Site icon

My own version of the Turing test (developed from a literary life), and no AI is even close

After a decade of studying and practicing the art of writing dialogue (for a novel), I learned a valuable lesson about finding good dialogue writing, and that lesson evolved into a deeper understanding of human communication, and the true work that will be required for immitating human communication.

First, a listener’s response to a speaker comes in two forms: simple and complex. Simple dialogue is a response to a question or a command. These responses are prescribed: the speaker, by asking a question or issuing a command, creates perameters for how the response should flow. Replying to the speaker means following the prescribed instructions inherent to questions and commands. A complex response happens in almost any other scenario, but let me provide some examples to show what I mean, and why these responses are complex:

1)

“The other day I was walking to work when I saw Joe for the first time in so long.”

“Oh my God I love Joe, I haven’t seen him forever.”

2)

“I have an interview tomorrow and I’m so nervous about it.”

“You have no reason to be nervous, you’ve got this.”

What was the purpose of the two responses? The purpose wasn’t to fullfill a command or answer a question. In the first example the responder decided to add information, the personal opinion, to the subject of what was spoken, Joe, and also to inform the speaker that they have a similar experience of not seeing Joe forever. In the second example the listener decided to debate the premise of the comment, with the goal of making the speaker feel better about the interview.

In my time of study I used this insight to narrow in on books that exhibited high levels of skill in dialogue writing. I eventually found that I could scan the dialogue of a novel and quickly determine if the dialogue was made up of primarily question and answer formats, or formats that involved complex responses. The ones with complex responses were inevitably better, and based on the rest of the writing I realized because complex responses require a lot more work from the writer. Personality, nuanced intention, and values come through in comlex responses, where as answering a question tends to show little more than a characters ability to answer a question properly.

My version of the Turing test (The Malcolm Test):

The Turing test involves a person observing a dialogue between a machine and a human, and testing whether they can tell the difference. Although the test isn’t based on the correctness of answers, it is still framed in terms of a question-answer conversation. I believe that all judgements today about AI, and their relatability to humans, is still based on question-answer or command-fullfillment conversations. The Malcolm Test, on the other hand, simply asks, is the AI capable of complex responses? Can the machine understand a statement, and provide a complex response?

Programming such a computer would begin by re-thinking the point when a computer decides to answer a question or fullfill a command. These two intentions need to be replaced by an infinite number of possible intentions for why the AI would be responding to the statement. That sounds like a monumental task, but I think there is a core intention for why humans respond to statements, that relates fundamentally to our species success through community-based problem solving: knowledge sharing. In both the above examples the responder could be said to have had the base-intention of sharing knowledge they know to be true. So I’ll add that concept to the Malcolm Test: can an AI, through complex responses, exhibit a base-intention of sharing knowledge for the purpose of participating in human-community-based problem-solving.

Please check out my mainpage for the novel that was the result of my work, and my essay on the nature of thought and time: pressureoflight.ca

Or buy them both in one book: https://books2read.com/u/bpqoWJ

Exit mobile version