When we introduced Alexa in November 2014, she could do 13 things – she could tell you the weather, play music, answer basic questions, set timers, and catch you up on news headlines. Customers told us the experience was magical – it was a completely new and improved way to interact with technology. But our vision was to invent the Star Trek computer – and Alexa needed to know a lot more before this vision could be realized. We’ve been hard at work to teach Alexa even more – from new languages, to better natural language understanding, to adding new features that make her more useful. There’s still a long way to go, but we’re excited about Alexa’s progress, which wouldn’t be possible without the teams across Amazon who are behind the scenes solving some of the hardest AI challenges in the world.
The ADS team and data associates play a critical role in making Alexa smarter. Many of the scientific breakthroughs we’ve made with Alexa wouldn’t be possible without their hard work.
Rohit Prasad, vice president and head scientist for Alexa AI

Of the many challenges, teaching Alexa the many nuances of human language is one of the most difficult, because what might be easy for a human to understand is hard for computer systems. It requires a dedicated team of Amazonians with high judgment constantly training machine learning models about the nuances of natural language. This is the focus of the Alexa Data Services (ADS) team, which is made up of data associates, language experts, and scientists who work together every day to create the foundation that Alexa AI is built on.

“The ADS team and data associates play a critical role in making Alexa smarter,” says Rohit Prasad, vice president and head scientist, Alexa AI. “Many of the scientific breakthroughs we’ve made with Alexa wouldn’t be possible without their hard work.”

The team reviews a small fraction of one percent of requests made to Alexa in order to train her to more accurately interpret requests and provide the best responses to customers in context. Data associates do not have access to identifying information from customer accounts as part of this process, which protects customer privacy while still allowing the team to do this important work. For example, a data associate reviewing a request for the weather in Austin can identify if Alexa misinterpreted it as a request for the weather in Boston. Or they can teach Alexa that if a customer in the UK asks about when the Spurs play, they are likely referring to the soccer team in London, versus the basketball team in San Antonio. The team’s work also helps teach Alexa to better understand customers with different accents or unique speech patterns, so she works well for everyone. This feedback helps Alexa understand the correct interpretation of customer requests and provide the appropriate response in the future.

Along with getting to work on exciting emerging technology, many ADS team members learn valuable skills that have helped propel their careers at Amazon and beyond.

Here are some of their stories