovations to propagate were: non-triviality (difficulty for humans), low transmission difficulty (learnability), and clear selective advantage.
LLMs as Agent Populations: Simulated cultural evolution with LLM populations organized in diverse social networks provides a testbed for manipulating personality, network structure, and transformation rules—revealing how generative agents can both mirror and differ from human cultural evolution (Perez et al., 13 Mar 2024).
Robust Real-Time Transmission: In reinforcement learning agents, memory mechanisms, attention to social cues, and adaptive curricula together support high-fidelity real-time cultural transmission from humans, even in novel contexts (Team et al., 2022).

6. Empirical and Analytical Challenges

Cultural transmission experiments face several fundamental challenges:

Data Completeness: Analyses that include only abundant or popular variants risk missing critical selection mechanisms. Full-spectrum data, especially on rare variants, are essential for distinguishing neutrality from bias-driven evolution (O'Dwyer et al., 2017).
Ecological Validity: Balancing experimental control with naturalistic complexity (e.g., using song reproduction tasks (Marjieh et al., 18 Feb 2025) or video games (Tessler et al., 2021)) is crucial for generalizability.
Parameter Identification: Controlled ablations and model comparison (e.g., Bayes-factor analysis (Fay et al., 2014)) are applied to isolate the effects of selection, memory, network topology, and transmission fidelity.
Application to Non-Human Systems: The principles derived extend to non-human animals, as demonstrated in models and experiments addressing cumulative cultural learning in pigeons and structured cultural divergence in animal populations (Dalmaijer, 2022, Tureček et al., 2022).

7. Broader Implications and Future Directions

Cultural transmission experiments have informed theory and practical applications across anthropology, psychology, linguistics, artificial intelligence, and network science. Current and future directions include:

Cross-domain application of models to artifacts (musical instruments, literary texts, technology) for reconstructing cultural phylogenies and historical processes (Youngblood et al., 2020, Camps et al., 2022).
Integration with machine learning and generative AI, raising questions about machine-influenced culture and the co-evolution of human and artificial innovation (Perez et al., 13 Mar 2024, Brinkmann et al., 21 Jun 2025).
Extensive parameter sweeps and dynamic environment modeling to test hypotheses about the emergence of cultural attractors, the resilience of traditions, and hybrid human-machine creativity.

Through methodological diversity and rigorous modeling, cultural transmission experiments have established a robust framework for empirically investigating how cultural traits emerge, spread, stabilize, and evolve, bridging individual cognition, social structure, and macro-evolutionary dynamics.