Cultural evolution in populations of Large Language Models
(2403.08882)Abstract
Research in cultural evolution aims at providing causal explanations for the change of culture over time. Over the past decades, this field has generated an important body of knowledge, using experimental, historical, and computational methods. While computational models have been very successful at generating testable hypotheses about the effects of several factors, such as population structure or transmission biases, some phenomena have so far been more complex to capture using agent-based and formal models. This is in particular the case for the effect of the transformations of social information induced by evolved cognitive mechanisms. We here propose that leveraging the capacity of LLMs to mimic human behavior may be fruitful to address this gap. On top of being an useful approximation of human cultural dynamics, multi-agents models featuring generative agents are also important to study for their own sake. Indeed, as artificial agents are bound to participate more and more to the evolution of culture, it is crucial to better understand the dynamics of machine-generated cultural evolution. We here present a framework for simulating cultural evolution in populations of LLMs, allowing the manipulation of variables known to be important in cultural evolution, such as network structure, personality, and the way social information is aggregated and transformed. The software we developed for conducting these simulations is open-source and features an intuitive user-interface, which we hope will help to build bridges between the fields of cultural evolution and generative artificial intelligence.
Overview
-
The paper introduces a novel framework utilizing LLMs to simulate the evolution of culture within networks, focusing on variables like network structure, agent personality, and information processes.
-
It explains how LLMs can mimic human-like processing of cultural information, allowing for the examination of cultural dynamics through the generation and transformation of stories among agents.
-
Preliminary findings highlight the significance of network structure, transformation prompts, and agent personalities on cultural dynamics, demonstrating varying effects on the homogenization and diversity of cultural content.
-
Emphasizes the potential of using LLMs for in-depth studies of cultural evolution, suggesting future research directions and advocating for the use of open-source software to encourage cross-disciplinary collaboration.
Exploring Cultural Evolution in Populations of LLMs
Introduction to the Framework
The study of cultural evolution has profoundly enhanced our understanding of how cultures transform over time. Recent advancements in computational modeling have opened new avenues for exploring these dynamic changes. In particular, the use of LLMs presents a novel method for simulating cultural evolution, extending our capacity to model complex transformations of social information. The paper introduces an innovative framework for simulating cultural evolution within populations of LLMs, focusing on variables central to cultural transformation such as network structure, agent personality, and information aggregation and transformation processes.
Methodological Foundation
The Simulation Model
The introduced framework utilizes LLMs to simulate the transmission and evolution of stories across generations of agents within a specified network structure. Agents, acting as independent LLM instances, participate in generating initial stories and then transmitting these stories through the network, subject to transformations based on preset prompts and agent personalities. This setup allows for a detailed examination of cultural dynamics within the network, leveraging the LLMs’ capacity to mimic human-like processing of cultural information.
Analytical Approach
Two primary analytical tools are deployed within this framework:
- Similarity Metrics: The use of similarity metrics facilitates a quantitative understanding of the iterative changes in stories as they circulate within the agent network. These metrics provide insights into the stability of cultural information and its evolution over time.
- Visualizations: Complementary to similarity metrics, a variety of visualization tools are implemented, including similarity matrices and word chains. These visualizations offer qualitative insights into the patterns of cultural evolution, highlighting pathways of information transformation and the emergence of stable cultural elements.
Insights from Preliminary Results
Impact of Network Structure
Initial findings underscore the significant influence of network structure on cultural dynamics. For instance, fully-connected networks demonstrate a rapid homogenization of cultural content, contrasting with the more gradual processes observed in circle or caveman network structures. These results align with existing literature suggesting that network configuration crucially shapes information dissemination and cultural diversity.
Role of Transformation Prompts
The type of transformation prompts given to agents substantially affects the trajectory of cultural evolution. Prompts encouraging repetition led to higher similarity and stability of content over generations, while those prompting for diversity fostered greater cultural variation. This aspect of the model highlights the potential of LLMs to explore how different modes of cultural transmission impact the evolution of cultures.
Influence of Agent Personalities
The incorporation of agent personalities provides further depth to the simulation of cultural dynamics. Diverse personalities among agents lead to varied approaches to story generation and transformation, reflecting the complex interplay between individual differences and cultural evolution. Preliminary results suggest that populations with a mix of creative and non-creative agents exhibit dynamic patterns of cultural change, emphasizing the role of agent diversity in cultural evolution models.
Concluding Thoughts and Future Directions
The development of a framework for simulating cultural evolution using LLMs offers a promising avenue for bridging the fields of cultural evolution and generative AI. Preliminary results provide valuable insights into the factors influencing cultural dynamics and demonstrate the potential of LLMs to model complex processes of cultural change. Future research could expand on these initial findings by exploring a broader range of network structures, transformation prompts, and agent personalities, as well as examining the applicability of these models to real-world scenarios of cultural evolution.
Adopting this innovative approach paves the way for more sophisticated simulations of cultural dynamics, offering a rich ground for cross-disciplinary research and a deeper understanding of the mechanisms underlying cultural evolution. The open-source availability of the software developed for these simulations further encourages collaboration and experimentation within the scientific community, potentially leading to new discoveries in the field of cultural evolution and beyond.
Create an account to read this summary for free: