I Made a Super Sauce Created by a Supercomputer

There’s an app to help us lose weight, an app to help us find love, and loads of apps that promise to help us cook.

As a digital designer who lives in the kitchen, I use cooking apps at an appalling rate: almost never. Could there, out there, be an app that truly helps cooks in the kitchen? No, I mean really, not just a glorified portable electronic cookbook or a timer with a sexy UI, but a real-life valuable tool.

By now you’ve probably heard of IBM’s Watson, a human-thinking supercomputer that can parse big data like a boss. Yes, the same one that won Jeopardy! four years ago. IBM has taken the same software and applied it to cooking to develop a beta app with Bon Appétit which was unveiled earlier this month. With a little input from the human cook (say you’re looking for a healthy pasta dish with Italian influence), Chef Watson generates lists of possible ingredient combinations with suggested cooking instructions.

The home screen for Chef Watson with Bon Appétit. Photograph courtesy of IBM Research

The home screen for Chef Watson with Bon Appétit. Photograph courtesy of IBM Research

I know, can a computer, a thing that normally eats ones and zeroes for lunch really come up with something humans might actually want to eat?

RELATED: “Can a Computer Cook?”

I pried myself away from the World Cup long enough to find out. The IBM Watson team came up with a Bengali barbecue sauce recipe and posted it online for anyone curious or delirious enough to try it.

The list of ingredients was intimidating. It contained tamarind, fresh turmeric, and Meyer lemon, the Where’s Waldo of citrus. Four grocery stores in on my quest, Watson had me stumped. I wandered through an Asian grocery store with my mouth agape looking for “coriander leaves.” (What the heck were those? What did they even look like?) I just couldn’t find them. Then, sadly, it donned on me. Cilantro. I was looking for cilantro.

In addition to its particular nomenclature, the recipe is listed entirely by weight, which requires either a scale or some deft conversions. I got to work. I chopped, peeled and weighed the butternut squash, cooked it down, then combined it with the vinegar, wine and spices and let it simmer. After about twenty minutes, I used an immersion blender to combine in the remaining ingredients into a golden, velvety sauce.

I took a spoonful from the saucepan. This was no elementary sauce. It was complex, deep, and delicious. The first note was a unique tang courtesy of the tamarind and vinegar, followed by a complex sweetness from the dates and squash, finished with a swift kick from the chilies. Everything played brilliantly together, and I couldn’t stop eating it.

There was one thing I was not expecting. This was a particularly creative feat for anyone let alone a computer. I wondered how this was at all possible. So, I called IBM. I spoke with Florian Pinel, a software engineer at IBM and one of the founding creators of Chef Watson.

I asked Pinel how the sauce came to be. The team, he said, knew they wanted a barbecue sauce, so they started there.

“First, Watson gave us a list of ingredients commonly used in barbecue sauce, and butternut squash was the one that drew our attention,” said Pinel. “Then it told us what cuisine that butternut squash was used in and in that list was Bengali, so we decided to create a Bengali butternut squash barbecue sauce.”

Watson then considered what actually goes into a barbecue sauce that makes it barbecue sauce, ingredients like vinegar and particular spices. Then it computed all the possible combinations of ingredients, keeping the ones that paired the best with butternut squash, had Bengali characteristics and kept the combinations that were enough to make a complete barbecue sauce. “We got a few thousand results,” said Pinel.

Scientist Florian Pinel (right) stands in front of the IBM Watson Food Truck at the South by Southwest (SXSW) Conference. Photograph courtesy IBM Research

Scientist Florian Pinel (right) stands in front of the IBM Watson Food Truck at the South by Southwest (SXSW) Conference. Photograph courtesy IBM Research

Then, the Chef Watson team worked with the Institute of Culinary Education to develop the recipe by including ingredients they liked, and crossing off ingredients they didn’t, until they arrive at a pared down list of ingredients to form a recipe they liked best.

“To help us in the process, we have three scores: surprise, pleasantness, and chemical pairing based on the flavor pairing hypothesis, that is the more flavor compounds the ingredients share, the more likely they are to go well together. For a given recipe, we can show you what flavor compounds are contained and what ingredients contain those compounds.”

To feed Watson required a lot of data. Initially, Watson pulled information from about 25,000 recipes, ingredients’ molecular data, nutrition, and data about what’s called hedonistic psychophysics, or what we humans find pleasant to eat. For each flavor compound, Watson can predict how well someone will enjoy eating it.

“Taking those recipes, we could find what ingredient combinations were the most frequent, what ingredients were found in what cuisine, and what ingredients you need to make a certain dish. This was all generated using machine learning and natural language processing to parse the text of the recipes.”

Pinel holds a diploma from the Institute of Culinary Education himself. I asked how Watson has changed the way he operates in the kitchen.

“It provides those unexpected ideas. I think as a chef you have those ingredients you like to work with. You have those things you like to make. Watson doesn’t have that bias. So when it comes up with a list of ingredients I wouldn’t typically make, it helps me diversify.”

Pinel is right. When I think of barbecue sauce, I think of red, sweet, and syrupy, but the Bengali sauce was nothing like that. This technology possesses the potential to give chefs a new way of thinking they never had access to before. You could spend years trying different flavor combinations, but Watson can help us break out of some of our ruts.  Hey, if it takes another uninspired pumpkin spice something or other off the menu, I’m all for it.

Photo of butternut squash.

Photograph by Jasmine Wiggins

Bengali Butternut BBQ Sauce
Approximate Yield: 550g

300g butternut squash, diced
200g white wine

100g rice vinegar
50g butter, unsalted
5g tamarind concentrate
40g water
10g chili paste (Sriracha)
4g soy sauce
50g dates, pitted and chopped
2g Thai chili
3g mustard seed
3g turmeric, fresh, thinly sliced
0.4g cardamom, ground
5g coriander leaves
2g Meyer lemon zest, grated
5g salt, to taste
10g Meyer lemon juice
6g molasses

1. Gently sweat the squash in the butter over medium low heat until softened, approximately 5-10 minutes.

2. Add the vinegar, tamarind, water, wine, chili paste, and soy; bring to a simmer and reduce heat to low. Add the dates, chili, mustard seed, turmeric, and cardamom. Continue to simmer and reduce to roughly 250g, for about 20 minutes.

3. Remove from heat; add the coriander leaves and lemon zest. Blend to a very smooth consistency and cool.

4. Season the mixture with salt, lemon juice, and molasses. Chill.

Created by @iceculinary and IBM Research

This story is part of National Geographic’s special eight-month Future of Food series.

Comments