Megabytes: supercomputer writes first cookbook
After culinary "training", IBM's Watson pens a recipe book
Touting such eyebrow-raising combinations as an Indian turmeric paella, a Turkish-Korean anchovy Caesar salad and a Creole shrimp-lamb dumpling, a new contender for hottest cookery writer of 2015 is preparing to elbow aside the likes of Jamie Oliver, Mary Berry and Nigella Lawson: IBM’s supercomputer Watson.
Four years after beating two human champions to win the US game show Jeopardy! IBM’s cognitive computing system is set to release its first cookbook - with the help of a few human specialists. Out in April from independent American publisher Sourcebooks, Cognitive Cooking with Chef Watson uses the supercomputer to generate exotic ingredient combinations from the “trillions” of potential groupings out there, with chefs from the Institute of Culinary Education then designing recipes based on Watson’s suggestions. The book will feature, promises its publisher, “unusual ingredient combinations that man alone might never imagine”.
“I think it’s fascinatingly different, that’s really one of the big reasons we chose to publish,” said Todd Stocke, editorial director at Sourcebooks. “Take a craft – an artform, really – like cooking, apply hardcore science. It makes for genuinely unique results with surprising and smart reasons behind them.”
IBM’s experts began by “training” Watson - which is named after IBM founder Thomas Watson - on tens of thousands of recipes, and on the chemical composition of foods, moving on to the flavours and ingredients which complement each other - such as rosemary and potatoes, or olives and gin. The computer then used three different metrics to analyse ingredients, rating them for surprise (suggesting ingredients which are rarely found together), pleasantness (“researchers have carried out studies on the flavours that give people pleasure at a molecular level,” says the book), and synergy (“studies indicate that foods sharing common chemical flavour compounds taste good together”).
“The system essentially worked as follows: the cook enters some basic elements that serve as a foundation for a dish, which Watson then processes using an extensive database of recipes, cultural studies, and chemical flavour composition, all of which results in an ‘output’ of ingredient lists that Watson finds interesting on scales of pleasantness, surprise, and flavour pairing,” said Michael Laiskonis, Institute of Culinary Education creative director.
“The cook then selected one of these lists of ingredients with which to create a dish. The system relies on human prompts and interaction, which has lent an interesting personal aspect to each of the dishes.”
So for one recipe, Watson suggested mushroom, strawberry, chicken and pineapple, four choices which “initially seemed incompatible”, writes Laiskonis in the book, but which all “share significant levels of the flavour compound g-dodecalactone”. He came up with the Vietnamese apple kebab, a “Vietnamese-inspired dish without the use of familiar flavours like fish sauce, chilli peppers, lemongrass, garlic and cilantro [coriander]”.
The Indian turmeric paella, meanwhile, “brings simple Indian flavours to a classic paella preparation”, while the Turkish-Korean Caesar salad adds traditional Korean ingredient doenjang, aubergine and artichoke to produce a unique take on the dish.
Chef James Briscione writes that he was surprised when Watson listed the ingredients which he would turn into the Greole shrimp-lamb dumpling. “Familiar ingredients like okra and filé powder make me happy, but putting lamb and shrimp together in the same dish was something I had never imagined doing, in any context,” he writes. “The flavours play surprisingly well together, and the result is far greater than the sum of its parts.”
There are more than 65 recipes in the book, also including the Spanish almond crescent - the first ever ingredients proposal from the computer, in early 2013, of pepper, saffron, honey, yeast, egg, coconut milk, cocoa, lemon extract, almond, pastry flour, heavy cream and oil - and the alarming-sounding hoof-and-honey ale. “The addition of veal stock makes this drink a one-of-a-kind creation,” writes chef Anthony Caporale. “Blending the time-tested combination of beer, beef and Burgundy wine with the sweetness of peaches and honey, it’s at once surprising and familiar.”
IBM hopes the cookbook will “help people discover the potential of cognitive computing systems”, showing how “Chef Watson” is able to understand vast amounts of data, and learn from interactions with people and information, “to help us discover brand-new knowledge and augment human creativity”. The project, said Florian Pinel, senior technical staff member at IBM Watson Group, shows “the possibilities of human and machine collaboration”.
“Through Watson, IBM demonstrated how a computer could understand and reason across vast amounts of data to provide experts with evidence-based answers. As part of our continued research we wanted to know if Watson could help us come up with brand new knowledge and answers,” he writes in a piece for IBM. “Going from making inferences about the world to generating new ideas is what led us to begin collaborating with professional chefs at ICE, to develop strategies to implement computational creativity in the culinary arts, and we are now using these lessons for other domains where innovation, discovery and design are necessary.”