Computational Basis of LLM's Decision Making in Social Simulation
This work addresses the need to regulate social concept encoding in LLMs for alignment and debiasing, with applications in academic and commercial social simulations, though it is incremental in extending existing probing techniques.
The study tackled the problem of understanding how character and context variables influence LLM decision-making in social simulations by proposing methods to probe and manipulate internal representations in a Dictator Game, resulting in substantial alterations in how variables like gender affect decisions.
Large language models (LLMs) increasingly serve as human-like decision-making agents in social science and applied settings. These LLM-agents are typically assigned human-like characters and placed in real-life contexts. However, how these characters and contexts shape an LLM's behavior remains underexplored. This study proposes and tests methods for probing, quantifying, and modifying an LLM's internal representations in a Dictator Game -- a classic behavioral experiment on fairness and prosocial behavior. We extract "vectors of variable variations" (e.g., "male" to "female") from the LLM's internal state. Manipulating these vectors during the model's inference can substantially alter how those variables relate to the model's decision-making. This approach offers a principled way to study and regulate how social concepts can be encoded and engineered within transformer-based models, with implications for alignment, debiasing, and designing AI agents for social simulations in both academic and commercial applications, strengthening sociological theory and measurement.