Above: Andrew Blumenfeld, co-founder of Telepath, and Lily Adelstein, creative project manager, discuss Amazon's Alexa announcement.
Among the many new announcements at Amazon’s recent fall event was an exciting update that provides a window into the future of machine learning technology. By now consumers are well aware of voice assistants, and Amazon’s Alexa leads the market. In fact, roughly one-in-four US adults owns a smart speaker, with 70-75% of that market share estimated to be captured by Amazon. This makes smart speakers in general, and Alexa in particular, one of the most recognizable and regular interactions that the average consumer has with machine learning technology. Machine learning is core to Alexa’s ability to understand natural language requests that are made by the user, and to predict how to most accurately respond to those requests.
So what’s new? Amazon has announced that Alexa will soon be capable of more explicit and complex learning in response to training provided by individual users, utilizing individual user’s data. For example, you can teach Alexa to recognize the sound your oven makes when it has preheated completely, so that Alexa can notify you or take other automated follow-up actions. Alexa can also “see” using the Ring video software, so you could train Alexa to recognize something like a door being ajar, or some other non-standard visual cue you wish Alexa to learn and respond to. Excitingly, Amazon indicates that you will need to train Alexa on as little as ten or so examples.
These are pretty stunning developments for machine learning in at least two ways. The first is technical: the ability for Alexa to learn on such a small training set of data represents major advances in the field of machine learning technology that will have the potential to make it far more accessible in many other contexts.
The other major shift here relates to consumer demand and expectations. As indicated at the outset, Amazon Alexa is a startlingly ubiquitous technology, and its newfound capacity to leverage machine learning to significantly increase consumer personalization is a potential game-changer. To-date Alexa has relied on machine learning to make a new type of technology (i.e., voice assistants) workable and less burdensome. Improvements in the technology have largely been most evidenced by reductions in error and churn rates.
The updates recently announced by Amazon are something different. They’re not about improving accuracy, they are about bringing a new level of personalization to the experience that opens up a whole new set of use cases and experiences. And because of Amazon and Alexa’s omnipresence, it could very well help to swell a new wave of user demand for the type of hyper-personalized experiences machine learning ostensibly has promised. So far that promise has not yet been realized in most consumer’s day-to-day lives. This announcement may signal the coming end of that.