A Survey on Human-Centric LLMs

TL;DR

The paper surveys human-centric LLMs, framing evaluation around two levels: individual capabilities (cognition, perception, analysis, executive function, and social skills) and collective dynamics (multi-agent interaction and societal-scale tasks). It catalogs a broad set of benchmarks and domain-specific studies across behavioral science, psychology, linguistics, political science, economics, and sociology, emphasizing how LLMs simulate human-like reasoning, social behavior, and cultural sensitivity while exposing gaps in real-time learning, empathy, and nuanced social understanding. Key contributions include a structured competency framework, a synthesis of methodological approaches (prompting, multi-agent prompting, fine-tuning, human-in-the-loop), and a roadmap of open challenges with concrete future directions (real-time adaptation, emotional intelligence, cultural competency, and domain-integrated systems). The findings underscore the potential of LLMs to augment human-centric research and decision-making while highlighting substantial work remaining to align AI behavior with human values, ethics, and cultural contexts, enabling safer and more effective human-AI collaboration.

Abstract

The rapid evolution of large language models (LLMs) and their capacity to simulate human cognition and behavior has given rise to LLM-based frameworks and tools that are evaluated and applied based on their ability to perform tasks traditionally performed by humans, namely those involving cognition, decision-making, and social interaction. This survey provides a comprehensive examination of such human-centric LLM capabilities, focusing on their performance in both individual tasks (where an LLM acts as a stand-in for a single human) and collective tasks (where multiple LLMs coordinate to mimic group dynamics). We first evaluate LLM competencies across key areas including reasoning, perception, and social cognition, comparing their abilities to human-like skills. Then, we explore real-world applications of LLMs in human-centric domains such as behavioral science, political science, and sociology, assessing their effectiveness in replicating human behaviors and interactions. Finally, we identify challenges and future research directions, such as improving LLM adaptability, emotional intelligence, and cultural sensitivity, while addressing inherent biases and enhancing frameworks for human-AI collaboration. This survey aims to provide a foundational understanding of LLMs from a human-centric perspective, offering insights into their current capabilities and potential for future development.

Figures (7)

• Figure 1: Our framework depicts how LLMs are evaluated on foundational human-like skills, divided into individual (e.g., cognition, perception, analysis, executive functioning) and collective (e.g., sociability) levels, and applied within various fields of study similarly categorized as individual (e.g., Behavioral Science, Psychology, Linguistics) and collective (e.g., Political Science, Economics, Sociology) domains.
• Figure 2: Overview of LLM Capabilities Across Individual and Collective Domains.
• Figure 3: Overview of LLM evaluations.
• Figure 4: Competency comparison between LLMs and humans. Blue bars represent skills where LLMs excel over humans, especially in structured tasks and predictable environments, while orange bars indicate skills where humans excel over LLMs, particularly in adaptive, nuanced, and real-world contexts.
• Figure 5: The conceptual framework for section \ref{['sec:studies_improve_llm']}, showing the human-centric domains that have been focused on using LLMs as a primary tool for investigation, the common LLM models that are employed, the most frequently seen techniques for using those models, which go from simple prompting to multi-agent systems and fine-tuning, or a combination of such methods, and finally, the research goals.