Don’t get me wrong. There was absolutely nothing elementary in IBM’s phenomenal work on Watson. The public debut of the machine (actually the real “magic” was in the software, rather than the hardware), was a triumph in a world that had been claiming, as far back as the 1980s, that “artificial intelligence” was just around the corner.
Still, there is indeed something about Watson that is clearly elementary: something that should give us great hope for the future—both Watson’s and ours.
The “Jeopardy Challenge” , in which IBM’s “Watson” computer handily defeated the two highest winning players in Jeopardy history, was only the latest in a series of Grand Challenges, in which IBM pushed the envelope of computer science to perform tasks that were previously considered beyond the realm of computers—the use of IBM’s Deep Blue in beating the world chess champion, Blue Gene’s role in decoding the human genome and even IBM’s role in enabling the U.S. the land a man on the moon.
Watson, however, went an order of magnitude beyond these previous triumphs of computer power. While the computer’s encyclopedic database and computational power certainly enabled its success, these capabilities were already available on off-the-shelf IBM hardware (2,800 cores and 15 TB of memory in 90 of its Power 750 servers and 20 TB of disk storage linked in a cluster).
Its real accomplishment was in its ability to interpret not just natural language, but the types of puns, metaphors and idioms that have come to characterize Jeopardy. This was enabled by a combination of off-the-shelf hardware and especially the secret sauce embedded in the Jeopardy-specific algorithms over which IBM researchers wrote, tested and tweaked over the last three years. And don’t forget the confidence rating and wagering algorithms which, while resulting in numbers that may have sounded strange to humans, were based on calculates of the odds for all types of contingencies.
Will the Real Watson Please Stand Up
Watson was certainly not perfect in its victory. In the first night’s contest, Watson modestly bested the score of one of its human competitors, and only tied that of the other. Night two, in the first round of Double Jeopardy, things got downright scary, with Watson being the first to buzz for, and correctly answer 24 of the 30 total questions. Watching the frustration of the helpless humans, one could be forgiven for thinking of 2001: A Space Odyssey’s HAL.
Then, with its blunder on its first round Final Jeopardy (Did Toronto recently secede from Canada and join the U.S.?) and its “merely human-level” performance (although it did reach a number of correct answers, but not in time to beat the other contestants) in round two, I got really scared. I began asking myself whether Watson consciously “backed off”, avoiding running up the score, either out of empathy for its flailing competitors, or out of fear that a machine that so dominated humans would be feared and shunned by society. While Watson did end up winning the three-night competition, the ultimate outcome wasn’t really determined until the last Daily Double, and the wager (that ensured it could not loose) that it made on the last Final Jeopardy question.
Why did I find this so frightening? Because I, who have been in the IT industry for more than 30 years, actually began to attribute human feelings to a hunk of silicon!
It is Indeed, Elementary
But I digress. As I discussed above, there was absolutely nothing in IBM or Watson’s Jeopardy performance that was “elementary.” It was, by any account, a stellar achievement.
So, what was so elementary about Watson’s triumph? The comparison of its success in winning a television game show, to:
- The enormous challenges that civilization faces (and, not coincidentally, that IBM is attempting to address with its Smarter Planet initiatives); and
- The contributions that Watson technology and learnings have the potential of making to addressing these challenges.
First, let’s recognize—Watson is a room size machine, residing in a specially designed and extensively cooled data center and that even its off-the-shelf components (without even accounting for the cost of developing the algorithms that were so fundamental to its success) cost hundreds of thousands of dollars. But, as Computer Intelligence guru Ray Kurzweil explained in his February 17 Wall Street Journal editorial, at the current rate of computer price-performance advances, Watson’s power is likely to fit within single server in about seven years and within a PC in a decade.
Just as importantly, a “real-life” system would not have to contain the sum total of world knowledge. These systems will be:
- Tailored to the needs of a specific discipline (such as medicine or finance) or the needs of a specific company;
- Will have access to the Internet, third-party search tools and external databases, rather than having to operate as a self-contained unit; and
- Will not be required to devise answers that meet its minimum confidence levels within the three seconds that are required for Jeopardy.
Watson-like capabilities, will, in other words, be available to the public (or at least some segments of the public) within the next couple years. Meanwhile, IBM has already partnered with Nuance Communications to bring speech recognition capabilities to Watson (initially, specifically for healthcare).
Watson’s Next Careers
After Watson’s first (albeit brief) stint as a television star, it is ready to explore more “mundane” careers. But what are these careers likely to be?
While the Star Trek computer was a model for at least some of IBM’s researchers, most of Watson’s opportunities will be much more down-to-earth. Many are based on the coupling of Watson’s “Deep Question Answering” technology and deep analytics in decision support applications. Possibilities—or indeed, probabilities—may include:
- Customer Service, which could improve service time and quality while simultaneously disrupting a business model in which so many call center jobs have moved to low-cost countries;
- Financial Analysis, such as in the combing of huge quantities of structured data and unstructured information to identify likely acquisition targets;
- Travel, such as in a new-generation navigation system in which drivers can ask for best ways of avoiding traffic, or more interestingly, to suggesting routes from X to Y that take one past attractions that best meet your profile, such as museums, restaurants or wineries that make 90+ point wines; and
- C-suite assistant, to identify and assess business trends, evaluate a broad range of contingencies or running what-if analyses, such as the impact different product and advertising mixes may have on revenue and profitability.
This leads to what is probably the most important and imminent of applications for Watson Technology—its use in health care. Although the potential applications are numerous, the first and highest-impact application is likely to be in diagnostics, such as where a doctor can input lists of symptoms, medical histories, and a broad range of other relevant information to identify possible illnesses.
Better yet, it could be used to review individual electronic medical records to identify symptoms that a doctor may miss or large volumes of electronic records to identify linkages that have not previously been discovered. Longer term, it could even be used to bring first-line diagnostics to remote, emerging country villages that do not have access to doctors, such as by allowing nurses or technologists to input systems into a computer, to a remote Watson-based diagnostic system.
Many potential applications, as in health care or engineering, could face big legal questions. What if Watson made a mistake in diagnosing an illness or in calculating tolerances for a bridge? Or what if Watson correctly suggested an option, which was dismissed by the doctor or engineer? Or have we taken the first step into the science fiction era, where computers may obviate the need for humans in even some of the most demanding of professions?
While the answers to such questions will have to wait, the application of Watson technology to these challenges will not. The day after Watson’s Jeopardy victory, Columbia University Medical Center and the University of Maryland Medical School announced a plan to work with IBM on health care analytics research, with a goal of launching a commercial diagnostic and treatment offering over the next 18-24 months.
We will have to wait to see whether Watson will be as successful in its future careers as it was in its first. My guess, however, is that Watson’s descendants will have as great an impact on society, business and the nature of knowledge work, as the Internet.