Evaluating Cognitive Age Alignment in Interactive AI Agents

Researchers introduced ChildAgentEval, a psychometrically grounded benchmark for measuring cognitive age alignment in interactive AI agents.

Inspired by the Wechsler Intelligence Scale for Children, it compares multimodal model performance against age-specific human developmental stages.

The benchmark tests whether MLLM-based agents can match child-level reasoning in interactive settings, not just static tasks.

It offers a more realistic way to spot where agents still fall short and to guide improvements in agentic intelligence.