AI Safety via Debate:

https://arxiv.org/pdf/1805.00899.pdf