AI Can Predict How Drunk You Are

Start practicing your tongue twisters, because artificial intelligence might be judging your sobriety on how well you recite them soon. At least that's what some researchers suggest after conducting a study analyzing intoxication levels based on speech that had remarkable accuracy. The Guardian walks through the findings in their paper, published in the Journal of Studies on Alcohol and Drugs earlier this month. In what may go down as the most fun experiment to sign up for, 18 adults of legal drinking age were given doses of vodka gimlets until they became intoxicated. The participants were then asked to recite tongue twisters every hour, while their breath alcohol levels were recorded in thirty-minute intervals.

Noting changes in voice pitch and frequency during different levels of drunkenness, the researchers then trained AI to analyze the findings—and the program was able to predict if someone was within legal sobriety limits of driving with 98% accuracy. "With the proliferation of smartphone sensors, we can now harness digital signals to more accurately predict when drinking episodes happen, enhancing our ability to intervene at the most effective moments," Dr. Brian Suffoletto, lead author on the study, told the Register. Suffoletto, an associate professor of emergency medicine out of Stanford, believes that several real-world applications will be easy enough to develop.

"The most obvious one is as a form of ignition lock on cars [that] would not allow someone to start their car unless they could pass the 'voice challenge,' which could be used in certain high-risk workplaces like school bus driver or heavy machine operator to ensure public safety," he said, per the Guardian. He added that restaurants and bars could enable devices to help manage when to cut off patrons from purchasing more drinks. While the tech behind the concept is interesting, the study only included a small sample size and racial makeup (all participants were white). "I believe that there is the potential for exciting developments that could eventually be really useful," said Petra Meier, a professor of public health. "But obviously one would first want to test this approach in larger and more diverse samples." (More AI stories).