But Alexa’s opinions would be neither of those things. They would be her own.
“This is the 2018 version of the video buff at the video rental store,” said Whitten. It’s like when you ask a friend what’s good to watch, and they offer a suggestion that’s not necessarily one of the top-rated shows – just something they are enjoying.
“This is the power of machine learning. One of the most interesting things we’re going at is how do you design an assistant that feels like you’re having a conversation with someone,” Whitten said.
Machine learning and deep learning networks are a key part of how Alexa will eventually offer more than just the fact graph – one of Google’s key strengths today. A separate team at Amazon has been rapidly improving Alexa’s ability to answer questions by adding more fact-based information to Alexa’s knowledge base, while also observing where holes still exist by analyzing users’ queries that the assistant can’t answer.
But while facts are important, what makes an assistant feel fun and friendly are its quirks, jokes, and other tricks that give it a personality.
For example, Alexa will belt out a cheesy country ballad if you ask her to sing for you, or she’ll tell you that she thinks “infrared is really pretty” if you ask about her favorite color.
Today, however, these sorts of things are more explicitly built into Alexa’s programming. The longer-term goal is that Alexa would come up with more answers on her own starting with her own set of opinions that aren’t curated by an editorial team inside Amazon.