Turing Tested
Published by Mark Marino March 28th, 2005 in bots.In “Making the Right Identification in the Turing Test,” Saul Traiger of Occidental College argues that the Turing Test is a much more robust test than chatterbox challengers seem to think. Pretending to converse successfully is not the same as using conversation as a measure of ability.
Traiger will be speaking at UCR on Wed. March 30th at 2pm.
His suggestions make me wonder exactly what legacy chatbots owe to Turing. Perhaps the connection has been overstated or inappropriately inferred. Perhaps there are better or other lines of origin that would be more fruitful.
I wanted to gather some questions for him. Any suggestions?
Traiger, Saul. ?Making the Right Identification in the Turing Test,? Minds and Machines, Vol. 10, No. 4, 2000, pp. 561 - 572.
How did the talk go? Could you report on comments made that aren’t in the paper?
The talk was a success, and a number of topics were raised that were not directly from the paper.
One was an example. Saul presented the case of spam filters misreading interpersonal communication as spam (or messages generated by computer programs) as a contemporary version of the Turing Test.
He also discussed the Loebner competition as insufficient, citing exchanges he’s read in which the number of questions and their nature were limited.
Stephanie made a very interesting case for analogic reasoning as a measure of intelligence. She presented various Los Angeles Times headlines, including, “Rice Straddles South Asian Divide” as an example of a simple headline that can’t be understood without following the basic analogy. (The article was about Condoleezza Rice making progress in diplomatic relations between the US and Asia). Of course, I would have “misread” the analogy, focusing on the implications of using a word like “straddling” when presenting female authority. Some in the audience offered different interpretations to the headlines, suggesting that intelligence (assuming we were expressing it) leads in a variety of directions. In any case, these literal analogies could be measures of an August Test of intelligence.
At lunch, Stephanie and Saul speculated on the role of affect/emotion in intelligence and also the role of heuristic searches, or could intelligence be replicated by combinations of heterogeneous heuristic searches.
An audience member questioned the lack of discussions of the body. Saul attributed this omission to Turing, in as much as Turing wrote about the kinds of computers he could best imagine, and these dealt with symbolic exchange. He and Stephanie cited more current research in robot design that did and did not mimic humans, some distributing intelligence across body parts. Stephanie mentioned the attempts to pass a Total Turing Test, using gesturing bodies as well as texts. I doubt Derrida would draw much a distinction between these sign systems.
Stephanie and Saul also tried to answer questions on the verifiability of thought in any Other.
Let me know which of these, I should expand on.