Jeopardy-winning supercomputer available for public use

Watson, the IBM supercomputer that famously beat champion players on Jeopardy, is to be made available to public developers for remote access.

IBM used Jeopardy as a test of the supercomputer’s parallel processing abilities. The gameshow, which supplies players with an answer and asks them to supply the question, was considered a particular challenge for a computer. That’s because it doesn’t simply involve recall of knowledge, but also parsing of natural language to discover its most likely meaning.

In a televised contest, Watson trounced former champs Ken Jenning and Brad Rutter, with only two real hiccups. One was when the human players were confident enough in a topic to buzz in first and then try to think of the answer in the few seconds before they had to give a response.

Another was when Watson mistakenly suggested Toronto was a US city. That seems to have been caused mainly by it having a lot of data about the Toronto Blue Jays, a team that plays alongside 29 US counterparts.

Since then IBM has looked into several ways Watson could be used for more practical purposes. It’s already been used for projects such as improving automated customer help systems, figuring out the best options for hospitals trying to buy medicines and equipment, and helping tailor lifestyle and health advice to individuals.

Other uses that have been suggested include figuring out possible medical conditions from a list of symptoms, and diagnosing technical problems with computer, again from (often vague) symptoms.

Now Watson will be opened up as a “development platform in the Cloud”, the idea being not only to help other organizations, but to get more ideas about how to make the most of being able to carry out human-like reasoning at the speed of a supercomputer.

As well as being able to run applications on Watson, developers will be able to access an online marketplace for data and resources. Accessing Watson itself won’t have an upfront fee: instead developer will pay a usage-based fee as and when they use Watson’s processing in their apps, along with a fee for getting help from IBM staff.

According to IBM, the API for using Watson will be relatively simple and developers won’t need to have a working knowledge of how machine learning works. They will, however, need to provide suitable “training data” to be able to go and use Watson for real.