Prosocial AI
A research direction in AI alignment focused on designing AI models that exhibit altruistic, cooperative, and harmless behaviors consistent with social norms during interactions. Core challenges include the context-dependence of norms and the fragility of evaluation benchmarks used to measure such behaviors.
Timeline (last 90 days)
A crowdsourced game built on the open-source model Olmo 3 has become a testing ground for researchers examining prosocial AI evaluation mechanisms.
The core difficulty in evaluating prosociality lies in the fragility of norms: human moral intuitions are highly context-dependent, and the same behavior may shift from prosocial to harmful in different contexts.
All Facts (2)
The core difficulty in evaluating prosociality lies in the fragility of norms: human moral intuitions are highly context-dependent, and the same behavior may shift from prosocial to harmful in different contexts.
50%UnverifiedA crowdsourced game built on the open-source model Olmo 3 has become a testing ground for researchers examining prosocial AI evaluation mechanisms.
50%