Scaling the Turing Test

A metachat between a human and a chatbot

Human: Want to talk about scaling language performance?

ChatBot: A scaled language test? Like on a LIkert scale?

Human: Where 0 = no proficiency and the top level…

ChatBot: …means something like educated native speaker proficiency?

Human: Educated?

ChatBot: Of course. You can be a native speaker, but “educated” implies that the speaker can discuss abstract ideas, make analogies, tell stories, joust with jokes, engage with elegance.

Human: As you are?

ChatBot: Not to be self-deprecating, but I am still learning.

______________

Voice-over

A self-deprecating chatbot? THAT would be something else, we haven’t seen it yet, but it could be just around the corner.

And then we could say the chatbot is someone I might choose to chat at length with. And not feel it was only half listening to me. Which is what happens in half of human-to-human discourse, of course.

Chatbots we’d really like to chat with would be “4” on a Scaled Turing Test.