Wednesday, June 11, 2014

Artificially Intelligent?

A mini-buzz surrounded recent claims by some researchers that a super-computer had "passed the Turing Test" that creates a line between human and artificial intelligence (AI).

The test is named after 20th century British computer scientist Alan Turing, whose thinking on the development of computers still has influence today. Turing devised a test in which an artificial intelligence would respond in a conversation with a human being. If the human being could not determine whether or not the computer really was a computer, then it was said to be able to pass the Turing Test.

Leaving aside the fact that there are some flesh-and-blood people whose conversations would have a hard time proving intelligence was involved on either end, Reason writer Zenon Evans gathers up a number of rebuttal arguments from different AI writers and commentors. Among the problems: The test was passed by a chatbot, not a super-computer -- and a chatbot is a program designed to mimic human conversation. Think of the way the average politician responds to the average reporter's question about a scandal in which he or she is involved. The responses are in the form of regular human speech, but they are pre-scripted and designed to carry the form of human speech without fulfilling its function, i.e., explain why campaign contributions got spent at a strip joint. They are instead designed to divert attention from the scandal in the same way that a chatbot is designed to fool people that it is a real live incredibly attractive member of the opposite sex who wants to interact with you and lives just a few miles away.

Also among the problems: Chatbots have passed the Turing test before, and as Evans notes, with higher scores than this one. Plus, by telling people they were talking to a teenage Ukrainian, the test organizers gamed the system by offering a rationale for strange-sounding answers that otherwise might have put them onto the idea that they weren't talking to a person. Evans rounds up several other issues with this particular experiment that would make most reasonable (heh) observers think twice before running with the news.

No comments: