For over fifty years, scientists have pursued the Turing test as the Maltese Falcon of artificial intelligence. English mathematician Alan Turing invented the test as a threshold for determining when artificial intelligence has reached equivalence with human intelligence. A human questioner communicates for five minutes by indirect means, such as a teletype, with two other players, a human and a machine. The machine player tries to deceive the questioner into thinking it is human, and the human player tries to help the human questioner identify the machine.

Turing based the test on the observation that humans necessarily rely only on the external appearance of intelligence when interacting with other humans. The same can be said for trust. We decide when to trust other human or non-human sources of information based on the evidence of our senses in the context of our previous experience.

The Turing test itself is structured on deception. Turing believed that yet-to-be-invented artificial intelligence programs would be able to lie like humans. The University of Alberta's poker-playing artificial intelligence program, Polaris, an excellent bluffer, proved it by beating top human players in 2008. So far, however, no artificial intelligence has passed the general Turing test, which is not limited to a specific field of knowledge.

The Internet has brought deception to the forefront of human concern because it hides important information about the communicator that we humans would otherwise use to evaluate information, and it allows false and misleading information to reach far more people far more quickly than ever before in human history. Jeremy Bailenson, founding director of Stanford University's Virtual Human Interaction Lab, has studied some not-so-obvious ways we can be deceived online. He found that subtle blending of the face of a human Internet user with that of a presidential candidate improves the user's opinion of the candidate. In another study, he discovered that an avatar (online cartoon human) that looked into the human user's eyes with a steady gaze made the human user uncomfortable but was also considered by the human to be persuasive. In a third study, a human user identified with his or her personal avatar's characteristics, feeling and acting friendlier if the avatar was tall or good-looking.

Artificial intelligence can help us cope with deception. Scientists at Canada's University of Waterloo have developed a model for an AI program to help potential buyers evaluate seller ratings in online marketplaces. It would help buyers decide whether to trust a seller and whether a purchase would benefit them. The model's developers make the important point that information can be unreliable because of the subjective quirks of the person making the rating or because of recent changes in seller behavior as well as because of deliberate deception, such as a rater colluding with a seller to boost the seller's ratings. Information can be unreliable for many reasons other than a deliberate attempt to deceive.

Some seller rating programs currently compare an individual rater's judgment to the combined ratings of all raters, and filter out those that are significantly different. The proposed program uses AI to evaluate consistency of a rater's ratings and gives greater weight to more recent information. These are guidelines a human might use, consciously or subconsciously, to evaluate information reliability. For example, a specific rater might always rate a seller high who delivers late and a seller low who delivers on time. (This may seem unusual, but consider that there must be some people who welcome email spam and read their junk mail first!) The AI model program extracts the useful information from unreliable ratings without letting them mislead the prospective buyer.

Can we use artificial intelligence to build a better detector of lying humans? The polygraph, invented about one hundred years ago, evaluates physiological changes which are linked to lying in some people, including heart rate, breathing rate, blood pressure and fingertip perspiration. It's considered to be about sixty to seventy percent accurate. Police in England are reportedly using an AI lie detector. Silent Talker combines a video camera with an AI program to detect small tell-tale movements that people make unconsciously when they lie but which are too small and quick for humans to detect consciously. Nemesysco, a security and fraud prevention company, offers another kind of AI-based lie detector, a speech analysis program based on AI learning algorithms that flags inconsistencies in a speaker's words and detects stress and uncertainty in the voice.

Stephen M. Kosslyn has studied brain patterns in lying and believes there will never be a perfect lie detector. As you've probably learned the hard way, liars have all kinds of devious skills. Some are excellent "on the fly" liars and others rely on memorized stories. Different kinds of lies use different parts of the brain. Even so, future advances in AI and other technologies (such as functional magnetic resonance imaging, which shows oxygen and blood flow within the brain) could produce a better lie detector, perhaps one that uses various high tech methods in combination.

What we really need is a foolproof spam filter, but one that's inserted between our computers and our brains. Until that technology comes along, we'll have to continue to rely on our human intelligence to detect deception. One often-overlooked part of the original Turing test is that the human player tries to help the human questioner identify the machine player. The AI poker ace, Polaris, has to be good at detecting bluffing by its human opponent. Likewise, we can help ourselves, and each other, cope with deception in cyberspace. Useful assistance, like Parental Controls and Junk Mail, is already built into our Macs. In addition, we can make greater efforts to teach people how to evaluate information for logical fallacies, raise internal alarms for suspicious behavior, and trap and expose deceivers. I remember vividly a picture that we first-graders colored with crayons; the goal was to help us understand the danger of that nice man in a car who was offering a big, bulging bag of delicious candy to an innocent lamb-child.

Not just a conduit for misconduct, the Internet is a tool that can help us evaluate and cross-check information if we pay some attention to how we're using it. As the finale for this column, I was going to quote Philadelphian Benjamin Franklin: "Honesty is the best policy." After I did some fact-checking on the Internet, however, I found out that this quote is probably falsely attributed to Franklin and really comes from Cervantes' Don Quixote.

Instead, I'll use another quote, which I'm reasonably certain is from American novelist, F. Scott Fitzgerald's essay "The Crack-Up," originally published in Esquire magazine in 1936: "The test of a first-rate intelligence is the ability to hold two opposed ideas in the mind at the same time, and still retain the ability to function." We can interpret this quote to mean that the human ability to hold the truth in mind while telling a lie is a sign of our intelligence. I prefer another interpretation, that a hallmark of human intelligence is our ability to hold in mind the possibility that we are being deceived while still trusting others sufficiently to function successfully in the world.

Sources and additional information:

Will an artificial intelligence pass the Turing test in the Turing centenary contest in 2012?
http://www.kevinwarwick.com/turing100.htm

"Police set to test new lie detector machine," Yorkshire Post,
http://www.yorkshirepost.co.uk/news/Police-set-to-test-new.430318.jp

Berkman Center for Internet & Society at Harvard University, Speaker Series, "The Psychology and Economics of Trust and Honesty":
Stephen M. Kosslyn, "Brain Bases of Deception: Why We Will Probably Never Have a Perfect Lie Detector,"
And Jeremy Bailenson, "Transformed Social Interaction in Virtual Reality,"
http://lawlab.org/conversations/trust-and-honesty

Kevin Regan, Pascal Poupart, Robin Cohen, "Bayesian Reputation Modeling in E-Marketplaces Sensitive to Subjectivity, Deception and Change," AAAI Twenty-First Conference on Artificial Intelligence, 2006,
http://www.aaai.org/Papers/AAAI/2006/AAAI06-189.pdf