IBM's Watson vs. the human brain: Tale of the tape
Which has the real computing advantage in "Jeopardy!" showdown?
- By Kevin McCaney
- Feb 10, 2011
History has witnessed some epic battles of man vs. machine.
Charlie Chaplin vs. the dehumanizing gears of the factory.
World Champion Garry Kasparov vs. Deep Blue at chess.
Homer Simpson vs. an electronic voting booth.
The battle moves to new ground Feb. 14-16, when Ken Jennings and Brad Rutter go up against IBM’s DeepQA supercomputer, better known as Watson, on “Jeopardy!”.
The basics are pretty well known.
In one corner is Watson, the result of four years worth of work by 25 IBM research engineers into building a system that analyzes natural human language – including the puns, plays on words and ironic deliveries used on “Jeopardy!” – and draws on about 200 million digital pages of books and other information sources to come up with the answer to a question (or, on “Jeopardy!”, the question to an answer).
IBM's 'Jeopardy!' match more than game playing
In the other corner are the all-time best "Jeopardy!" players, Jennings (74 straight wins) and Rutter ($3.25 million in winnings).
So what are Jennings and Rutter are up against? And for that matter, what is Watson up against? Here’s a tale of the tape.
Raw processing power
Machine: Watson is a Linux system powered by 10 racks of IBM Power 750 servers with 2,880 processor cores. It is capable of operating at 80 teraflops, or 80 trillion floating point operations per second.
Man: The computing specs of the human brain involve some guesswork, but many estimates, based on the estimated number of neurons and synapses in the brain, put the processing power at about 100 million MIPS, or 100 million million instructions per second. In floating-point terms, estimates have made it 100 petaflops (100 quadrillion flops). Of course, since the brain also is occupied with running the functions of the human body, how many of those flops are focused on the question at hand at any given time is a mystery.
Advantage: Machine, because of the focus question. Jennings and Rutter have pretty much proved to be machines at playing “Jeopardy!”, so we assume their ability to focus is pretty good, but Watson’s processors don’t have to worry about regulating body temperature or otherwise dealing with the subconscious, for instance.
Machine: Watson has 15 terabytes of RAM. It doesn’t really have the equivalent of a hard drive; its in-memory database doesn’t require external retrieval, which only speeds things up.
Man: Even more of a mystery than raw power, but estimates have ranged from 750 gigabytes to 6.4 terabytes. Most likely it can’t be determined by any reasonable measure.
Advantage: Machine. At least it knows what it has.
Natural language processing
Machine: For Watson, this is the real test. Solving equations has always been natural for computers. Processing speed can be added easily. But interpreting language is a big step forward – and the reason Watson was built in the first place. IBM scientists have spent four years teaching Watson to read, listen and think. The machine won a practice round against the humans, but this time, money will be on the line.
Man: Watson has learned a lot, sure, but it still starts out with ones and zeroes. Jennings and Rutter will be working in their native tongue, which has to count for something.
Machine: Bright. IBM says Watson represents a great leap forward in what is called natural language processing and statistical machine learning, and could lead to advances in search and analytics. Among the examples the company cites are improving health care diagnoses, examining potential drug interactions, speeding up legal searches for case law, and developing what-if scenarios involving regulatory compliance. And its popularity could be only beginning. Studies have found that 18-month-old babies considered a shiny silver robot to be human, as long as it was friendly and that people developed warm feelings for their Roombas – robotic vacuum cleaners that look like a round weight scale – even to the point of treating them like pets. A gentle-voiced supercomputer that answers all your questions will be a big hit on Facebook. People will be calling it “he” in no time.
Man: Undetermined. The human race has produced Albert Einstein and Mother Teresa, but it’s also produced the characters on “Jersey Shore.” Whether we’re going up or down has been debated for centuries. But humans did also build Watson.
Advantage: Maybe Watson can answer this one.