IBM's Watson, a Jeopardy Champ, Could Deliver Natural Voice Recognition That Actually Works

Jim Motavalli

Jim Motavalli | Jun 10, 2011

IBMers take on Watson at Jeopardy & get creamed. (Jim Motavalli photo)
IBMers take on Watson at Jeopardy & get creamed. (Jim Motavalli photo)

When Mr. Spock needed an answer from the Star Trek computer, he'd say something like, "Computer, how much power is left in the dilithium crystals?" After what sounded like a shuffling of punch cards (it was the 60s, after all) the computer's disembodied voice would get back to him with an answer that often spelled doom. "Five minutes until the self-destruct sequence is initiated."

I want voice recognition to work that way, but in fact although it's getting better (especially as seen in systems like MyFord Touch) it's still pretty clunky. Even the more sophisticated versions, which at least don't require users to input tedious lists of words, want you to ask for something with laser-like precision.

As cars get more complicated, and demand more of our attention, they also increase the possibility of driver distraction. Automakers are increasingly relying on voice recognition to combat that problem, but the technology is sometimes lacking. I saw one demo go completely awry in front of a roomful of journalists.

Because I don't remember if my friend lives on Homestead Street or Homeland Road, I don't meet voice recognition's high standards. And how can I ask for a song if all I can remember is that it has the word "love" in the title? I'm always getting these pregnant pauses, and the ultra-polite voice (male or female, your choice) asks, "Could you repeat your selection?" The whole car has to be quiet while you do this, too.

I despaired of the possibilities for voice recognition until in Orlando for the "Innovate 2011" conference, I saw Watson take on two panels of seasoned IBM programmers in a game of Jeopardy, and beat them handily. Actually, it slaughtered them--correctly unscrambling all the state capitol anagrams. And this was after its win earlier this year, when it took down "titans of trivia" Ken Jennings and Brad Rutter.

Watson is a super computer, powered by 90 servers and network-attached storage with 21.6 terrabytes of data. Through the raw power of 2,800 Power7 cores, it can perform at roughly 80 teraflops (which would be more impressive if I knew what a teraflop was). But I'm definitely impressed by futurist Ray Kurzweil's claim that it can process data at roughly 80 percent of a 100-teraflop human brain, with somewhat reduced data storage. That means it can, and does, answer questions in a very human way, in real time. Here's Watson explained:

On Jeopardy, Watson was hit with this from Alex Trebek, "Wanted for a 12-year crime spree of eating King Hrothgar's warriors; Officer Beowulf has been assigned to the case."

And he correctly answered, "Who is Grendel?"

Wouldn't that be perfect for auto-based voice recognition? "Natural language processing will change the world of business," said David Shepler, Watson's program manager. I just want it to help me find a song "by that guy with the beard who sounds kind of like Bob Dylan."

MyFord Touch: one of the better voice recognition systems. (Ford photo)
MyFord Touch: one of the better voice recognition systems. (Ford photo)

My basic concept is this: Since all that teraflop capacity won't fit into my car, the much-anticipated cloud can connect me to a Watson-like supercomputer in the sky, and I'll be able to get my intuitive answers in real time. What's more, I'll be clued in even though my original question lacked a vital element, such as the artist's name. I talked with Kal Gyimesi, automotive lead for the IBM Institute for Business Value, and he seemed to be saying that my solution is possible, but not just yet.

"At some point we'll be able to interact with our vehicles using natural, instead of rigid, language," Gyimesi said. "GPS wants exact addresses, when what you want to have it do is find the sushi restaurant you saw advertised in the paper. We've had some conversations with auto companies, but it's still three to five years away."

Watson and his friends, obviously, could change the world of search by simply answering your search request rather than giving you hundreds of pages to sort through. Ask a question, and like the Star Trek computer or 2001: A Space Odyssey's Hal 9000, it simply gives you the answer. And maybe that form of search could be less energy intensive. IBM's Grady Booch told me that a single Google search uses as much energy as heating up a cup of Starbucks coffee. That's a scary thought.

Get the Car Talk Newsletter

Got a question about your car?

Ask Someone Who Owns One