Alexa Roundup: Amazon Improves Its Wish Fulfillment Device

From its earliest days, Amazon has built its business on removing the barriers between people and the things they want to buy.

New voice computing technologies and business models unveiled at the company’s annual developer conference in Las Vegas this week represent the continuation of that vision. Alexa and the ecosystem of services and skills growing up around it amount to something approaching the magical genie in the lamp, or cylinder, or refrigerator, as the case may be—albeit one that grants wishes using your credit card. It’s frictionless commerce nearing perfection.

Alexa’s continuously improving ability to understand what customers say—aided recently by university teams competing to build a better chat bot—and to act on those commands, allows Amazon to dispense with some of the trappings of e-commerce’s first two decades. Why bother logging in and clicking “add to cart” when you can simply speak your wants and trust the company’s system of systems to turn those spoken words into physical objects packed in brown boxes, arriving on your doorstep in days or hours?

The technologies and business models Amazon deploys to make this near-instant wish fulfillment real are still maturing, but the incremental improvements are relentless.

Here’s a selection of Alexa-related announcements pouring forth from the re:Invent developer conference this week, focusing on Alexa’s expansion geographically and into the world of business; its new initiatives to spread the money-making potential to developers; and a big win for a University of Washington team in Amazon’s Alexa Prize competition.

—Amazon (NASDAQ: AMZNsees a big opportunity for voice technology in the workplace, and announced a series of new Alexa capabilities and integrations to make it more useful in business applications. Amazon envisions companies putting Alexa-enabled devices at workers’ desks, in conference rooms, and elsewhere in businesses, provisioned and managed centrally like traditional corporate IT infrastructure.

Through integrations with service providers such as Concur, Polycom, and Salesforce, employees can ask Alexa about upcoming business travel, initiate video conferences, and get an update on sales leads, respectively. Companies will also be able to build custom skills for use by their employees.

This looks like good news for Seattle, uniting its legacy in enterprise software and the current blossoming of voice technologies.

(DeVore is managing director of Techstars Seattle, which hosted the Amazon Alexa Accelerator earlier this fall.)

—Speaking of startups, Amazon re-loaded the Alexa Fund—a corporate venture investing vehicle focused on voice—with $100 million, earmarked for investments in companies outside the U.S.

In other international expansion news, Alexa and the devices it runs on will be available in Australia and New Zealand by early next year, the company says.

—In addition to fostering startups that thrive on its voice platform, Amazon wants to enable developers building skills for Alexa to have profitable businesses.

The company introduced a way for developers to monetize the Alexa skills they build that’s akin to the freemium model the casual games business is built on. Developers can continue to offer free content in their skills—trivia games, for example. Soon, they will be able to offer customers the option to pay for premium upgrades—access to Double Jeopardy, say—on a one-time or subscription basis. These in-skill purchases will become available early next year.

Developers already make money through Alexa by integrating it with apps and other businesses. Amazon is trying to make that easier with a new instance of its payment platform, Amazon Pay, that will allow people to order and pay for things like food and movie tickets by talking to Alexa.

Amazon also directly pays some developers who build popular skills for use on its voice platforms by people in the U.S., U.K., and Germany. Those payments will soon be offered for skills used in Japan, India, Australia, and Canada, the company said.

—Underpinning many of these voice applications is the ability of the machine intelligence to carry on a conversation with a human. Amazon and its competitors are racing for advantage in this area, as it is fundamental to making voice a successful new user interface.

Amazon enlisted the help of top computer science students from around the world through the Alexa Prize, a competition to build a better chat bot using the Alexa Skills Kit and AWS cloud services. The winner, announced earlier this week, is from the University of Washington.

The UW team, Sounding Board, built a chatbot that could engage in conversation with a human for an average of 10 minutes, 22 seconds. Judges scoring the conversations rated it 3.17 out of 5.

The company had challenged teams to build an A.I. system that could carry on coherent, engaging conversations with humans for 20 minutes—dangling before the competitors a $1 million research grant for a winner that could accomplish this. Amazon plans to run the competition again, with the 2018 Alexa Prize competition opening for applications on Dec. 4.

The 15 semifinalists winnowed from a field of more than 100 entries could then develop their bots further through interactions with Amazon customers. Beginning in July, users could say “Alexa, let’s chat,” to begin conversations. This provided the teams a stream of usage data and real-world human feedback that was invaluable. The chatbots with the best customer scores during this phase of the competition became finalists.

Amazon used professional conversationalists and judges to evaluate the finalists over two days in November at its Day1 headquarters in Seattle, says Ashwin Ram, senior manager, AI Science and Alexa Machine Learning at Amazon, in a blog post announcing the winners.

From the company’s perspective, the contest was a huge success. Alexa “took a great step forward in conversing coherently and engagingly with humans on popular topics and news events,” Ram writes.

Ram says Amazon will publish academic papers by the semifinalist teams, detailing the computer speech advances they developed. Even though the work will be shared with the rest of the computer speech community, Amazon seems best positioned to take advantage of these advances—in things like language models and dialogue management—given that they were made using its own tools and infrastructure.

The Alexa Prize winning Sounding Board team from University of Washington. Photo via UW

The UW Sounding Board team (pictured above, in part) is comprised of electrical engineering and computer science PhD students Hao Feng, Hao Cheng, Elizabeth Clark, Ari Holtzman, Maarten Sap, and professors Mari Ostendorf, Yejin Choi, and Noah Smith. The students will share the $500,000 first prize. The runner up was the Alquist team from Czech Technical University in Prague, which took home $100,000. Third-place, winning $50,000, was the team from Heriot-Watt University in Edinburgh, Scotland.

“One of the biggest challenges we faced when designing Sounding Board was the diversity of people we interacted with,” says Clark, in a blog post on the UW Paul G. Allen School of Computer Science & Engineering website. “We needed to handle a wide range of topics, conversational styles, and user personalities and interests.”

The Sounding Board bot was built to develop an understanding of its individual human interlocutors through analysis of a person’s comments and stated interests. It adjusts “parts of the conversation based on who it thinks the user is,” Sap says.

In other words, the genie in the lamp wants to get to know you better. (Amazon also announced new personalization features that enable Alexa to give tailored responses when it recognizes an individual customer’s voice.)

Credit: Feature image by JD Hancock via Flickr used under a CC BY 2.0 license

Trending on Xconomy