Alexa, Amazon.com Inc.’s increasingly ubiquitous digital voice assistant, is about to get a lot chattier.

Today at the company’s re:MARS conference in Las Vegas, Amazon said it has developed a new approach using deep learning, which has already produced new capabilities in voice and image recognition in recent years, that will help developers create more natural voice experiences more easily. Amazon also promised the approach, called Alexa Conversations, will use a third as many lines of code as the current approach and use 10 times less data to train the algorithmic models.

The result, the company said, will be more natural and flexible dialogs both within individual skills, as Alexa’s capabilities or apps are called, and allow multiple skills pretty seamlessly in a single conversation. They’re available now in a developer preview.

Rohit Prasad (pictured), vice president and head scientist for Alexa, demonstrated the new capabilities onstage at re:MARS, which includes a new “cross-skill action predictor (below).” In a demo that didn’t appear to be live, a woman asked Alexa what movies are playing nearby, after which Alexa proactively not only suggested booking tickets but also dealt easily with changes in movie times desired, asked if she wanted to book a restaurant nearby, gave options for Chinese food she wanted, booked a reservation on OpenTable and then asked if she needed a cab or Uber and booked that.

“This is a big leap for customer-focused AI,” Prasad said. Before, it would have taken 48 “user turns” or requests but now took only 13, shifting cognitive load from customers to Alexa.

More specifically, an artificial intelligence-driven “dialog manager” with an advanced dialog simulation engine that automatically generates synthetic training data. Using developer-provided application programming interfaces for a skill, along with annotated sample dialogs with prompts for Alexa and actions customers are likely to take, Alexa Conversations generates dialog flows and variations using a recurrent neural network.

“At runtime this neural network takes the entire session’s dialog history into account and predicts the optimal next action or step in the dialog, improving accuracy and reducing your design and code efforts,” Amazon explained in a blog post. “It is trained to interpret dialog context in order to handle multiple user workflows, accommodate natural user input (like out-of-sequence information or corrections), address common business transaction errors, and proactively recommend additional API functionality.”

The backdrop for all this is that customers are continually asking more of Alexa for actions such as ride booking, shopping, ticket purchasing and making reservations, and more than that, expect Alexa to converse more fluidly on multiple topics.

Amazon also said today that Alexa has became 20% more accurate in understanding requests.

Photo: Robert Hof