How a Computer Can Tell If You’re Drunk

By Paul Gaita 05/22/15

This could be the start of a Skynet-like takeover of drunk driving enforcement. Or not.

computer algorithm.jpg
Translation: You're drunk. Shutterstock

Researchers in Greece have developed a computer algorithm that can determine whether a person is intoxicated beyond the legal limit.

The algorithm instructs the computer to examine infrared images of individuals’ faces to determine if temperature levels within regions of the face have increased. The temperature spike is the result of alcohol’s effect on blood vessels, which relax and expand. The change in size brings the blood vessels closer to the imbiber’s face, which becomes hot and flushed. If the individual has a deficiency in aldehyde dehydrogenase—an enzyme that helps in the metabolism of alcohol—the flushed response deepens.

Using that information, the Greek researchers took two thermal images of 41 study volunteers’ faces, one before drinking four glasses of wine and another after the volunteers consumed the alcohol.

A second test determined that all of the participants had a blood alcohol level content of at least 0.05, which is below the legal driving limit, but high enough to increase by 38% the volunteers’ chances of getting in an accident if they got behind the wheel of a car. The researchers then used artificial neural networks to compare the facial temperatures shown in the thermal imagery, and applied that information to an algorithm that was able to determine which participant was drunk and which was sober with 90% accuracy.

The researchers stated that they hope the information provided can aid law enforcement in determining if alcohol contributed to individuals’ illegal behavior. They also envisioned a scenario in which the cameras and algorithm could be installed in vehicles to prevent driving while intoxicated.

Please read our comment policy. - The Fix

Paul Gaita lives in Los Angeles. He has contributed to the Los Angeles Times, Variety, LA Weekly, and The Los Angeles Beat, among many other publications and websites.