x
The need for multi-agent experiments — LessWrong