So if the conversation fails, and the bot can switch the topic to “don’t you hate how our conversation fails like x” then the conversation continues?

Given that the goal of the bot is to make conversation, teaching it to discuss conversational failures abstractly seems like good strategy.

The ability of bots to make meta-conversation reminds me somewhat of recent advances in integrated Interactive Fiction help / tutorials - particularly the recent Emily Short’s City of Secrets and Dreamhold by Zarf (Andrew Plotkin).

One of the things that most impressed me about COS was the way it traps the most common unhandled text typed by users and gives a general response. In some ways, I suppose that is a good portion of the entire bot development ethic, but until I saw it I hadn’t really thought about how little that had been done in IF….

Frustration, understood in this way, is not a way of distinguishing good from bad art - frustration may simply be more or less present in a given work.

I agree. I translate this to bot writing as the “Make bugs into features”-rule. A bug becomes a feature if the bot can detect it (either directly or by inferring the bug’s present from the context) and can bring it up as a topic of discourse.