(Diagram after Dubberly Design Office)
That’s not a typo — A Turning Test. For Conversation.
We all know the Turing Test for Intelligence — when a human will judge if a machine is “intelligent.” I want a machine that will judge if a conversation is “intelligent.”
That’s right, “intelligent” — the quotes mean it’s ambiguous and we ought to discuss it. The meaning of “intelligent” is subject to opinion and personal values. For me, intelligent conversations are forward-moving, collaborative, generative — they go to new and interesting places. And for you?
I like this proposal for a Turning Test because:
- It’s a constructive thought experiment. I want to provoke conversation about what it takes to have an intelligent conversation — What are the elements? How does it move and flow? When does it generate new ideas and value? Why does it keep us engaged? And how can we improve our conversations — can we design for better conversations?
- It’s a vision for a new service. I want to build such a machine — to write conversational heuristics, and see others write them, to evaluate the qualities and benefits of a given conversation. Then we might input the deluge of all possible conversations and get help in identifying the good ones. How about on social media feeds, just for a start. And on commenting threads and on Slack and on long email threads and political debates — anywhere that conversation is open for review and evaluation.
Because the more we have conversational agents and chatbots, the more we need a Turning Test. And maybe we would steer our own conversations to be better.
CUIs (conversational user interfaces) and VUIs (voice user interfaces), are all the rage today. Companies and techies are trying to put them anywhere. A lot of people are “talking” to their devices — they are “taking turns”, terminology that comes from scholarly work. For example, Me: “Would you like to talk about having great conversations?” You: “Sure, as long as it goes somewhere interesting!” That’s why I’m calling it a Turning Test.
By the way, do you know the etymology of “conversation”? It means “to turn together.” Nice.
I’ve said what I want: your turn.
The notions of intelligence (thinking-based?) and forward moving (directional?) are challenges. They can be gamed by emotions and biases. As we see in politics, once someone chooses a goalpost, it defines the direction the manner, direction, and often limits of the conversation. How we keep things from being one-sided will help keep both parties in the conversation.
“Turning test”: Let me offer a sidestep. I learned that Carnot erred in coining ‘entropy’ as entropy. He apparently mean ‘u-tropos’–the state where things do not and will not turn. He mistook ‘en–‘ for ‘in–‘(as, for example, inactive…) He did not mean the internalized vital state of tropos, turning.
PP’s lovely reflection above seeks machinic judgement of intelligent conversation. This makes me now wonder whether Carnot’s mistake could open something valuable, with a gentler disposition for the definition of the word entropy, several steps away from the frustrating, lawless wildernesses and incorrigible refusals of order and the inexorable marches toward dust and cessation under eternal laws of thermodynamics that are implied by the industrial-revolution treatment of that word. Could an alternate disposition involve a renewal of what some of us actually mean and seek when we say ‘entropy’, closer to its implied ‘-tropos within’ etymology than its industrial-revolution imposed definition? In an optimistic mood, I imagine this invoking the kinds of arisings that ancient atomists saw as the beginnings of life–bemused, curious, precarious, shifting and swerving. Might that serve as a useful term within tests for intelligent turns of conversation? Turning together.
Tango.
It takes two to turn.
Two part test?
First, is it (the exchange) a conversation?
It’s not a conversation if nothing turned.
Two party limit?
Otherwise a discussion or too many bifurcations to follow? Duration matters? Long enough to reach a turn or agreement?
“Hi.”><"Or a white hole?, ha, ha!" [Reverse]
Then, did intelligence happen? This would require transparent access to the thinking provoked by the exchange. Can we measure agreement in degrees or kind? Did something novel get generated? Does conversation need to have a creative or have an improvisational nature? The telemarketing and service call "keystroke" interrogation follows a script. Is intelligence present in those calls? Could it be?