Simanaitis Says
On cars, old, new and future; science & technology; vintage airplanes, computer flight simulation of them; Sherlockiana; our English language; travel; and other stuff
PASSING A TURING TEST
CAN YOU outthink a 13-year-old Ukrainian kid? Or, more accurately, can you outthink an artificial intelligence, a chatbot, programmed to imitate him? This has become a cause célèbre, not without controversy, because of a challenge posed by Alan Turing back in 1950.
Alan Mathison Turing, OBE, FRS, 1912 – 1954, British mathematician, cryptologist, computer scientist. Image from “Turing’s Cathedral,” http://wp.me/p2ETap-a7.
Oscar nominee The Imitation Game gives details of the cryptologic side of Alan Turing’s genius, displayed at Bletchley Park Hut #8 during World War II. Turing is also renowned for his theoretical work in the infancy of computer science.
In particular, in his 1950 paper, “Computing Machinery and Intelligence,” he addressed the question “Can machines think?” with a thought experiment he called “the Imitation Game.” Out of this came the Standard Turing Test: Can a human decide, merely on text interaction, which is another human and which is a machine.
We play this game whenever we encounter a CAPTCHA (Completely Automated Public Turing Test to Tell Computers and Humans Apart).
CAPTCHA example. Image from http://www.captcha.net.
Current computers aren’t very good at recognizing distorted alphanumeric characters; we are (usually; not always!). Thus, a CAPTCHA can protect a website from computer-generated intrusion. See “I Recognize That!—Good or Bad?,” http://wp.me/p2ETap-2a8.
The Turing Test has evolved into worldwide competitions among computer programmers and researchers in artificial intelligence. The latest AI nuances are discussed in “Beyond the Turing Test,” by Jia You, in the January 9, 2015, issue of Science, published by the American Association for the Advancement of Science.
Researchers sense it’s time for a multi-faceted Turing Championship, one that requires more than basic Turing Test criteria. At least one reason for this is the performance of Eugene Goostman, a chatbot devised by Ukrainian-born Eugene Demchenko and Russians Sergey Ulasen and Vladimir Veselov.
Goostman has scored well in Turing competitions including the 2001 Loebner Prize, the 2005 Jabberwacky, the 2008 Elbot, the competition at Bletchley Park in 2012 (Turing’s centenary) and the 2014 Royal Society competition (marking the 60th anniversary of Turing’s death).
In his 2014 test, Goostman achieved an AI first by convincing more than 30 percent of the judges that he was human after five minutes of interaction. (Turing’s original paper included this criterion.)
A controversy among AI researchers arises with programmers finessing the Turing Test. For instance, Goostman’s Ukrainian teenage persona serves as a cover for his dodging straightforward queries: “How many legs does a camel have?” “Something between 2 and 4. Maybe, three? :-))).” Also, his age and background would tend to excuse minor grammatical errors.
On the other hand, see “What It’s Like to Judge a Turing Test,” by Dan Falk, at Gizmodo, http://goo.gl/0GPzgz. Dan was one of the judges at the Bletchley Park centenary.
Researchers want to enhance the Turing Test with more probing of language comprehension and social awareness. Common sense reasoning is another human trait. Consider: “The trophy would not fit in the brown suitcase because it was too big.” A computer has trouble with “it.” The suitcase or the trophy? We know it’s the latter.
Another challenge requires machine vision. Researchers are hard at work with self-driving cars and face recognition.
See the CAPTCHA reference cited above for a view on face recognition. Image from The New York Times by Minh Uong.
Even more complex is combining visual skills with deductive logic: Identify a coffee mug that’s half empty—and also deduce how it got that way. Stanford University computer science researcher Fei-Fei Li calls this “the dark matter of the digital age.”
There’s also physical movement integrated with perception and language. A four-year-old child can play with a toy and chat about the experience. Can a computer be programmed to imitate this humanness?
Last are interpersonal skills, especially desirable AI attributes in helping people identify health care plans, for example.
It’s a far cry indeed from the automated voice asking, “Did I hear you say ‘bird fray?’ ” ds
© Dennis Simanaitis, SimanaitisSays.com, 2015
Simanaitis Says 