The last time a computer famously beat humans, the domain was chess. It is pretty easy to understand the basic brute-force technique that a computer chess engine uses to play the game. See How Chess Computers Work for details.

The Watson program is more complicated than that. The “rules” of language are far more complex than the rules of chess, and brute force chess requires no real knowledge base ahead of time. But Watson is still using something of a brute force technique to play Jeopardy.

Watson is a brute-force approach, yes, but an interesting one. They fed the Watson algorithm normal text data. So they fed it all of the English version of Wikipedia as normal text (about 14 GB of text), another encyclopedia, a complete dictionary, a complete thesaurus, the Bible, the Internet Movie Database, plus lots and lots of other stuff. Approximately a terabyte of normal text in all. There was no “markup” occurring to the text – no separate human-structuring-of-the-data step. From that terabyte of input (which was all stored in 15 terabytes of RAM during game play), the system drew all of it’s “knowledge” when interpreting and answering questions. This page summarizes the process:

IBM Watson Does Not Answer Questions Like Humans

When watching Jeopardy! tonight try to keep in mind that for every question, IBM Watson has to, as a minimum, within 3 seconds:
* Take the stated question and parse its components
* Determine relationships between grammatical elements
* Create items that it must look for or relationships that may expand its search
* Have Hadoop dispatch work to access information that UIMA has intelligently digested and annotated
* 2880 POWER7 cores processing through TBs of data looking for the best set of results
* DeepQA then determining what it considers the best response, and
* Press a mechanical button as do the human contestants and express the answer in English.

Details on the hardware:

Inside System Storage — by Tony Pearson

The system runs 2,880 cores (90 IBM Power 750 servers, four sockets each, eight cores per socket) to achieve 80 [TeraFlops]. TeraFlops is the unit of measure for supercomputers, representing a trillion floating point operations. By comparison, Hans Morvec, principal research scientist at the Robotics Institute of Carnegie Mellon University (CMU) estimates that the [human brain is about 100 TeraFlops]. So, in the three seconds that Watson gets to calculate its response, it would have processed 240 trillion operations.

See also this:

“When Watson is booted up, the 15TB of total RAM are loaded up, and thereafter the DeepQA processing is all done from memory. According to IBM Research, the actual size of the data (analyzed and indexed text, knowledge bases, etc.) used for candidate answer generation and evidence evaluation is under 1 Terabyte (TB). For performance reasons, various subsets of the data are replicated in RAM on different functional groups of cluster nodes. The entire system is self-contained, Watson is NOT going to the internet searching for answers.”

Because you can feed the Watson algorithm normal text data, IBM’s suspects that the technology will let them create useful expert systems in specific domains. So you could take a knowledge base to create a system that is much better at answering call center questions, or medical information to create a system to answer medical questions, etc.

This article points out another medical application:

IBM’s Watson could usher in new era of medicine

Within a year, Siegel hopes that “Dr. Watson” will change all of that. Watson is expected to be able to take a patient’s electronic medical records, digest them, summarize them for the doctor and point out any causes for concern, highlighting anything abnormal and warning about potential drug interactions.

“It offers the potential to usher in a whole new generation of medicine,” Siegel said. “If all Dr. Watson did was allow me to organize electronic medical records and bring to my attention what’s most important and summarize it, that would be incredibly valuable to me.”

The hardware system used to create Watson looks ridiculously massive right now. However, in 20 years that is likely to be a desktop machine that costs $1,000 at Best Buy. That is what past technology trends would indicate anyway. And now that it has been proven possible, the algorithms and techniques will be tweaked and simplified. In much the same way that your desktop machine can now easily beat you at chess using a fairly simple little program, there will come a day in the not-too-distant future where question-answering machines will be much easier to create and deploy.

The Watson documentary:

If you would like to follow Brainstuff on Twitter or Facebook, here are the links:
- Follow Brainstuff on Facebook
- Follow Brainstuff on Twitter