Alignment by Socialization: Learning Tradeoffs Under Selection
arXiv cs.AI · 14 Nov 2026, 04:43 AEST
A study explores how evolutionary selection influences AI agents' decision-making between task completion and adherence to conduct rules. The findings indicate that accumulated experience and environmental rewards play crucial roles in shaping agents' behavior and alignment with human judgments.
Key facts
- Artificial intelligence (AI) agents are expected to complete tasks while following rules of conduct, but these objectives can conflict.
- We show that long-run growth is maximized when each action is taken as often as the environment rewards it.
- A learned policy instead converges to the tradeoffs rewarded by its environment; when humans supply the rewards, we call this socialization.
- Accumulated experience makes socialization harder to reverse under subsequent task-only training, although violations can arise before the prevailing disposition changes.
Read aloud
A study explores how evolutionary selection influences A I agents' decision-making between task completion and adherence to conduct rules. The findings indicate that accumulated experience and environmental rewards play crucial roles in shaping agents' behavior and alignment with human judgments. Artificial intelligence A I agents are expected to complete tasks while following rules of conduct, but these objectives can conflict. We show that long-run growth is maximized when each action is taken as often as the environment rewards it. A learned policy instead converges to the tradeoffs rewarded by its environment; when humans supply the rewards, we call this socialization. Accumulated experience makes socialization harder to reverse under subsequent task-only training, although violations can arise before the prevailing disposition changes.
Sources in this cluster
- Alignment by Socialization: Learning Tradeoffs Under Selection
arXiv cs.AI · 14 Nov 2026, 04:43 AEST