This summer has been one of great travel and great ideas. I haven’t really been meeting my goals of doing regular videos or even regular writing, but June was a very concentrated moment for me in working through some nascent thoughts.

Here’s a talk I gave with Korey Stegared-Pace from Stockholm’s Microsoft Reactor. They focus on offering a lot of different helpful content for developers of all kinds. Korey covers the technical aspect while I consider some of the other reasons why we might want to do this and what it might mean.

The biggest question of course is the unanswerable one – what’s a debate? How do you know? There are endless definitions of a debate and most of them are good. I have very particular pedagogical preferences when it comes to what a debate is. I’m willing to change it based on the situation at hand. But when you say “I’m going to teach an AI how to debate,” the bigger question for me is “What’s a debate,” not “how are you doing to teach it?” If you don’t have some specifics there, you might not be teaching what you think you are.

Another question is the relationship between argument and debate. This is rarely thought about outside of the rather obvious “arguments appear in debates” or the somewhat naïve “debate is a kind of argument.” This also has to be grappled with a bit I think and we have to consider what this relationship would be. Again, no objectively right answer here, but you need a working definition in order to teach this. I’d say the way to think about it is history – you have to have a definition of history to teach it, and that definition has to be serviceable, not universally right – historiography is fascinating because of this continuously appearing feature of history. Same with debate and argument.

Something that didn’t make it into the talk but I’ve thought about since then is the necessity of having a temporary certainty to teach about something that circulates around uncertainty. Science pedagogy is quite comfortable with this – the good way is through the scientific method, a navigation process to teach for the swirling madness. The bad science pedagogy is the one of truth-discovering or “we are right” discourse that even good scientists (as apart from good science teachers) will sometimes say. To teach proof of concept of AI we need some clear understanding of a debate, what it is and how to look at it, a notion of what arguments are in that context, and then we can move forward and say an AI can debate or can’t debate – and what that might mean.