Start practicing tongue twisters because AI might soon assess your sobriety based on your ability to recite them fluently. Following a study that accurately gauged alcohol levels from speech patterns, some experts suggest a potential application. The Guardian examines the findings from a recent article in this month’s Journal of Studies on Alcohol and Drugs. 18 adults of legal drinking age were given alcohol doses until they reached intoxication levels in what could be considered a rather enjoyable research endeavor. Subsequently, they were tasked with reciting tongue twisters daily while their breath alcohol levels were monitored at half-hour intervals.
By observing changes in voice modulation and pace at different stages of intoxication, researchers taught AI to interpret the outcomes. The algorithm demonstrated a 98% accuracy rate in predicting whether individuals were within the legal limits for driving. Dr. Brian Suffoletto, the lead author of the study, stated, “With the prevalence of laptop cameras, we can now leverage digital signals to more precisely identify instances of drinking, thereby enhancing our intervention capabilities at critical junctures.” Suffoletto, an associate professor of emergency medicine at Stanford, envisions numerous practical applications stemming from this technology.
The vocal challenge could potentially be implemented in high-risk environments such as school bus transportation or heavy machinery operation to ensure public safety, as per the Guardian. “One prominent application could be as a safety mechanism in vehicles,” he suggested. He further proposed that establishments like bars and restaurants could utilize this technology to regulate alcohol consumption among patrons. Despite the promising technological advancements, the study’s limited sample size and homogeneity (all participants were white) raise concerns. Professor Petra Meier, specializing in public health, remarked, “There is potential for intriguing innovations that could prove genuinely beneficial in the future.” Nonetheless, a more extensive and diverse dataset would be imperative for further evaluation of this approach.