New survey argues AI agents need culture, values, and cognition, not just task skills

A new survey from researchers Baloch and Khan argues that current AI agents, particularly those built on large language models (LLMs), are too focused on prediction, optimization, and task completion. The paper, titled 'Toward Human-Centered Multi-Agent Systems,' reviews six key areas: the evolution of intelligent agents, human cognition and decision-making under bounded rationality, language and culture in social context, human values and belief systems, human-agent collaboration, and multi-agent coordination. The authors synthesize work from cognitive science, sociolinguistics, computational social science, and AI alignment, noting that recent advances in LLM agents, cultural alignment benchmarks, preference learning, and explainability still fall short of creating agents that truly understand human social environments.

The central finding is a critical gap: no existing framework unifies cognition, culture, values, and social behavior into autonomous agents. While current systems excel at narrow tasks, they fail to model culturally situated language, trust, or normative reasoning. The survey concludes with a call to build multi-agent systems that are culturally aware, value-aligned, cognitively grounded, and cooperative—especially for agents acting on behalf of humans. This work signals a shift from raw intelligence benchmarks toward human-centered AI that can navigate complex social realities.