Steering the Herd: A Framework for LLM-based Control of Social Learning
Raghu Arghal, Kevin He, Shirin Saeedi Bidokhti, Saswati Sarkar
We introduce, analyze, and simulate (via LLMs) a model of controlled social learning to study how algorithms can influence social beliefs, decisions, and welfare via information design.
Abstract
Algorithms increasingly serve as information mediators -- from social media feeds and targeted advertising to the increasing ubiquity of LLMs. This engenders a joint process where agents combine private, algorithmically-mediated signals with observational learning from peers to arrive at decisions. To study such settings, we introduce a model of controlled sequential social learning in which an information-mediating planner (e.g., an LLM) controls the information precision of agents while they also learn from the decisions of earlier agents. The planner may seek to improve social welfare (an altruistic planner) or to induce a specific action the planner prefers (a biased planner). Our framework presents a new optimization problem for social learning that combines dynamic programming with decentralized action choices and Bayesian belief updates. In this setting, we prove the convexity of the value function and characterize the optimal policies of altruistic and biased planners, which attain desired tradeoffs between the costs they incur and the payoffs they earn from induced agent choices. The characterization reveals that the optimal planner operates in different modes depending on the range of belief values. The modes include investing the maximum allowed resource, not investing any resource, or the investment increasing or decreasing with increase in the belief. Notably, for some ranges of belief the biased planner even intentionally obfuscates the agents' signals. Even under stringent transparency constraints—information parity with individuals, no lying or cherry‑picking, and full observability—we show that information mediation can substantially shift social welfare in either direction. We complement our theory with simulations in which LLMs act as both planner and agents. Notably, the LLM-based planner in our simulations exhibits emergent strategic behavior in steering public opinion that broadly mirrors the trends predicted, though key deviations suggest the influence of non-Bayesian reasoning—consistent with the cognitive patterns of both human users and LLMs trained on human-like data. Together, we establish our framework as a tractable basis for studying the impact and regulation of LLM information mediators that corresponds to real behavior.
Framework studying strategic control of social learning by algorithmic information mediators with theoretical analysis and LLM-based simulations.
- Framework for analyzing strategic control of social learning by information mediators
- Proves convexity of value function and characterizes optimal policies for altruistic and biased planners
- Shows constrained planners can substantially impact social welfare in either direction
- Dynamic programming
- Bayesian belief updates
- Game theory
- LLM simulation
Study lacks human data to verify fidelity of LLM-human simulators
from the paper
Generalize model to larger state spaces, more general signal structures, more general loss functions, and heterogeneous agents
from the paperDevelop mechanisms to prevent welfare decrease caused by biased planners through regulation or incentive alignment
from the paper
Author keywords
- Social learning
- LLMs
- optimal control
- information design
- dynamic programming
Related orals
LLM Fingerprinting via Semantically Conditioned Watermarks
Introduces semantically conditioned watermarks for robust and stealthy LLM fingerprinting robust to deployment scenarios.
Every Language Model Has a Forgery-Resistant Signature
Ellipse signatures function as forgery-resistant model output identifiers based on high-dimensional geometric constraints.
Gaussian certified unlearning in high dimensions: A hypothesis testing approach
Analyzes machine unlearning in high dimensions showing single noisy Newton step with Gaussian noise suffices for privacy-accuracy.
Differentially Private Domain Discovery
WGM-based methods provide efficient domain discovery with near-optimal guarantees for missing mass on Zipfian data.
What's In My Human Feedback? Learning Interpretable Descriptions of Preference Data
WIMHF uses sparse autoencoders to extract human-interpretable features from preference data, enabling better understanding and curation of human feedback.