Here’s a taste of what IBM’s Watson supercomputer now can do: it can devise a recipe for a Swiss-Thai fusion asparagus quiche, or a cayenne-infused papaya custard with orange juice reduction.  

During a day of demonstrations and talks Wednesday about the future of cognitive computing at IBM’s T.J. Watson Research Center in Yorktown Heights, New York, the computational feat of building recipes in response to custom requests–and drawing on recipe databases and knowledge of human taste and smell preferences to do so–was clearly a light moment. But it also spoke to the challenges the company faces in finding killer applications for the question-and-answer technology.

IBM’s Watson made big headlines two and a half years ago when it won the Jeopardy game show against two human competitors.  That was the culmination of a long effort at customizing the system to amass the right databases, process language, and act fast in a question-answer format.  Since then, IBM Research has put considerable effort behind translating this feat into practical products, such as decision-support tools for overwhelmed medical and financial workers (see “Watson Goes to the Hospital”).  The company has also created a way for the technology to be used as a customer service agent (see “Trained on Jeopardy, Watson is Headed for Your Pocket”).

I had a chance to talk with Doug Johnston, a cardiologist at the Cleveland Clinic, one of several institutions working with IBM to hone the technology for medical applications.   What’s already apparent is that Watson can be very good at doing things like boiling down medical literature to aid medical students. What’s tougher for Watson to digest are the narratives written at the point of care by clinicians. These chunks of text are written in different styles, and often contain incorrect or incomplete information.

Watson and other analytic technologies will get better if such records are formatted in clearer ways–with distinct fields for patient symptoms, actions taken, and outcomes, he said.  With this in mind, IBM has been trying to customize business software to be Watson-ready (see “Watson’s New Job: IBM Salesman”).  A larger point was articulated by Thomas Malone, director of the MIT Center for Collective Intelligence. The future, he said, lies in building systems that can best leverage the capabilities of humans and computers.  A growing body of research is finding that answers gleaned from a combination of humans and computers are more accurate than those generated by either group alone, he said.

 Jeopardy  “was a game, and we did an incredible demo of computer and human ingenuity. But in the end, it was a game.  This is no longer a game – this is serious,” said John Kelly, the head of IBM Research.  IBM today also announced a new research collaboration in cognitive computing with MIT, RPI, New York University, and Carnegie Mellon University aimed at  ”driv[ing] this technology forward,” Kelly said.

Then there was the quiche. Kelly explained that IBM was exploring how Watson might help the pharmaceutical and chemical industries come up with new compounds. As a demonstration, his team tested Watson in the kitchen. To come up with new recipies, it “takes all of the world’s known recipies, breaks them down into ingredients, breaks those down into the chemical constituents,” he said.  Then they threw in a dash of data about human preferences for various compounds that provide taste and smell.

To come up with new recipies, you ask Waston questions that might specify country names and preferences, such as dietary restrictions.  In the case of the quiche, asking for dishes merging Swiss and Thai influences led Watson to offer a dish that included ingredients such as pimento, tamarind, galangal, basil, gruyere cheese and mint.  IBM offered samples from “Chef Watson,” abd I found the papaya-cayenne-orange custard particularly good.

While the recipe technology is just a research prototype right now, it provided one possible taste of cognitive computing’s future–providing cool commercial services we didn’t know we needed.