Amazon today detailed a capability coming to Alexa called Teachable AI that will enable the assistant to ask questions to fill gaps in its understanding. First announced during the company’s September virtual press event, Teachable AI leverages machine learning to determine whether a request can be a trigger for a teachable moment. If Alexa makes this determination, it will ask a customer for information to help it learn.
Amazon says Teachable AI will become available in the next few months for smart home devices before expanding to other areas.
Scientists at Amazon’s Alexa AI research division have long pursued semi-supervised and unsupervised learning techniques, in which AI systems learn to make predictions without ingesting gobs of annotated data. Semi-supervised and unsupervised learning have their limitations, but both promise to supercharge Alexa and other voice assistants’ capabilities by imbuing them with a humanlike capacity for inference.
In the case of Teachable AI, if a customer says something like, “Alexa, set the living room light to study mode,” the assistant might now respond, “I don’t know what study mode is, can you teach me?” Alexa will extract a definition from the customer’s answer and when they later make the same (or a similar) request, reply with the learned action.
As another example, Alexa might not automatically know that a customer’s preferred reading setting is 40% brightness when they say, “Alexa, set the light to [my] reading mode.” Using Teachable AI, the assistant will ostensibly learn these definitions and corresponding actions and associate them with a particular account for future use.
“This is an exciting step forward not just for Alexa, but for all AI services that rely on end users to teach them,” Alexa VP and head scientist Rohit Prasad wrote in a blog post. “I am excited to [see] … AI advancements bringing us closer to our long-term vision of making interactions with Alexa as simple as speaking to another person.”
Teachable AI builds on latent goal inference, which enables devices powered by Alexa to infer goals implicit in requests but not directly expressed. For instance, if a user says “How long does it take to steep tea?” Alexa might follow up with “Five minutes is a good place to start” and the question “Would you like me to set a timer for five minutes?” Latent goal inference and Teachable AI also complement Natural Turn Taking, an Alexa feature that lets users converse with the assistant without having to repeat a wake word. (Three AI models run in parallel to power Natural Turn Taking, which will initially only be available in English when it launches sometime next year.)
