A lot of hoopla is being made about the supposed passing of the Turing test. Supposed because it passed a particular test set up, which may or may not have been an adequate simulation of Turing’s abstract test. Then there is the question of whether passing is actually an indicator of any real intelligence or thinking. Regardless, this is a milestone of sorts, though we’ll have to see if it is repeatable now.
Personally, I think a properly rigorous version of the Turing Test would be a good test for intelligence, but I doubt this was rigorous enough. From previous transcripts of when judges were fooled I could see how in some cases they might be if the actual human wasn’t cooperative. I wasn’t able to find transcripts between the judge and the human confederate for recent tests to see how they compare. However, to be an adequate test, the humans would have to be cooperative. If they are trying to fool the judge too, then it is more of measure of the human than the computer. The depth of the depth and nature of some of the interactions also seem to be too limited. The 5 minutes does come from Turing, but I don’t think he was envisioning roughly 1 simple exchange per minute. I certainly don’t think that is sufficient interaction. Perhaps when the transcripts to this test are released we’ll see something different though.
The limited interactions of this set up of the tests allows programmers to write programs to fool the judges instead of programs to exhibit intelligence. In this case it was to pose the program as someone who could not communicate intelligently. While this is a clever tactic to “winning” it doesn’t meet the purpose of the test. With more depth a judge would be able to dig enough to make the distinction between an entity that didn’t communicate intelligently because of cultural/language/etc barriers and an entity that wasn’t capable of intelligence.