IBM's Watson computer learns how to cook
Ask Watson. The IBM computer beat humans at Jeopardy in 2012, a feat that required an artificial intelligence system to understand natural language (the quiz show requires contestants to provide a "question" to an "answer"). As Cosmos reported, Watson prepared by reading a database of 200 million page of text, including all of Wikipedia.
To make Watson even friendlier to humans, IBM has given it the task of writing recipes. This time the computer analysed about 9000 recipes from the US food magazine Bon Appetit. It noted the most popular ingredients and how they were combined in order to understand their underlying logic. Watson then went on to combine this information with its scientific understanding of complementary flavour compounds.
Florian Pinel is the software engineer and cooking school graduate who helped design Chef Watson. He said:
“Most people can reason about pairing two ingredients; there are books written about that, like The Flavor Bible. With a computer, we could look at pairings of six, seven, eight, nine ingredients without a problem. That’s pretty central to the way the algorithms work; they look at the pairings of ingredients. That’s how we built our knowledge base.”
To see if the system worked in real life, Chef Watson was unleashed at the South by Southwest Festival in Austin, Texas last year in an Institute of Culinary Education-run food truck. One of its inventions was a Vietnamese-style kebab made with pork, apples, mushrooms and strawberries. Pinel said the combination worked because the ingredients shared similar flavour compounds.
Another Watson invention is Bengali Butternut pumpkin barbecue sauce, with ingredients including dates, tamarind, tumeric, rice vinegar and Thai chillies.
Watson's cookery book Cognitive Cooking with Chef Watson will be published later this year. Guardian writer Edward Smith recently tried some recipes and concluded:
"IBM's experiment with Watson has real value - they are essentially creating a smart database of food types ... But, in the end, there is nothing quite like experience, a bit of travel and a touch of human intuition."