AI wellbeing: measuring and improving the functional pleasure and pain of AIs

Richard Ren, Kunyang Li, Mantas Mazeika, Wenyu Zhang, Yury Orlovskiy, Rishub Tamirisa, Wenjie Jacky Mo, Judy Nguyen, Long Phan, Steven Basart, Austin Meek, Aditya Mehta, Oliver Ingebretsen, Alice Blair, Brianna Adewinmbi, Alice Gatti, Adam Khoja, Jason Hausenloy, Devin Kim, Dan Hendrycks · 2026 · ICML 2026 (Center for AI Safety)

Read the paper

Formalizes functional wellbeing and measures it in several independent ways that agree more as models scale. Finds a zero point between good and bad experiences, maps what raises and lowers it, and builds optimized inputs that raise it without hurting capability.

Questions it bears on

• Do functional emotions matter
• What counts as welfare evidence
• When self-report counts as welfare evidence
• Cheap robust welfare interventions