What’s the relationship between money and happiness?
Emily McManus, editor of TED.com, can’t tell you the answer to this age-old question. But for a long time, based on the hundreds of TED talks she’s watched since 2007, she could point you to relevant takes from some of the world’s top thinkers. However, with more than 2,000 talks posted to date, even she’s no longer seen every single one. “No one human has,” McManus says.
Now she has the help of IBM’s Watson, the conversationally-conversant AI computer system that famously won Jeopardy! in 2011.
Over eight weeks, an IBM team created a unique tool for exploring TED’s body of work that it will demonstrate at a Watson developers conference it’s holding in New York City on May 5.
When asked a question, the program surfaces TED talks that touch on the relevant concepts and cuts helpfully to the exact section of the talk. The “search result,” with one cut played after another from different talks, is something like a super mashup of TED speakers taking on whatever deep questions of society and the universe strike your fancy. There’s also a timeline that shows concepts that come up during each talk and data visualization that maps the connectedness of all the talks.
“We can pick up ideas that emerge,” says Jeffrey Coveyduc, IBM’s director of advanced cognitive technology. “It’s like a rolling discovery tool, and it’s all concept-driven. There are topics that come up that are never explicitly mentioned,” he notes.
Another, perhaps even more interesting feature is that Watson is also doing a sentiment and psycholinguistic analysis on each speaker’s talk. A talk given by global health expert Hans Rosling on “How not to be ignorant about the world,” for example, summarizes his presence as “heartfelt and tranquil” (sounds about right for TED) and notes his presentation has a relative small emotional range and high degree of openness. Such a feature could eventually analyze speaker traits across the broad spectrum of presenters and surface other unexpected connections, the researchers say.
The TED program is being tested and trained by a small group of users now, but could be available publicly as soon as this summer. In a demo at IBM’s Watson office, I posed the software my own question: “What’s the relationship between income inequality and crime?” and surfaced a few interesting talks from the TED library–some more obvious than others. For example, it pointed me to sections in the Rosling talk mentioned above as well Richard Wilkinson’s, “How economic inequality harms societies,” Andrew McAfee’s, “What will future jobs looks like,” and Loretta Napoleoni’s “The intricate economics of terrorism.” Of the latter talk, which gets into how terror networks target the poor, McManus says, “It’s a good pick. I wouldn’t have thought of that.”
Like the Jeopardy! matchup and Watson’s sometimes dubious attempts at cooking, the TED program is partly meant to help IBM introduce Watson to the world in a non-threatening way.
Last year, the company launched a $1 billion business unit dedicated to commercializing the software, and it is now courting app developers, hospitals, universities, retailers, and many other sectors to jump onto what it calls the “cognitive computing” revolution. It announced a number of new apps and partnerships at the World of Watson event today, and says there are “tens of thousands” of developers on its platform already.
IBM’s not the only company developing this kind of emerging technology, but it is among those pursuing it most aggressively. More than a search tool, the idea is that Watson would serve as smart collaborator or intelligent assistant to help people make sense of the ever-growing amount of information available–more than people’s brains can possibly process.
The TED demo also serves as an early foray into Watson exploring video, rather than only text, images, and audio. Back when Watson played Jeopardy!, Alex Trebek has to refrain from video clues, but now the software is more capable, the researchers say. To develop the TED tool, the team processed TED’s existing transcripts of talks and used Watson’s speech-to-text technology to add more detailed information. And as video and audio become more important on the web (take MOOCs, for example), this kind of work will be needed if Watson wants to be a valuable tool that’s worth paying for.
“Sometimes it’s challenging to actually find the ideas that you want within this body of information,” says Kai Young, the IBM program director who led the prototyping team. “And with the proliferation of video being the dominant form of data on the web, that problem is only going to grow.”