Uber Files Patent For AI Device That Spots Drunk Passengers

by Samuel Abasi Posted on June 12th, 2018

San Francisco, California, USA: The ride-sharing company Uber has filed a patent application with the United States Patent and Trademark Office (20180157984) for a device that uses machine learning to help warn drivers about a drunken passenger and better match those inebriated to drivers with relevant training.

The patent application, published recently, details the use of a machine that would use an algorithm to determine the state of a passenger by identifying unusual behavior.

Uber described it as “Predicting user state using machine learning” and “a system coordinates services between users and providers. The system trains a computer model to predict a user state of a user using data about past services. The prediction is based on data associated with a request submitted by a user. Request data can include current data about the user’s behavior and information about the service that is independent of the particular user behavior or characteristics. The user behavior may be compared against the user’s prior behavior to determine differences in the user behavior for this request and normal behavior of prior requests. The system can alter the parameters of a service based on the prediction about the state of the user requesting the service.”

The patent application uses technology that would hone in on typos, how precisely a user clicks on links and buttons, walking speed and how long it takes to request a ride. It would also consider the time of day and where a ride is requested.

If a passenger is deemed drunk, that person also might not be allowed to be part of a shared ride. Some riders may not be able to get service at all.

Uber is a safe way for an intoxicated person to get home, but many drivers do not want to deal with mishaps that accompany drunken riders, such as throwing up in the vehicle.

Also, the majority of sexual assault cases filed against drivers are from riders who had been drinking. During the past four years, 103 Uber drivers have been accused of sexual assault and in many of the cases, police reports note the passengers were drunk.

Below is Uber’s summary of the device.

“A travel coordination system identifies uncharacteristic user activity and may take an action to reduce undesired consequences of uncharacteristic user states. The system uses a computer model to identify user and trip characteristics indicative of the unusual user state. A system receives information about transport services (e.g., trips) requested by users and/or provided by providers. The information includes information about trip requests submitted by users, data obtained by monitoring trips as they occur, and feedback regarding trips received from users and providers. Such monitoring and trip feedback may indicate whether a user’s state was unusual during the trip or when picked up by the provider. The system uses the data about past trips to train a computer model to predict a user state of a user submitting a trip request. That is, the model is trained to predict whether the user is acting uncharacteristically by analyzing the submitted request as well as data collected about the user before and/or during the time the user submits the request (and in some instances, after submitting the request).

When a user enters information to generate a ride request into a user’s device, the user’s device collects information about the user’s activity on the user’s device during data entry and includes the user activity data in the ride request to the system. The system compares the user activity data to determine the user’s state. For example, the user activity can include text input characteristics, interface interaction characteristics, and device handling characteristics. For example, text input characteristics may include the number of typographical errors entered by a user or the number of characters erased by a user while entering a search query. Examples of interface interaction characteristics include the amount of time for a user to interact with the user interface after new information (or a modified display) is shown to the user, or the user’s accuracy in pressing an interface element on the device. Example device handling characteristics include the angle at which the user is holding the device, movement of the device during the request entry, or a user’s travel speed.

The system receives the trip request from a user and generates a prediction about the current state of the user using the computer model. To predict user state, the system compares data associated with the trip request to data about past trip requests submitted by the user. Past trip information may be parameterized to a profile of the user and identify how the user activity of the current trip request deviates from previous (or “normal”) behavior for that user. The system can also compare data associated with the trip request to past trip requests submitted to the system by other users. For prior trips, such as those in which there was an incident due to unusual user behavior, deviations of user input information from a corresponding user profile are featurized to train a model that predicts user state. In an embodiment, the prediction of user state is a value representing a probability that a user is acting uncharacteristically when submitting a request.

Responsive to a user state prediction, the system may alter parameters associated with trip coordination for the requested trip. Some examples of trip variations include matching the user with only certain providers, alerting a provider about the user’s possible unusual state, and modifying pickup or dropoff locations to areas that are well lit and easy to access. Different trip parameters may be altered in different situations, such as depending on the predicted likelihood that a user is acting uncharacteristically. For example, when the likelihood is comparatively very high, the user may not be matched with any provider, or limited to providers with experience or training with users having an unusual state. Similarly, when the likelihood is comparatively low (but not non-zero), the system may match the provider normally and provide a notification to the matched provider of the possible state.

The features and advantages described in this summary and the following detailed description are not all-inclusive. Many additional features and advantages will be apparent to one of ordinary skill in the art in view of the drawings, specification, and claims hereof.”

Author

Samuel Abasi

Samuel Abasi

A computer programmer who is also very proficient with hardware, Sam follows sports, science and tech news and everything else in between with an unrivaled passion that keeps readers coming back. Sam is also proficient in use of most online journalism tools and Social media management
Phone
Email

Leave a Reply