Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Be.FM: Open Foundation Models for Human Behavior

Yutong Xie, Zhuoheng Li, Xiyuan Wang, Yijun Pan, Qijia Liu, Xingzhi Cui, Kuang-Yu Lo, Ruoyi Gao, Xingjian Zhang, Jin Huang, Walter Yuan, Matthew O. Jackson, Qiaozhu Mei

TL;DR

Be.FM introduces an open, foundation-style model specifically trained to model human behavior by integrating four broad data sources—literature, experimental, survey, and observational data—within a formal framework $y = \mathcal{F}(\mathbb{K}, x, c)$. Built on open LLMs (LLaMA-3.1 backbones) and fine-tuned with LoRA/SFT on Alpaca-formatted data, Be.FM is evaluated through a comprehensive benchmark spanning behavior prediction, personality inference, contextual insight, and behavioral reasoning. Across tasks, Be.FM demonstrates stronger alignment with human distributions and improved inference on individual and population traits, while also generating plausible contextual hypotheses and supporting research workflow reasoning. The work highlights Be.FM as a step toward a new AI-assisted interface for behavioral science, capable of scalable analysis and cross-domain applications, with future directions including expanded data, reinforcement learning, and broader community engagement.

Abstract

Despite their success in numerous fields, the potential of foundation models for modeling and understanding human behavior remains largely unexplored. We introduce Be.FM, one of the first open foundation models designed for human behavior modeling. Built upon open-source large language models and fine-tuned on a diverse range of behavioral data, Be.FM can be used to understand and predict human decision-making. We construct a comprehensive set of benchmark tasks for testing the capabilities of behavioral foundation models. Our results demonstrate that Be.FM can predict behaviors, infer characteristics of individuals and populations, generate insights about contexts, and apply behavioral science knowledge.

Be.FM: Open Foundation Models for Human Behavior

TL;DR

Be.FM introduces an open, foundation-style model specifically trained to model human behavior by integrating four broad data sources—literature, experimental, survey, and observational data—within a formal framework . Built on open LLMs (LLaMA-3.1 backbones) and fine-tuned with LoRA/SFT on Alpaca-formatted data, Be.FM is evaluated through a comprehensive benchmark spanning behavior prediction, personality inference, contextual insight, and behavioral reasoning. Across tasks, Be.FM demonstrates stronger alignment with human distributions and improved inference on individual and population traits, while also generating plausible contextual hypotheses and supporting research workflow reasoning. The work highlights Be.FM as a step toward a new AI-assisted interface for behavioral science, capable of scalable analysis and cross-domain applications, with future directions including expanded data, reinforcement learning, and broader community engagement.

Abstract

Despite their success in numerous fields, the potential of foundation models for modeling and understanding human behavior remains largely unexplored. We introduce Be.FM, one of the first open foundation models designed for human behavior modeling. Built upon open-source large language models and fine-tuned on a diverse range of behavioral data, Be.FM can be used to understand and predict human decision-making. We construct a comprehensive set of benchmark tasks for testing the capabilities of behavioral foundation models. Our results demonstrate that Be.FM can predict behaviors, infer characteristics of individuals and populations, generate insights about contexts, and apply behavioral science knowledge.

Paper Structure

This paper contains 48 sections, 6 equations, 2 figures, 14 tables.

Table of Contents

  1. Introduction
  2. Overview of Be.FM
  3. Behavioral Data
  4. Literature data.
  5. Experimental data
  6. Survey data
  7. Observational data
  8. Model Configuration and Training
  9. Generalizable Capability of Be.FM in Various Behavioral Science Tasks
  10. Be.FM's ability to predict and simulate behaviors across scenarios
  11. Predicting behaviors as a function of characteristics and settings in economic games.
  12. Predicting Big Five personality traits.
  13. Be.FM's inference of characteristics of subjects and populations
  14. Be.FM can generate insights about contextual factors
  15. Be.FM's application of behavioral knowledge in reasoning and problem solving
  16. ...and 33 more sections

Figures (2)

  • Figure 1: Be.FM is a foundation model designed for modeling human behavior. Trained on a diverse portfolio of behavioral datasets, Be.FM has the capabilities to: (1) Predict and simulate behavior across diverse scenarios; (2) Infer motivations and other characteristics of individual humans as well as broader populations from their behaviors; (3) Generate insights about contextual factors that influence human behaviors; and (4) Represent and apply behavioral knowledge in reasoning and problem-solving.
  • Figure 2: Behavior distributions of human players (the first row) and models (remaining rows) in classic behavioral economic games. Each column of histograms represents a specific game scenario. Be.FM (the second and third rows) narrows the gap between AI-generated and human behavior distributions, demonstrating improved alignment and more accurate population-level behavior simulation.