A simulated Federal Reserve meeting that used artificial intelligence agents modeled on real-life policymakers showed political pressure polarised members of the board in their rate-setting deliberations.