Earlier this summer, we introduced you to Chef Watson, the cool new "cognitive cooking" technology that IBM is developing in partnership with Bon Appétit. Using the system's capabilities—built partly on our database of 9,000 recipes and partly on Watson's understanding of flavor combinations—the BA test kitchen created a surprising July 4th–themed menu: fennel-spiced short ribs, a grilled corn salad, and an herbaceous berry cobbler. (For an in-depth look at how this all actually goes down, see here.) And that's when the challenges began.
First, we asked you, our readers, to invent a potato salad using Watson's ingredient ideas (here's the winning recipe), and then, once IBM opened up Chef Watson to beta-testing by a select (and ever-widening) group of users, we dared them to develop a dish starring the king of summer produce—the tomato, obvs—in whatever form they liked. With one exception: It couldn't be a salad.
Now, after a week or two of testing (and refining, and retesting) by our senior food editor, Dawn Perry, we have some winners to announce: a light gazpacho, a savory tart, and an ambitious seared scallop dish with a flavorful tomato water. And perhaps more so than in previous Chef Watson experiments, we learned what the system is capable of, where human hands (and brains) remain absolutely essential, and how Watson can continue to evolve in the future. In a minute, the winners—but first, the lessons:
Appearances Matter
Tomatoes and avocado are a brilliant combination—creamy, sweet, acidic, and fresh. Add in a little spice from a chopped jalapeño, and you'll knocked it out of the park. But throw it all into a blender and you get a murky-looking soup. Perry tested a Watson-bred gazpacho recipe that called for puréeing all of the above, plus some fresh dill. "It was muddy," she explains. The flavors were spot-on, but with so much focus on creating awesome flavor combinations, the app wasn't really equipped to think visually. As a cook, Perry knows that blending avocado with tomato will result in an unappetizing color. She might instead choose to dice a ripe avocado and add it to the finished gazpacho for a similar flavor profile with much more aesthetically appealing results.
Creative Doesn't Always Mean Tasty
Chef Watson puts together flavor combinations based on two variables: a "pleasantness" and a "surprise" factor. The pleasantness of a dish is determined by similar and complementary flavor compounds, and the surprise element is all about putting together ingredients that, although we don't often pair them, go well together—scientifically speaking, at least. While Watson's suggestions are usually totally on-point, users should consider how the finished dish will taste. Case-in-point: Perry tested a Watson-generated recipe that called for red curry paste, coconut milk, basil, and…mozzarella cheese. "Coconut milk and mozzarella are not happy together," Perry said, adding that she tried the recipe first, just to confirm her suspicions. Remove the mozz and voilà—it's a masterpiece. Lesson learned: a little common sense goes a long way.
Every Recipe Needs an Editor
The flavors in each recipe had Perry excited to test things out, but some of the ingredient proportions definitely needed editing, a challenge that some of the beta testers skillfully navigated. A delicate tomato tart, for example, called for three-quarters of a teaspoon of smoked paprika, which was just enough enough to overwhelm the other flavors. Another kink that provided a little good-natured humor was the app's at-time vague language—which is definitely improving with time. One recipe submitted called for trimming the "white bulbs" from cherry tomatoes before cooking. Uh, white bulbs? "Exactly!" says Perry, laughing. She snipped off their green tops and called it good. In time, we all expect, Watson will learn better how to handle each ingredient, whether by explicit programming ("note to Watson: no white bulbs on cherry tomatoes") or by a more advanced, and so maybe more complicated, approach to machine learning (i.e., Watson figures it out on its own).
It's All About User Experience
So what does this all mean? For Perry, it was a reminder that although Watson's recipes are created by a computer, they're implemented by humans. Perry admits that she prefers simply to look at the suggested ingredient combinations, determine the proportions and methodology herself. "But then, I don't like being told what to do," she says, acknowledging that for someone who doesn't spend every day in the kitchen, the suggested steps definitely prove helpful. And, not for nothing, Chef Watson had a few great takeaway tricks that Perry will definitely be using in the future. In lieu of a fussy dough, the tomato tart recipe called for a frozen puff pastry sheet, which is way faster than making a dough.
So are we ready to let a computer take over our kitchens, à la The Jetsons? No, but thankfully that's not the point. As Perry says, "It's all about inspiration."
Want to make Watson's winning tomato recipes at home? Check them out Below:
Best Real-Life Application ____
