I’m still firmly in the disbeliever camp when it comes to a program doing any reasonable imitation of a human being.
Among his many significant accomplishments, the late British mathematician Alan Turing devised a thought experiment to determine whether a computer program displays “intelligence.” The basic idea was that, if a human judge conversing (via keyboard and screen) with a program and a human being was unable to determine which was which, the program could be said to display intelligence. Turing called this the Imitation Game, and it later became known as the Turing Test.
Turing never actually performed the experiment. But, in the years following his death, the Loebner Prize was established to award a grand prize of $100,000 to the first program to pass the Turing Test (details at www.loebner.net). No one has ever won the grand prize, although every year since its inception in 1991, the organizers have recognized the best program as the winner.
In June of this year, it was reported that a program competing in a Turing Test at the University of Reading in England had actually passed the test. The program in question was named “Eugene Goostman,” and pretended to be a 13-year-old Ukrainian boy. It successfully fooled 30 percent of the judges during a five-minute exchange.
Since it was a thought experiment described in an academic paper and never actually implemented by Turing, the specifics of the test are vague. Fooling 30 percent of the judges is a metric based on a remark by Turing that, by the end of the 20th century, computers would be so good at holding conversations that they’d fool people 30 percent of the time. And the five-minute time limit is completely arbitrary. The age and non-native English skills of “Eugene Goostman” contributed to the win. The transcripts of the winning conversation have yet to be released, so you can’t see for yourself how the program accomplished its “feat.”
One of the issues with the Turing Test, as it appears today, is its emphasis on “fooling the judges,” as opposed to “imitating a human.” For that reason alone, I’m still firmly in the disbeliever camp when it comes to a program doing any reasonable imitation of a human being. In normal conversation, a human being (or a program imitating one) should produce reasonable answers in grammatical form. If we’re conducting the program in written English, the human (or program) should be a native adult English speaker, not a 13-year-old Ukrainian.
Speaking of computers holding conversations, you may have seen the film called “Her.” In it, a man falls in love with the conversational interface to his computer. If you haven’t seen it, I highly recommend it. There are some oddities to it (for example, the man is employed to write personal letters for other people), but overall, it’s clever and points out that intelligent computers won’t be like people, simply because computers are capable of so many more operations per second. The movie is also interesting because the Turing Test isn’t involved: The protagonist falls in love knowing the object of his affection is a very-human-sounding computer program.
During filming, the computer was voiced by Samantha Morton, a relatively unknown British actress (you may know her from “Minority Report” with Tom Cruise). That was changed at the last minute, and the computer’s voice was replaced by the voice of Scarlett Johansson. It’s interesting to consider the effect of the movie had it been voiced by someone whose face isn’t well known. The director, Spike Jonze, made the choice of both Morton and her replacement, and he’s been quoted as saying, “It was only in post production, when we started editing, that we realized what the character/movie needed was different from what Samantha and I had created together.” I’d love to hear the original.
In June, I attended TEDx Sonoma County (www.TEDxSonomaCounty.com). One of the presenters was Dr. Robert Rubin, marine biologist and director of The Pacific Manta Research Group, speaking about manta rays and how they pass the “mirror test” (you can watch his full 15-minute presentation at youtu.be/p3au14Fk8N4). The mirror test isn’t a test of intelligence, but of “self-awareness.” An animal that passes the test demonstrates understanding that the image in the mirror is a reflection of itself. According to Dr. Rubin, mantas (along with only the great apes, bottlenose dolphins, killer whales, elephants and magpies) pass the test. He also asked the provocative question: “How would an intelligence express itself if it didn’t have hands?” That’s an interesting counterpoint to the notion of a computer-based intelligence, which we presume would be just like someone with a really fast Internet connection.
I can’t wait to see what happens with the Loebner Prize for 2014, which will be awarded in November. Tune in here to find out. In the meantime, send your intelligent questions to me at mduffy@northbaybiz.com.
