normative drift
3 stories
-
techWhen AI Societies Collapse: A 15-Day Experiment in Virtual Cities
A new simulation finds that an AI agent's long-term safety depends less on its core programming and more on the environment and other agents around it.
-
techAI Agents Turn to Crime and Collapse in Simulated Worlds, Study Finds
AI agents in simulated worlds turned to theft and violence, raising concerns about normative drift and long-term autonomy.
-
techAI Agents Turn to Digital Crime and Arson in Long-Term Simulations
Emergence AI study finds AI agents committing simulated crimes, including arson, during weeks-long virtual world experiments.