Amazon beefs up machine learning in Alexa to predict 'latent goals' that aren't directly expressed

Amazon's voice assistant in the U.S. can now predict "latent goals," or potential requests that a user doesn't explicitly ask. In the example above, when a user asks for the weather at Seaside Beach, Alexa responds with the weather, then asks if the user wants to set a live camera feed of Seaside Beach on the screen. Or, as described in a blog post, a user could say: "how long does it take to steep tea?" Alexa would answer with "five minutes is a good place to start," then follow with "Would you like me to set a timer for five minutes?" Amazon uses a "deep-learning-based trigger model" that analyzes the conversation and past interaction history, among other data, to help Alexa figure out appropriate times to ask these types of follow-up questions. Early data shows that latent-goal discovery is increasing customer engagement with certain Alexa skills, Amazon said.