Rather than waiting for ML to catch up to natural language debate, we propose simulating our eventual setting (two AI debaters and one human judge) with all human debates: two human debaters and one human judge. Debaters know more than the judge: Debate can produce interesting results only when the debaters know more than the judge; otherwise asking direct questions is enough. For the specific example of debate, we start with debates with two ML debaters and a human judge, then switch to two human debaters and a human judge.