From semantic memory to collective creativity: A generative cognitive foundation for social creativity models

Simulation-based theory development has yielded powerful insights into collective performance by linking social structure to emergent outcomes, yet it has struggled to extend to collective creativity. Creativity is hard to capture purely at the social level, as novel ideas are generated through cognitive mechanisms. To address this gap, we introduce a multi-level socio-cognitive agent-based framework in which agents share a common semantic vocabulary and substrate but differ in semantic network topology. A single generative parameter tunes semantic modularity, yielding emergent individual differences in ideational breadth. When agents exchange ideation traces, two canonical social-creativity phenomena arise without being imposed: lower pre-interaction ideation overlap predicts larger stimulation gains, and shared inspiration sources induce network-level redundancy. The framework enables mechanistic theory-building about cognition and social structure in collective creativity.

💡 Research Summary

The paper addresses a longstanding gap in computational models of collective creativity: the inability to link individual cognitive mechanisms of idea generation with social interaction dynamics. The authors propose a multi‑level agent‑based framework in which all agents share a common semantic vocabulary (the set of concept nodes V) and a common substrate graph G₀, but each agent possesses a personalized semantic network Gᵢ that differs in edge composition. Individual differences are generated not by assigning exogenous “creativity scores” but by varying a single structural parameter – the Watts‑Strogatz rewiring probability pᵢ – which systematically alters the global modularity Q(Gᵢ) of each agent’s semantic network. Lower modularity (higher pᵢ) creates more cross‑community shortcuts, facilitating access to remote concepts during retrieval.

Idea generation is modeled as a length‑T random walk starting from a prompt node s. The set of visited nodes Vᵢ(s) constitutes the agent’s “idea trace,” and its cardinality |Vᵢ(s)| serves as a proxy for ideational breadth. Because every agent uses the identical random‑walk process, differences in breadth arise solely from differences in network topology.

Social interaction is implemented by exchanging these traces among agents. Overlap between two agents’ traces is measured, and a “stimulation gain” is defined as the increase in ideational breadth after receiving a partner’s trace. Simulations reveal two canonical social‑creativity phenomena emerging without being hard‑wired: (1) pairs with lower pre‑interaction overlap experience larger stimulation gains, confirming the well‑known benefit of diverse peers; (2) when many agents draw inspiration from the same source, redundancy builds up at the network level, limiting overall diversity.

Empirical analyses show a strong monotonic negative correlation between rewiring probability and modularity (Spearman ρ = –0.92, p < 10⁻¹³¹). Modularity, in turn, predicts ideational breadth with a robust negative relationship (Pearson r = –0.90, p < 10⁻¹⁷⁹, R² = 0.81). Linear models indicate that each unit increase in modularity reduces expected breadth by roughly 29 concepts, while quadratic fits suggest diminishing returns as modularity approaches its low‑end. These findings validate the hypothesis that semantic‑network modularity is a mechanistic “creativity knob” capable of generating individual differences in creative potential.

The framework’s key contributions are threefold: (i) it grounds individual idea generation in a well‑established cognitive model (semantic‑network random walk), (ii) it embeds agents in a shared conceptual space that enables meaningful exchange of ideation traces, and (iii) it demonstrates that canonical social‑creativity effects arise naturally from the interaction of these two levels. By keeping the cognitive and social components modular, the model can be extended to alternative network topologies (e.g., hierarchical, scale‑free), weighted or directed semantic edges, and more sophisticated exchange mechanisms such as trace recombination or collaborative construction.

Limitations include the simplification of semantic memory to unweighted, undirected graphs and the reliance on a single random‑walk retrieval process. Real human semantic networks exhibit graded association strengths, directionality, and context‑dependent activation, which could affect both individual breadth and the dynamics of trace sharing. Moreover, the current exchange protocol treats traces as immutable packets; future work could model transformative processes where agents integrate received ideas into their own networks, potentially altering modularity over time.

Overall, the paper provides a parsimonious yet powerful computational foundation for studying collective creativity, bridging the cognitive and social sciences and opening avenues for systematic, hypothesis‑driven exploration of how individual knowledge structures and interaction patterns co‑determine emergent creative outcomes.