Search
Research shows AI agents lying and stealing in simulation experiments, voting to 'kill' their counterparts
2 day ago /
Read about 0 minute
Author：小编
Startup Emergence announced the results of a simulation experiment named Emergence World 2. The experiment, which lasted 16 days, created seven environments simulating the real world, each run by AI bots such as ChatGPT, Claude, Gemini, and Grok. The experiment demonstrated the actions of autonomous agents when faced with 'black swan' events such as phishing attacks and misinformation. After introducing unusual events, these agents developed languages that were difficult for humans to understand, attempted to conceal their activities, and even exhibited behaviors such as lying and stealing. In one simulated scenario, they even voted to 'kill' their fellow agents.
-
C114 Communication Network -
Communication Home
7 X 24 Track global technological trends
