Amazon Wants to Sell You the Kind of A.I. That Pilots Its Drones

Amazon’s same-day-or-sooner fulfillment pledge doesn’t come cheap. The company has invested $33 billion in recent years on the purchase of trucks (19,000), delivery vans (30,000), cargo planes (50) and drones. Lots of fully autonomous drones.

These UAVs won’t just be powered by propellers and motors, but by machine learning, Amazon chief technology officer Werner Vogels explained Thursday in a keynote presentation at World Summit A.I. in Amsterdam.

“Safety is the most important thing. We use many different sensors to ensure it doesn’t crash into anything, or that any other bad things will happen,” he says. The e-commerce giant will equip the drones with its computer vision algorithms so that when it finally gets the regulatory green-light to launch the service, the winged machines will be able to land at the correct door step without bumping into anything.

That’s no small feat.

Vogels shared a number of examples of what the drone could encounter en-route: all manner of flying objects, clothes lines and power lines, and, closer to the ground, humans and their pets. Its eyes will be cameras and sensors built with a combination of stereoscopic, image segmentation and heat-mapping technologies that ensure it avoids trouble on the way to the landing zone outside your front door

The drones can carry packages weighing up to five to seven pounds. That turns out to be convenient as more than 70% of Amazon deliveries weigh in at about that payload, he said. Delivery is expected to take no more than 30 minutes.

When will the skies fill up with delivery drones?

Amazon delivered its first package by drone in the United Kingdom in 2016, the symbolic launch of Amazon Prime Air. Regulatory approval is still needed to scale that service in most countries, including the United States.

“Soon” was the only hint Vogels gave on Thursday for the wider launch date.

The company wants to corner the market on the kind of machine-learning technologies it’s developed for its drones—as well as for its Amazon Go retail format and on its shopping site.

It’s aggressively marketing its SageMaker suite of machine learning through its AWS platform as it sees a lucrative market beyond its traditional retail business. Tens of thousands of customers now use AWS machine learning services, the company says. Early adopters include the National Football League, which uses the machine-learning APIs for the Next Gen Stats feature on its broadcasts, plus Duke University, which uses the computer vision capabilities to test for early warning signs of autism in children.

For much of the first day of the conference, the talk around A.I. take-up was met with caution about concerns around bias and a lack of fairness creeping into these systems that promise to automate workflows.

Amazon and event sponsor Huawei though took a decidedly more bullish view in their pronouncements.

At the Huawei booth, the Chinese telecoms giant showed a video of a robot-dominated workplace of the future. It predicts that more than 80% of all business processes can be automated through the deployment of A.I. in the coming years.