“AI models are likely to fail when applied to markets and politics because these domains are time-varying, rule-varying, and adversarial, which contrasts with the static, rule-based systems where AI has succeeded.”