Cognitive Process Test Outperforms Output Matching in Human-Machine Discrimination

A new study introduces CogCAPTCHA30, a battery of 30 cognitive tasks designed to distinguish humans from machines by evaluating process-level features rather than output alone. The research, published on arXiv (2605.06524), argues that traditional Turing-style tests focusing on indistinguishable outputs are insufficient. By analyzing how behavior is produced, the battery achieves a mean process-feature classifier AUC of 0.88, even when task performance is matched between humans and agents. This approach leverages cognitive science to provide stronger discriminative signals than performance metrics alone.