Artificial intelligence isn’t all fun, games and entertaining gadgets.

When people put trust in AI technologies that aren’t proven 100% reliable, consequences can be tragic. The fatal autonomous Uber crash and it’s fallout have underscored this for the public. But some games, toys and low-stakes experimentation are a great way to help produce AI for prime-time, real-life use.

“A little bit of competition really gets developers juices going,” said Mike Miller (pictured), director of AI devices at Amazon Web Services Inc.

Take AWS’ DeepRacer — a 1/18th-scale car with trainable machine learning technology built in. Developers at all skill levels have demonstrated great enthusiasm for the product. They relish regular contests where they can display how they’ve trained the mini autonomous vehicle to drive faster and more safely around a track. And their discoveries and accomplishments could make their way into the AI real people will depend on, according to Miller.

Miller spoke with John Furrier and Dave Vellante, co-hosts of theCUBE, SiliconANGLE Media’s mobile livestreaming studio, during the AWS re:Invent event in Las Vegas. They discussed the gamification of AI development. (* Disclosure below.)

From just playing to prime time

DeepRacer employs a form of ML called reinforcement learning that improves models through a reward function. Developers may choose to reward a model for achieving an objective or behaving a certain way in a specific context. In this way, they can train the car to stay closer to a track’s center line, take fewer turns and the like. (The record for fastest trip around a track — 7.44 seconds — was set by sola@DNP, the first female winner of a DeepRacer contest.)

AWS has added sensors to the updated DeepRacer Evo that allow for new ways to train the car. There’s a lidar — a laser-powered technology for range detection — on its hood and stereo cameras on its front for depth sensing.

“Developers can now be challenged by integrating depth sensing, object avoidance, and head-to-head racing into their machine learning models,” Miller said.

In the AWS DeepComposer keyboard, AI takes on AI with generative adversarial networks. Neural nets work against each other to produce original music.

There is a bridge between the recreational use of these products and serious R&D at tech companies, Miller pointed out. DeepRacer development could find its way into robotics; DeepComposer’s into product development and more.

“It’s really about reducing the learning curve and making it easy for developers to get started,” Miller said. 

Here’s the complete video interview, part of SiliconANGLE’s and theCUBE’s coverage of AWS re:Invent. (* Disclosure: Amazon Web Services Inc. sponsored this segment of theCUBE. Neither AWS nor other sponsors have editorial control over content on theCUBE or SiliconANGLE.)

Photo: SiliconANGLE