Amazon has always been at the forefront of Internet innovation, including Artificial Intelligence. I’m not just talking about Alexa, Amazon’s intelligent assistant and perhaps the most impressive consumer technology of 2016. Amazon also utilizes AI in its core product: the personalization engine that drives Amazon.com. Whenever Amazon’s website recommends an item to you (e.g. “Customers Who Bought This Item Also Bought”), part of the technology behind that is AI. More specifically, deep learning via neural networks.
Indeed, with the popularity of Alexa this year and its continuing dominance in e-commerce, Amazon has arguably been the most successful of the tech companies at commercializing Artificial Intelligence. Google, for all its promise, has yet to deploy AI in a big way outside of its search engine. Apple’s Siri never reached its potential. Microsoft has had mixed results with AI. Facebook’s AI backbone is an impressive feat of engineering, but it only impacts certain things in the News Feed (such as organizing images). Whereas Amazon’s recommendation engine is the core of its retail experience.
When it comes to making ‘smart’ consumer Internet products, Amazon just seems to have an edge. Which is why things look good/ominous (depending on your point of view) for Amazon’s retailing future. Allow me to explain…
Amazon’s AI Destiny
The technology behind Amazon’s recommendation engine was open-sourced in May of this year. It’s called DSSTNE (pronounced: destiny), which stands for Deep Scalable Sparse Tensor Neural Engine. “We use DSSTNE to train neural networks and generate recommendations that power various personalized experiences on the retail website and Amazon devices,” Amazon engineer Kiuk Chung wrote in a blog post.
It’s important to note that Amazon only recently started using AI in Amazon’s product recommendations. DSSTNE was developed in 2014, according to a presentation in July by its lead developer Scott Le Grand. In a separate comment on Hacker News, Le Grand said that it was created “specifically for product recommendations from Amazon’s catalog.”
Before 2014, Amazon’s recommendations came from a technique known as “item-to-item collaborative filtering.” In a January 2007 post on ReadWriteWeb (the blog I used to run), Alex Iskold analyzed some of the ingredients Amazon used in its recommendations: item relevance, personalization, and social signals. With DSSTNE, we can assume that Amazon made the system much smarter. Hopefully that means avoiding some of the glitches of the past, such as recommending more Ronan Keating CDs if you happened to buy your mother one for her birthday.
It’s also worth noting that while DSSTNE has been open sourced, Amazon’s vast and proprietary data set hasn’t. Or as Le Grand put it, Amazon hasn’t released “the actual networks, and definitely not how they prepare customer purchase histories.” The truth is, Amazon is supremely confident that no other online retailer can come close to matching its retail database. Therefore no other company will do recommendations as well, even if it uses DSSTNE. So there was little risk in Amazon open sourcing this technology, but there’s big potential upside. Amazon is hoping an external developer will extend DSSTNE in unforeseen ways; and thus give it an even bigger edge in online shopping.
Upskilling With Alexa
If DSSTNE sounds like the name of an obscure rapper, Alexa sounds like the next big pop music sensation after Beyoncé. And sure enough, Alexa has been Amazon’s star performer in AI.
Alexa doesn’t use the DSSTNE framework, according to Amazon executive Rohit Prasad. Instead it uses its own speech recognition system. While that technology hasn’t been open sourced, Alexa has launched a couple of platforms for third party developers to tap into. One is the Alexa Skills Kit (ASK), a collection of APIs and tools that allows developers to build new functions into the Amazon Echo.
Think of Alexa “skills” as akin to apps on a smartphone. There are now over 1,900 third party skills for Alexa (you can search them all here). One example is Honeywell’s Wi-Fi Total Connect Comfort thermostat, which you can control using voice commands. Just say “Alexa, set home temperature to 72” (or 22, if you live in my part of the world) and the temperature adjusts without you having to get up from the sofa.
Alexa Voice Service (AVS) is another third party platform for Alexa. It gives hardware makers a toolkit to integrate Alexa into their Internet-connected devices. It is early days for this service, but one example I found is Petnet. This company created an automatic feeder for cats and dogs, called SmartFeeder. According to ProgrammableWeb, “customers can use the [Alexa-powered] voice-activated assistant to ask specific questions about their pets feeding activities and manage other aspects of pet feeding.” (One of our cats caught a bird yesterday, so I would’ve loved to ask him how he did that – but I don’t think that’s what SmartFeeder does…)
Alexa, What Should I Buy?
The reason why Alexa has captured so much of our attention this year is because it has that rare combination of utility and a magical user experience. Tim O’Reilly wrote a great analysis of Alexa, calling it “the first winning product of the conversational era.” In particular, he showed how Alexa is far superior to Google’s equivalent voice interface, Google Now. That’s spot on, but I think there’s another angle to Alexa too: shopping conversations.
I’m willing to wager that Alexa will increasingly be your go-to source for conversations about what to buy. Think about how hard it is to choose a new TV. I’m in the process of doing that right now, and even with The Wirecutter and CNET‘s help I still find all the choices and features overwhelming. It’d be great to have an automated assistant to ask questions to and help me whittle down the choices. Alexa is self-learning, so over time she’ll develop the knowledge required to be your shopping assistant. Many times, she will point you to an item that just happens to be available on Amazon.com. Ok, other retailers will presumably tap into Alexa as well, through the third party services – but Amazon has a head start and will probably do it better.
Let’s face it, Amazon is amazingly good at stealthily coming to dominate everything retail. So it’s reasonable to assume that Alexa is another weapon in Amazon’s e-commerce arsenal. Of course it’ll be used for much more than just buying stuff, but you watch: Amazon will soon have a conversational interface direct to your wallet.