IBM’s Supercomputer Has Implications for Healthcare, Information Tech and More
Wouldn’t it be nice to have your very own supercomputer in your pocket?
If your laptop crashed while you were working on a major presentation, you could ask your portable expert to help diagnose the problem. If you wanted to bone up on Middle Eastern history, you could ask it to comb every document available and then wrap it all up in a simple summary (annotated, of course).
Best of all, instead of typing out basic questions on a cramped keyboard, you could speak to it in natural human language and it would understand.
Thanks to the researchers behind Watson – IBM’s whiz-bang computer designed to compete on ‘Jeopardy!’ – that sci-fi-like scenario is a little bit closer to reality.
In less than a week, the world will watch as Watson takes on the top human contestants on the popular trivia gameshow. But artificial intelligence experts say it may not be too long before the technology powering Watson spills over into doctors’ offices, businesses and, eventually, maybe even your phone.
For the past four years, researchers at IBM have been grooming their computer program (named after IBM founder Thomas J. Watson) to compete on ‘Jeopardy!’ against the game’s most succsessful champions, Ken Jennings and Brad Rutter.
Watson Understands Questions in Natural Language
Not only does the program need to be able to recall facts and figures across a wide range of topics, it needs to understand the puns and metaphors commonly found in ‘Jeopardy!’ clues. And, to win, it needs to do it all faster than the fastest human.
It’s an information-seeking tool that’s capable of understanding your question to make sure you get what you want, and then delivers that content through a natural flowing dialogue, David Ferrucci, IBM’s Watson team leader, said on the NOVA special ‘Smartest Machine on Earth,’ which premieres on PBS tonight. I don’t think the world has seen a machine quite like Watson and, frankly, I’m thinking where can we go from here?
The magic of Watson is that beyond being able to search formal databases and tables for information based on keywords, it uses many different algorithms to understand and process natural human language.
In Medicine, Watson-Like System Could Be ‘Doctor’s Assistant’
Classic expert systems tend to be brittle, they tend to be very narrow, they tend to be able to solve problems only in the way it was expressed in that formal math, Ferrucci told ABCNews.com. What you see underneath the hood in Watson is a way to do that reasoning but right over the natural language content itself.
That might not sound so tricky at first, but think about all the nuances and wordplay woven into human language. Homonyms, inflection, double entendres – they might be familiar to you, but they’re foreign to even the most advanced computers.
Dealing with natural language is a very, very hard task for a computer, Ferrucci said. But then, moreover, there’s tremendous potential if we can continue to chip away at this task.
In medicine, for example, a Watson-like system could serve as a kind of doctor’s assistant, he said.
Let’s say a patient suffers from a rare disease, the medical Watson could listen in on patient interviews and combine those conversations with the patient’s medical record, family history and test results. Not only that, but it could cross-reference that material against relevant journal articles, research and other published information.
Applications of Watson Technology Could Extend to Government, Engineering
Finally, it could generate a list of the top conditions the patient might be suffering from, along with a list of all the relevant sources.
What the doctor can do is just consult this and say, am I missing anything? There’s a huge amount of information out there, Ferrucci said. I imagine that it will help the doctor to adjust and refine and rationalize and document both the diagnostic process as well as the treatment process with a lot more confidence.
Watson-like systems may not be as precise as a rule-based system working over a specific database, he said, but they can cover a wider collection of information and then make it more digestible for humans.
Ferrucci said the same process could be used by information technology professionals to unravel complicated computer problems.
IBM’s immediate challenge might be to best world-champions on ‘Jeopardy!,’ but, ultimately, the company is looking to apply Watson’s technology to areas as varied as government, engineering and business.
‘Jeopardy!’ Challenge Could Spark Public Conversation, Drive Research
Eric Nyberg, a professor in the Language Technologies Institute at Carnegie Mellon University, said he hopes the ‘Jeopardy!’ competition will not only open up a conversation with the public about artificial intelligence, but also drive more research in the field.
As far as consumer applications, he said, “I think the logical next stop beyond Watson is going to be systems that can advise you on selecting certain kinds of products that meet your personal needs.”
For example, we may not be too far away from a system that could read through all the camera reviews available and then, based on its knowledge of a user’s preferences, recommend the best choices, Nyberg said.
The system could be accessed in a retail shop where you would buy the camera, but it could also be accessible through a cell phone, he said.
“We could build applications like that today. For example, if there was a manufacturer that wanted to create a version of Watson that could answer questions about its entire product line that would be a very easy thing to do,” he said, adding that the range of trivia and language used in ‘Jeopardy!’ actually poses a more difficult problem.
But though Watson may represent a ground-breaking step in so-called question-answering systems, researchers say it’s still not the ultimate goal in artifical intelligence.
Goal of AI: Build Machines With Human Intelligence
“From the science point of view, the goal of artificial intelligence, when it started 50 years ago was to build machines that exhibit human intelligence,” said Boris Katz, the principal research scientist at MIT’s Computer Science and Artificial Intelligence Laboratory. “One could argue that answering certain questions is part of that but I think it would be especially interesting to build something that not only performs certain tasks but maybe even does it in a way that a human would do it.”
Those machines would not just provide answers to questions, but be able to explain how they arrived at the correct answer. And, given the trend toward mobile devices, he said, eventually, those machines will likely find their way into your hands.
Some smartphone applications can already understand limited voice commands and execute basic tasks like dialing contacts in your phone book but, Katz said, advanced systems could potentially turn your computer into a hand-held buddy.
“Your pocket friend – more than Watson,”he said. “Not only [to] answer simple questions, but actually do things. – [You could instruct it to] “Please tell my friend to do this,” “please find this information and summarize it.”
That scenario is probably still decades away, he said, but it starts with the natural language research that put Watson on ‘Jeopardy!’.
So while you might want to be loyal and root for the human race next week, the future may not be such a bad consolation prize if man ultimately does get defeated by machine.