IBM Watson not only succeeded in subjugating its human opponents on Jeopardy but Watson might someday be the motivator for Sonny and the re-programmed robots (http://en.wikipedia.org/wiki/I,_Robot_%28film%29) to move from Chicago, USA to Toronto, CA :-)
All kidding aside, industry pundits have been seriously speculating about alternate uses for IBM Watson. While dramatic extrapolations to HAL9000, SkyNet, and I, Robot abound, less has been said about Watson's immediate usefulness. Rather than to dwell on the Jeopardy “version” of Watson with its ninety IBM POWER7 servers (http://www-03.ibm.com/systems/power/advantages/watson/) clustered together with an aggregate memory size of 15TB, it may be more useful to look at non-game show “versions” . We generally know what Watson had to accomplish within the three second (see previous blog) Jeopardy response time rule. What may be more interesting is to consider classes of problems where response time demands are in the order of minutes, where a 15TB data set is not required, or if a Watson-like construct can aid in narrowing down solutions to sets of possibilities, rather than a single exact response – and let humans decide and execute on a critical choice. Versions of IBM Watson/DeepQA can be architected and have its data digested as a function of the problem being addressed. Alternate, larger data set versions or versions requiring a response in less than a second can be designed. Let's look at some examples from a 50,000 foot level.
Medical and Health Services
A researcher is confronted with a set of patient symptoms never learned in school; some symptoms look familiar to seasoned colleagues, but not all of them at the same time. Is this a new disease, a mutation, something that may have always existed but never categorized in a way that was recognized? Traditional hospital databases can be scanned, results correlated as best can be, but still nothing definite. This is what might take place today, assuming these databases where constructed for more than just billing purposes. A Watson-like derivative could be designed to ingest patient data with specific annotations allowing correlations that would greatly enhance the chance of narrowing down the knowledge required to identify and eventually treat what appears at first be an uncategorized disease. This capability may be vital for health services located in rural areas where a Watson-like system has the proven knowledge of millions of medical experts and studies. Imagine making a query on a surgical procedure and finding out that a technique abandoned twenty years ago has a better chance of being successful than what is used today because of an heretofore unsuspected interaction from combinations of patient symptoms or new hormonal balances resulting from subsets of prescribed modern medicines. Replace bacterial, microbial or ontological diseases, etc., with determining patterns between psychiatric symptoms and effectiveness of classes of past treatments – this is an another variant of a Watson solution. Would Watson completely replace a doctor, probably not, but it can start off as a trusted advisor and the role of a doctor may be changed forever.
Financial and Economic Analysis
Pumping through piles of financial and economic data looking for patterns, uncovering relationships between seemingly related events already consumes vast amounts of computing power in such places as Wall St, London, and Kong Kong. The ability to act on distilled, structured information is generally left to analysts – except for programmed trading. Programmed trading systems react faster than humans to prevailing conditions, but lack the capability to respond to exogenous events outside of its rule-based model. It acts more like a chess program than Watson on Jeopardy. The experts on Wall St cannot possibly program all the rules of the particular game with the hope that combinations of dynamic market and economic data will hit one of them. A system designed to dynamically digest unstructured data (examples including libraries of texts on economics, university lecture notes, radio and TV programs, blogs, etc) create relationships with static data, and purposely distribute this information across processing nodes to minimize redundancy and maximize processing is much more capable of efficiently ascertaining risk than having a rule for every possible combination of known financial and economic data. Having a machine with near instantaneous access to machine learned data from world's leading economists and financial analysts might make a nice companion on the trading floor, considering it appears reluctant to bet the house if it is not confident of a position. Investors Business Daily has an interesting take on Watson-like capabilities (http://www.investors.com/NewsAndAnalysis/Article.aspx?id=562978).
Tech Support and Help Desks
IBM Watson could cannibalize most forms of current consumer technical support. It could it be worse than what goes for telephone-based tech support today.
Law, Patent and Trademarks
Not only could a Watson-like capability minimize filing through existing databases on laws, prior cases, rulings, hearings, opinions, it could also be used as a method of testing witness questions. or suggest a series of inquiries and questions for litigation. It could be used to simulate certain judges, prosecution and defense lawyers, based on prior cases.
A Watson-like system could generate questions for a prospective patent claim based on it's ingestion of the entire patent and trademark database.
National Defense Planning and Intelligence
The amount of structured data and especially dynamic unstructured data that can be associated with military and defense planning is enormous and expanding rapidly. Military decision support systems augmented by a system that has data on all previous military campaigns, past and current international relationships, archives of all international military school generated data including books, theses, lectures, military doctrines, etc. Continuous ingestion of real-time data is added expanding existing relationships. An inquiry that might be forwarded to such a system might include, “What would Sun Tzu do given the immediate crisis?”
Taking just the US. Internal Revenue Service – the most serious problem at the IRS for taxpayers is getting somebody at the agency to answer telephone calls, http://bucks.blogs.nytimes.com/2010/01/07/top-service-problem-at-irs-not-picking-up-the-phone/ Judging from Watson's ability to address very nuanced Jeopardy questions, it could address this problem now.
Other areas include manufacturing, homeland security, local law enforcement agencies, etc
A clear pattern is emerging here. Tasks that traditionally involve humans remembering, making intelligent guesses and informed estimates, even if backed up by filing through mountains of data, could be greatly enhanced or even replaced by Watson and its derivatives.
As with most new technologies, something is gained but something is lost. Most people reading this used to remember important telephone numbers. Today, with perhaps hundreds stored in cell phone, the ability to recall telephone numbers is almost a lost talent. However, nobody seems to be complaining!