An Israeli student’s new app uses IoT and machine learning to keep dangerous drivers off the road
With New Year’s Eve just around the corner, millions of people are beginning to ask themselves the same question. No, it’s not if they can really lose 20 pounds in 2017 or any of the other resolutions (read lies) that we tell ourselves every year.
It’s how the hell are we going to get home from the party after. If you have been drinking, then please, please, please, do not get behind the wheel of a car. It is estimated that some 10,000 people are killed every year as a result of drunk driving.
In recent years, there has been a spurt of breathalyzer devices aimed at letting people know if they have had a few too many to drive. Devices paired with apps like the BACtrack line of breathalyzers help users to make smarter decisions. However these kinds of devices generally run at around $100, and depend too heavily on the driver actually remembering to use it.
Looking for an alternative that could save lives, an Israeli Masters of Sciences student out of Ben Gurion University named Ben Nassi has developed an app that can detect intoxication based on the user’s movement as monitored by their smartphone or wearable.
The idea for his research came to him after reading an article that was published two years ago about a drunk driver who killed a father and daughter near the southern city of Arad. He remembers seeing a photo of the car with a bottle of whiskey inside that he couldn’t quite shake, leading him to wonder if there was a way to test for intoxication automatically.
Along with his advisors Professors Yuval Elovici and Lior Rokach of Ben Gurion University, Nessi developed as a part of his thesis a machine learning algorithm that is capable of telling the difference between a person’s behavior when they are either intoxicated or sober.
The team ran their tests with a combination of devices, looking for an array of options. With a Samsung Galaxy S4 in their right rear pocket, an LG G-watch on their left hand, and a Microsoft Band on the right, the IoT-laden test subjects were topped off with Google Glass — the wearable that first set the taboo for wearable cameras until Snap Inc proved that good taste should never get in the way of your ability to upload pictures of your lunch — as they walked with the app tracking their movements, judging how intoxicated they were.
Interestingly, instead of asking the user to perform classic police checkpoint tests like walk a straight line, etc, the app follows the driver’s more basic movement, checking to see how they are walking, and if it varies from their normal gait.
“Unlike others,” Nessi tells Geektime, comparing their study to other kinds of tests, “We didn’t use any specific challenge like the walk and turn test. We told ourselves that we didn’t want to use these kinds of tests, but use more natural activity like the person’s gait, to test if the person is intoxicated.”
Since the app uses data from the user’s phone or wearable, it is constantly collecting data on the person, learning how they behave under normal (read sober) circumstances. Since the app runs silently in the background, the idea here is that the user would not have to actively decide to turn it on before they start drinking or get up to drive, essentially making it an automated solution.
So far, they have been able to attain 93% accuracy with their tests on mid to late stage millennials, having taken it out to three bars in central Tel Aviv for some in-the-field experimentation that they say matches the standards of a police run check.
Nessi points out that their choice of test subjects, a group of 20 somethings, was quite deliberate as this is the group that is both more likely to be driving while intoxicated and already has an affinity to wearables.
“We actually used the kind of technology that they already love since they are already using it,” he tells Geektime, making breathalyzers appear like cameras, where the best one is the one that you have on you. Nessi hopes that the fact that the users can utilize existing technology will help make its use more widely spread.
Actually getting people to use a breathalyzer, especially if they are drunk, is probably the hardest part of making a dent in the issue of intoxicated driving. Even his solution, he points out, is still a voluntary process since the person would have to download the app and use it.
However there are situations like those in the US that require people who have already been convicted of drunk driving to use a breathalyzer.
Speaking with Geektime, Nessi believes that while autonomous cars are probably the future, it will likely take a long time until they become the only kind of vehicle on the road. As such, stop gap measures like his will be necessary for keeping us safe. Ideally as the connected car becomes more widely adopted, the app will be able to communicate with the car, alerting it that the driver is intoxicated and that it should not turn on. He notes though that the level of effectiveness of apps like his will be dependent on how they are able to interact with the car, and if the auto manufacturers will develop a sufficient API for permitting this.
Similarly, this kind of app could branch out into identifying other kinds of driving impediments like the influence of drugs or drowsiness, a less spoken about yet deadly problem on the road.
In spite of the strong showings during their trials, Nessi says that he is still hesitant to upload his app to the Google Play store for use. Deeming it to still be in the research stage, he feels that it would have to be more accurate before allowing people to depend on it for getting behind the wheel safely.
Looking at Nessi’s work, there appears to be a wealth of potential use cases for his technology but more so for his approach of constantly collecting data through IoT. This is hardly a new concept since most health trackers, including FitBits and Apple Watches, are already doing this. However, bringing the idea of providing automated warnings before risky behavior feels like a very good use for this kind of machine learning and monitoring tech.
Looking further ahead, this could be used to maybe even pick up on signs of something like an upcoming heart attack, letting the user know to get medical attention.
Some cities may find this app as a cheaper and more effective alternative to keeping tabs on those with DUIs. Parents could make having the app on their teen’s phone a condition of them driving. The more options there are for taking decision making and paying attention to detail out of the hands of people, I argue the better.
Now as a software engineering intern at Google, Nessi is continuing to develop smart solutions that one can hope will save lives in the future.
At the end of the day, it is still up to the user — even the less than thoughtful one — to put the right systems into place to prevent them from making stupid choices. This can be things like downloading the app, paying attention when it tells you that you are too drunk to drive, or once the connected car becomes advanced enough that it can interact with the app, requiring the vehicle to receive the okay from the app before turning on. All of these are steps that a responsible person who is less likely to drive drunk in the first place will probably take. To address the issue of the wider public and reach those that are more likely to drive while intoxicated, it will likely require legal regulations, such as was what occurred with seat belts or child seats.
So while we may need to wait a little longer for Nessi’s app to start impacting our lives, this New Year’s Eve, do yourself a favor and order a cab or Uber if you must, but please avoid the wheel if you’ve been drinking. 2016 has been tragic enough, and we all need some better vibes heading into 2017.
To read the study, use this link.