What is John Giannandrea, Google’s former head of search and AI, going to do at Apple? The short answer is “a lot.” The more pressing answer is “fix Siri.” Luckily, Giannandrea seems to be the right person for the job.
The 53-year-old Scottish engineer was poached by Apple in April, but his new title and position were only announced yesterday. He is now “chief of machine learning and AI strategy,” which means he will be overseeing a team that combines two of the iPhone maker’s most important AI products: Core ML and Siri. Core ML is Apple’s machine learning framework for apps. It helps developers run AI tools smoothly on mobile devices, and it’s essential for Apple to maintain its lead as the home of the world’s most advanced apps. And Siri... well, you know Siri. It’s Apple’s underpowered AI assistant that was first to market among the big tech companies, and it has been losing ground ever since.
Which is where Giannandrea comes in.
Looking at the new Apple AI chief’s history, it’s easy to pick up on some consistent themes. Time and time again, Giannandrea has worked on technology that involves processing speech and text and providing useful information to users. That seems perfect for someone who is tasked with upgrading a computer assistant.
Giannandrea’s past work is pure Siri: speech, text, and answering questions
One of Giannandrea’s early roles was at a startup named TellMe, which ran a phone line that used voice recognition to answer callers’ queries on things like the weather and the stock market. (Sound familiar?) After that, he co-founded a company named Metaweb Technologies, which pitched itself as a “database of the world’s knowledge.” Google bought Metaweb in 2010 and integrated its tech into the Knowledge Graph. (Those are info boxes that pop up when you Google a specific question like “how old is Tim Cook?” As of 2015, they answer roughly one-third of all Google search queries, suggesting that whatever Google paid for Metaweb, it was worth the price.) Giannandrea also worked on RankBrain, which helps Google answer search queries it’s never seen before, and Smart Reply, which suggests automated replies to email and texts.
Broadly speaking, all this work falls under the category of “natural language understanding,” or NLU, a sub-field of artificial intelligence that’s all about teaching computers to understand language like humans. It’s incredibly challenging, but Giannandrea has said he finds it to be the most exciting part of AI. In a 2017 interview, he even referred to natural language understanding as the “holy grail for applied artificial intelligence.”
But how does this all relate to Siri? Well, in many ways, Siri is just an answering machine. When we use it, we’re either trying to retrieve information (like finding out what the weather will be like tomorrow) or carry out a command (like playing a song). In both cases, it requires the computer to understand what you want and how to get you what you need.
This may seem relatively pedestrian compared to, say, the exciting worlds of robotics or augmented reality. But giving computers the ability to sort information with the same aptitude as a human is a core competency of AI. As the analyst Benedict Evans explained in an insightful blog post called “Ways to think about machine learning,” we can understand that contemporary AI can do three things: deliver better answers to the questions we ask our data, allow us to ask new types of questions, and allow us to query new types of data altogether (e.g., audio, images, and videos as well as words and numbers). By all accounts, this is what Giannandrea is extremely good at.
In that same 2017 interview in which Giannandrea called NLU the “holy grail,” he also referred to his admiration of Douglas Engelbart, an inventor who, in the 1960s, developed many foundational technologies that we still use today, like the mouse and hypertext. Engelbart’s guiding principle was to develop tech that “augments human intelligence,” and Giannandrea echoed those comments.
Searching for this dream in the Siri we know seems a little laughable, given the assistant’s many limitations. But it’s not a bad way to think about how Siri (and other assistants) might develop in the future. As AI gets better at processing data, these could become truly useful assistants that perform ever more complex tasks for us. Google has shown it’s thinking along these lines with its Duplex tech, and Apple is pursuing greater complexity through its new Shortcuts app, which allows users to set up complex, multistep instructions using Siri.
Obviously, there’s a lot of work to do before AI assistants fulfill Engelbart’s dream of augmenting human intelligence. But with Giannandrea now in charge of Apple’s efforts, it seems like we might have more of a race on our hands.