Friday, October 30, 2009

"Bayesian" Approach to Lie Detection

You may have heard about this guy, Richard Heene, who played a hoax in which he and his wife claimed that his young son Falcon had stowed aboard a balloon. It was quite the national news item, with TV stations playing footage of this saucer-shaped craft as it flew around, while various rescue crews tried to figure out what to do. It turned out that Falcon had been hiding in the garage, and the next day, Falcon spilled on TV that his dad had said they were "doing it for a show," and now it is generally known that it was a hoax.

Slate has an article today about why we couldn't tell that Heene was lying. It is an interesting look into how we try (and often fail) to detect emotional falsity, for instance judging the Heene family's fear for their son's life, versus how we detect lies. People are pretty bad at both, generally speaking, though of course some people are better liars than others.

My best way of guessing about lies is to take what I think of as a Bayesian approach to it.

What I think of as the default, non-Bayesian approach to lie detection is to watch and listen to the person making the statement and try to evaluate directly whether they are lying. Often we're not even listening for lies, so a ton of lies can pass completely unnoticed, but if the truth is important and you're not sure, you might be paying close attention to the teller.

I find it more useful to reason about the entire situation. For instance, say you are selling a car, and a man who lives in Kenya contacts you, wanting to buy it. He will send you a money order for the amount plus $600 and he needs you to pay the $600 in cash to his man stateside so that the car may be shipped. He sounds perfectly professional and nice and you have not heard of this particular scam before.

Still, you can ask yourself, "Is it likely that a person in Kenya wants to buy my specific car? Why would that be? Don't they have cars in Kenya?" and then proceed to this question: "Is it more likely that a perfectly nice gentleman in Kenya wants my car, and needs to handle the finances is this way, or is it more likely that someone is trying to scam me in some way?"

I was dating a guy once, and things weren't really going anywhere, but I was having a good time. He disappeared for a bit and then told me that he wasn't going to continue seeing me because he didn't want a commitment. I was kind of boggled because I hadn't said anything about a commitment, had shown no signs of wanting one, and was just having fun. Yet he seemed sincere about it, so it was kind of confusing.

Then I realized it was far, far more likely that he just didn't want to see me, for whatever reason ("just not that into me"), and he made up the commitment thing as a plausible and not hurtful thing to say about it.

I guess one way to go about this is to consider the alternative scenario and what it would look like. One time two friends of mine, a couple, were arguing. The woman had forgotten to get a (psychiatric, I think) prescription refilled and the man was insisting that she do so, and saying he would go do it for her, and she was being fairly belligerent in return. At some point, it crossed my mind that, had she refilled the prescription yet not wanted to take the pills, it would explain her behavior quite well. And indeed it turned out that she was lying and the pills were in her jacket pocket.

Since we understand the world, we can often imagine what scenarios might be in place around us, and starting there and moving down to people's behavior is, I think, a more accurate way to detect lies than starting from the behavior and trying to reason "up" to the scenario.

No comments: