No products in the cart.


No products in the cart.

Study: AI Can Tell if You’re Stoned by Reading Your Smartphone Data

A new study has used artificial intelligence to predict cannabis impairment based on biometric data picked up by smartphone sensors with a surprisingly high level of accuracy.

Researchers at the Stevens Institute of Technology just recently published a study in Drug and Alcohol Dependence which analyzed the smartphone data of cannabis users and non-cannabis users. Cannabis users self-reported the times they consumed and what level of intoxication they experienced based on a simple 1-10 scale.

By comparing and contrasting over 100 different sensory inputs including time, location, noise and movement levels picked up from the phones of the cannabis users to the non-cannabis users, the researchers claim to have identified noticeable differences between the datasets during the times that the cannabis users were intoxicated, differences which regular human senses could not identify on their own. The same technology has also been used to study and predict impairment from alcohol and other drugs as well. 

“Smartphones with mobile sensors are universal and can track our behavior in an unobtrusive way,” said Sang Won Bae, an assistant professor at Stevens Institute of Technology who led the study. “They are not a distraction, you don’t have to wear them, and the data they collect can potentially prevent poor decision-making when under the influence.” 

The differences in datasets were then used to train an artificial intelligence learning model which may one day be able to detect if someone is under the influence of cannabis in real time through the information detected by their phone sensors. This could hypothetically allow the phone to intervene in some way in the form of a notification suggesting rideshare services, etc. The researchers claimed their AI model could predict cannabis intoxication with 90% accuracy after being trained by the smartphone data.

“It’s important to give people the chance to change their behavior before something negative happens,” Bae said. “This study aims to predict human behavior as a way to support people while physically or cognitively impaired.” 

The study claimed to be able to predict cannabis impairment using only the smartphone data with about 67% accuracy, but when paired with time data like day of the week and time of the day, the artificial intelligence learning model “Light Gradient Boosting Machine” was able to predict cannabis impairment with a greatly increased accuracy of 90%. Impairment was measured using a 0-10 scale in which a score of 0 meant “not intoxicated,” a score of 1-3 was considered “low intoxication” and a score of 4-10 meant “moderate-intensive” levels of intoxication. 

“We tested the importance of time features (i.e., day of the week, time of day) relative to smartphone sensor data only on model performance, since time features alone might predict ‘routines’ in cannabis intoxication,” the study said, indicating that the AI learning model could predict impairment with 60% accuracy based on time factors alone. 

There are some noteworthy limitations to the way this data was collected which may have impacted its results including small population size and reporting bias among other factors. The study monitored smartphone data from 57 cannabis users who consumed on a total of 451 different occasions throughout the study. The times of consumption and degree of intoxication were also self-reported by the participants which is, needless to say, a very subjective experience user to user. Both of these factors could sway the results of the study a bit, which the authors of the study acknowledged. 

This is not the first such attempt to detect real time cannabis impairment. Most blood, saliva or urine tests are not able to predict current impairment, only recent use. A Montana-based company was in the process of releasing an eye-scanner for police to use in the field that detects cannabis impairment by analyzing eye movements but that has not been released to the public yet. Regardless, very few if any options exist to accurately detect levels of impairment or time of impairment, which is what the study claimed to have demonstrated is at the very least possible to achieve. 

“This exploratory study demonstrated the feasibility of using smartphone sensor data to detect subjective cannabis intoxication in the natural environment among young adults,” the study said. “Smartphone sensor data contributed unique information, over and above time features, to detect subjective cannabis intoxication.”

Cannabis users needn’t hide their phone before getting stoned just yet as the results of this study are still preliminary and more review is needed before a firm assessment can be made. 

Source link

Leave A Reply

Your email address will not be published. Required fields are marked *

Related Posts