Uber's new algorithm can tell if passengers are drunk when they hail a ride

We get it, drunk people are no fun to deal with if you’re an Uber driver. Maybe they fall asleep, or throw up in your car.

Hitching a ride in an Uber can be dangerous for intoxicated passengers, too. In the past four years, riders have accused more than 100 Uber drivers of sexual assault or abuse, and no one can say for sure how many other incidents go unreported. This is, obviously, not great for the passengers, but it’s also not great for the company, which gets slapped with lawsuits and bad publicity. Best for everyone just to avoid all this, right?

So, intoxicated passengers are a problem for Uber. And based on a patent application, the company thinks it’s found a way to solve it: Just don’t give drunk people rides.

On Thursday, as CNN Money reported, the U.S. Patent and Trademark office published an application Uber filed in December 2016. The application details a system that uses machine learning to predict an Uber user’s “state.”

The application never outright claims to focus on intoxicated riders. Instead, it assumes an overly “tired” passenger who “might have difficulty locating the provider’s vehicle.” But take a look at the factors about a passenger’s ride request the system considers when making its prediction, and see what you think:

How accurately and quickly the passenger types in their information the angle at which the passenger holds their device, the passenger’s walking speed, the passenger’s location, the time of day and the day of the week

Based on our experience, this seems pretty on point for signs of inebriation (not sure what the angle a person is holding their phone has to do with how sleepy they are, but maybe Uber knows something we don’t).

Once it has this info, the system then compares it to a passenger’s history with Uber. So, an abundance of typos, for instance, might not raise any red flags if a passenger makes typos every time they use Uber. If the proposed system determines that a passenger is in an “unusual” state, however, it can then adjust how Uber responds to the ride request. For example, it might:

Let the driver know the passenger is possibly in an unusual state. Only match the passenger with a driver with experience dealing with people in unusual states. Change the pick-up or drop-off location to somewhere well-lit and easily accessible. Prevent the passenger from joining a carpool Choose to deny the passenger a ride

That’s right. According to the patent application, “When the likelihood [that a user is acting ‘uncharacteristically’] is comparatively very high, the user may not be matched with any provider.”

While several aspects of Uber’s proposed system could actually help cut down on unwanted incidents — training providers to deal with people in unusual states, for example — outright denying passengers rides is kind of a bad idea. This could lead to lawsuits from injured or disabled passengers denied rides based on their walking speed, or how the company could still find itself on the receiving end of public outcry if it denies a passenger a ride and that passenger then ends up in a drunk driving accident.

So, sure. Cleaning puke out of your car mats for the 10th time this week might not be fun, but, hey, carting around drunk passengers is a cornerstone of Uber’s business. Unfortunately for drivers, it comes with the territory.