IBM's Watson supercomputer made history in 2011 by trouncing the two best human Jeopardy players ever. Not content to rest on its laurels, IBM is moving headlong into commercializing Watson technology as it boldly pushes the technology through work with partners to create Watson applications. Healthcare could be a key beneficiary of such efforts.
Is 2014 the year of Watson? It might be.
First, if you haven't yet seen the Jeopardy tournaments in question, do please watch the video below. Watson's speed, knowledge base, and, especially, natural language processing ability will blow you away (made even more amazing given the often riddle-like form of Jeopardy questions).
EDN spoke with Rob High, IBM fellow and Watson CTO, and Kristen Dattoli, global external communications for Watson Solutions, about the Watson technology.
Before getting into any technical details, career opportunities for the engineering community were discussed. The Watson ship may have already sailed, but the flotilla is just getting started.
Indeed, opportunities are rife to work with Watson technology. According to High, IBM is hiring. But you'd better be good if you want to work on the multidisciplinary Watson team. Skills such as linguistics, behavioural science, cognitive science, NLP (natural language processing), grammar, and machine learning are in demand in addition to the usual CS talent.
If your interest lies more in using the technology, IBM is planning for nothing less than a full flowering of the Watson ecosystem. You can write programs that make use of the Watson API, sending user or program generated queries to IBM's Watson cloud, and receiving responses in plain English. In fact, IBM is working with the freelance portal Elance to set up a pool of qualified developers.
In regard to healthcare, High was not overly familiar with the expert systems we've been hearing about for decades. Still, his impression is that doctors have regarded them as "gimmicks" of a sort, doing little to help them in their daily real-world activities. When demonstrated to doctors, Watson, however, was perceived as a truly useful technology.
Most doctors report that they spend less than five hours per month reading journals. With Watson, doctors can access personalized recommendations for patient therapies, which take into account patient and family histories, genetics, current medications, and more, read directly from the patient record. Watson will also compare cases to results in the larger population, and report possible adverse effects.
IBM Power architecture servers make up Watson, which High remarked are unmatched in their combination of raw compute power and massive data I/O rates.
Watson is largely a learning machine, charged with acquiring domain knowledge. At a top level, High explained that there are three elements involved in teaching Watson.
In the medical example, Watson first acquires the core literature and treatment standards. Training then takes place by evaluating patient cases, where Watson must find supporting evidence and rationale for treatments. Once commissioned, Watson continuously acquires new knowledge as standards and clinical experience evolves.
So, is Watson an AI or something unique? High confirmed that Watson research has taken a path away from classical AI.
For example, current AI research hovers around "25 to 35% human language understanding. Where classic AI focuses heavily on ontologies and rules, Watson focuses heavily on linguistics." This allows it to derive understanding from a combination of linguistics and context. High said Watson's design is in "a class of its own," and achieves 75 to 85% understanding -- something eminently clear after watching the Jeopardy contest. In addition, Watson seems to have stimulated interest anew in the AI community.
According to IBM's website, Watson can now execute in a single 16-core server with 256 GB of RAM. That's a far cry from the 2,900-core, 15TB machine that played Jeopardy. What's changed?
Watson's architecture has been re-engineered so that the entire processing pipeline can run on this one server. The system scales for more users by adding servers. Despite this, IBM doesn't have plans to sell Watson systems to customers, preferring to treat the technology as a cloud service. That just may be a good idea. Imagine the tech support calls: "I think my Watson is depressed... It's been giving bad answers all morning."
As EDN's chat with the IBM Watson experts approached its too-soon conclusion, it became clear that we're on the threshold of something big and new in computing. High envisions a healthy ecosystem of applications and developers leveraging "cognitive systems" -- Watson or otherwise -- birthing applications that can answer questions in an ever-widening field of domains.
Watson right now can only respond based on its knowledge base, but it may soon understand questions of "speculation and conjecture" -- what-if questions, if you will, a phrase made infamous by a certain killer app ca. 1980. High also imagines these systems becoming more responsive to human input; for example, with multisensory processing that can react to body language, facial expressions, speech inflection, and so on.
And when Watson can process images, it will allow both a deeper degree of training completeness, as well as greater utility in the field, such as being able to comprehend X-rays, MRIs, EKG signals, and more.
By this time next year, don't be surprised if there's already a spate of applications out there ranging from life-saving to frivolous. IBM is planning nothing less than to change the computing landscape with cognitive systems like Watson. Whether it's behaving as a powerful research and analytical assistant for your doctor, or acting as your personal shopping adviser, Watson may be talking to you from your computer or smartphone in the very near future.
Equally exciting for engineers are the opportunities to get involved. Even if you're not part of the lucky few able to work directly on Watson technology, you can join the elite corps designing Watson-based applications. All you need is an idea. Me, I'm looking forward to iWatson on my iPhone.
This article is part of EDN's Hot Technologies: Looking ahead to 2014 feature, where EDN editors examine some of the hot trends and technologies in 2013 that promise to shape technology news in 2014 and beyond.