Crazy! I was playing around with the exact same idea of having AIs debating each other and me learning from that by listening to their debates.

I used OpenTTS and Ollama for doing that (using locally installed models).

I'm now working on the Humankind Survival Challenge for getting AIs to discuss how humankind could survive the various crisis facing us.

The idea is to discover whether AIs have our best interests at heart or whether they have biases that favour their survival. SciFi stuff but also a kind of Turing test to ensure that AIs remain in their place!