AI models show human-like anhedonia when reward valuation circuits are perturbed

Researchers have developed a new framework to assess reward valuation in vision-language models, drawing parallels to human anhedonia and motivational deficits. By adapting clinical tests used for major depressive disorder, they identified and perturbed reward-anticipatory units within these AI models. The study found that disrupting these units led the models to favor low-effort, low-reward choices, mimicking symptoms of anhedonia without impairing general task capability. This work reveals functional reward valuation circuits in AI that closely mirror those observed in humans. AI

IMPACT This research could lead to AI systems that better understand and respond to human emotional and motivational states, potentially improving human-AI interaction and therapeutic applications.

RANK_REASON Academic paper detailing novel research findings on AI model behavior. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• anhedonia
• major depressive disorder
• MAP-SR
• Nucleus Accumbens
• vision-language model

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →