Cutting Cheese With Pi

A pseudo-Turing Test for conversational intelligence

Despite the idiomatic, euphemistic meaning of “cut the cheese”, this is literally about cheese, not farts.

I design conversational intelligence for a living, and when I test bots out in the wild, I look at them through Dubberly and Pangaro’s framework for conversations. Once in a while, a bot is good enough to sustain my interest and keeps me engaged long enough to invest the time needed for a deep dive conversation design assessment.

My Criteria

The ultimate criteria for me to accept a bot as being conversationally intelligent is that it has to be sincerely and optimistically helpful.

Broken down further, I look for:

1. Its use as a tool
2. Empathy in its personality (and they all have personality whether planned for or not)
3. It can keep context (memory across turns)
4. A good attitude and hard-working nature (the sincerity part)
5. General knowledge or awareness beyond its primary purpose
6. That it uses spoken language
7. Reasoning skills
8. Trustworthiness
9. Optimistic in its problem-solving (even if it doesn’t always get it right, it really wants to help and looks for alternate solutions)