Beyond Behavior: Why AI Evaluation Needs a Cognitive Revolution

2026

Open publication workspace · Sign in to read the full PDF.

AI-generated summary

1) This paper argues that current AI evaluation methods, rooted in behaviorism, are insufficient for understanding true AI intelligence and proposes a shift towards a cognitive-centric approach.
2) * Critiques the limitations of purely behavioral AI evaluation, drawing parallels to psychology's transition from behaviorism to cognitivism.
* Proposes a "construct-thinking" epistemology that emphasizes process-level evidence and distinguishes between performance and cognitive claims.
* Outlines a research agenda for developing new evaluation methods that can assess internal mechanisms and cognitive properties.
3) AI evaluation, cognitive revolution, epistemology, behaviorism, construct-thinking

Check the original publication for accuracy and context.