I remember the much talked about 1996 & 1997 Chess
championship matches between then reigning champ Garry Kasparov and IBM’s “Deep
Blue” super-computer. Kasparov beat Deep Blue in ’96 series only to have his
tail handed to him in ’97. What made the difference a year later was pure “brute force”:
It was a massively
parallel, RS/6000 SP Thin P2SC-based system with 30 nodes, with each node
containing a 120 MHz P2SC microprocessor for a total of 30, enhanced with
480 special purpose VLSI chess chips. Its chess playing program was written in C
and ran under the AIX operating system. It was capable of evaluating 200
million positions per second, twice as fast as the 1996 version. In June 1997,
Deep Blue was the 259th most powerful supercomputer according to the TOP500
list, achieving 11.38 GFLOPS on the High-Performance LINPACK benchmark.
My teenage self was impressed.
But looking back, Deep Blue’s achievement seems not so awe-inspiring. After-all there was no real artificial intelligence (AI) at work, and
Deep Blue could not do much else than play chess. Chess -- though a complex game--is confined to a 64sq board and fixed rules of play. If we can build a computer
fast enough to process all possible lines of play after each move, to figure
out the optimal strategy, then such a computer can become nearly unbeatable at
chess.
Even IBM admitted as much, and at the end of the day Deep
Blue was a publicity stunt more than a real advancement for AI.
Enter Watson. IBM’s latest super-computer, and it's no gimmick.
IBM’s latest invention is the fastest, smartest super
computer on the planet and what really sets Watson apart is its ability to
understand natural language.
IBM put Watson to the test last year by making it compete in
a Jeopardy challenge against two of the smartest brains to compete on the game
show. Watson took on Ken Jennings, who holds the record for longest winning
streak on the show, and Brad Rutter, the biggest all-time money winner.
And Watson won!
And Watson won!
IBM chose the Jeopardy challenge to test Watson, since to
compete effectively on the show, Watson needed to be fast, be able to
understand human conversational language and the nuances associated with it,
form a statistical analysis of the correct possible answers after querying it
databases, and then make a risk/reward assessment as to whether to buzz in. In
the initial test runs Watson’s performance was dismal. It failed to understand
subtle differences in language and elementary distinctions easy for the human brain to process.
But what enabled Watson to improve was
its ability to learn from examples as it goes, using a complex statistical algorithm, powerful processors and memory, and vast database of human knowledge.
The key proved to be it's ability to learn from it's mistakes and building on patter recognition to process natural language. In other words, Watson learns as it goes. Once
Watson is fed the right answer it incorporates that pattern into its databases
to make a better analysis and return a better result next time.
“Watson is a workload optimized system designed for complex analytics,
made possible by integrating massively parallel POWER7 processors and the IBM
DeepQA software to answer Jeopardy! questions in under three seconds.
Watson is made up of a cluster of ninety IBM Power 750 servers (plus additional
I/O, network and cluster controller nodes in 10 racks) with a total of 2880
POWER7 processor cores and 16 Terabytes of RAM. Each Power 750 server uses a
3.5 GHz POWER7 eight core processor, with four threads per core. The POWER7
processor's massively parallel processing capability is an ideal match for
Watson's IBM DeepQA software which is embarrassingly parallel (that is a
workload that is easily split up into multiple parallel tasks)."
"Nova on PBS" recently
aired an episode on Watson and its trials with Jeopardy. You can watch it here:
In my future posts I will look at some of the business application
of Watson including its use in healthcare.
No comments:
Post a Comment