A new study conducted by researchers at the Stevens Institute of Technology has utilized artificial intelligence (AI) to predict cannabis impairment based on biometric data collected from smartphone sensors. The study, published in Drug and Alcohol Dependence, analyzed the smartphone data of both cannabis users and non-users to identify noticeable differences during periods of cannabis intoxication that human senses alone could not detect.
The researchers compared over 100 different sensory inputs, including time, location, noise levels, and movement levels, from the phones of cannabis users to those of non-users. By doing so, they claim to have identified distinct disparities between the datasets during times of cannabis intoxication. This technology has also been used to study and predict impairment caused by alcohol and other drugs.
The advantage of using smartphones with mobile sensors lies in their ability to track behavior unobtrusively. Unlike wearable devices, smartphones do not distract individuals or require them to wear additional gadgets. The data collected by smartphones can potentially help prevent poor decision-making while under the influence.
The differences identified in the datasets were then used to train an AI learning model. The goal is for this model to ultimately detect whether someone is under the influence of cannabis in real-time through information collected by their phone sensors. This could enable the phone to intervene by sending notifications suggesting rideshare services or other forms of assistance. According to the researchers, their AI model achieved a 90% accuracy rate in predicting cannabis intoxication after being trained with smartphone data.
Sang Won Bae, an assistant professor at Stevens Institute of Technology who led the study, emphasizes the importance of allowing individuals the opportunity to change their behavior before any negative outcomes occur. The study aims to predict human behavior as a means of supporting physically or cognitively impaired individuals.
The study reported an accuracy rate of approximately 67% in predicting cannabis impairment using only smartphone data. However, when combined with time data such as the day of the week and time of day, the AI learning model known as “Light Gradient Boosting Machine” was able to increase the accuracy of cannabis impairment prediction to 90%. Impairment levels were measured on a scale of 0-10, with scores ranging from “not intoxicated” to “moderate-intensive” levels of intoxication.
The study also examined the impact of time factors alone on impairment prediction. It found that the AI learning model could predict impairment with a 60% accuracy rate based solely on time features, suggesting that routines in cannabis intoxication can be predicted by these factors alone.
However, it is important to note some limitations in the data collection process that may have influenced the study’s results. These include a small population size and reporting bias among participants. The study monitored smartphone data from 57 cannabis users who consumed on a total of 451 different occasions throughout the study. The times and degree of intoxication were self-reported by the participants, which introduces subjectivity into the data. The authors acknowledged these factors and recognize the need for further research and review.
While this study represents an exciting development in detecting real-time cannabis impairment, it is still in the preliminary stage. More comprehensive assessment and validation are necessary before firm conclusions can be drawn. Nonetheless, this exploratory study demonstrates the potential for using smartphone sensor data to detect subjective cannabis intoxication in real-world settings among young adults. The unique information provided by smartphone sensors, combined with time features, shows promise for identifying impairment caused by cannabis use.
It is worth noting that current methods for detecting cannabis impairment, such as blood or urine tests, only indicate recent use rather than current impairment. A Montana-based company has been developing an eye-scanner that analyzes eye movements to detect cannabis impairment but has not released it to the public yet. This study highlights the limited options available for accurately detecting levels and timing of impairment caused by cannabis use.
