A dash of flavor, a pinch of pepper, and a spoonful of AI
The secret to great flavor? It might just be AI.
McCormick & Company
“I’ve been at IBM for 20 years and worked on a lot of interesting things,” Richard Goodwin told Industrious from his Yorktown Heights office. “The coolest thing so far is this project.”
A principal researcher with a PhD in artificial intelligence who once worked on a mobile robot project funded by NASA in grad school, today he’s focused on the convergence of AI and creativity in a new arena: flavor.
By his own account, it’s a challenging domain, because the science of how humans experience flavor is not well understood.
Computers can’t taste or smell, but they can work with humans to apply data and insights to flavor creation for food products—which is exactly what Dr. Goodwin and his team have done with global flavor company McCormick.
Dr. Hamed Faridi has served as McCormick & Co.’s Chief Science Officer for more than two decades. He leads a global lab of 500 scientists and developers who work on a broad range of products; clients include both consumers and manufacturers.
A few years back, he was looking for a technology partner.
“When you look at 500 scientists and developers, they have different levels of skill sets and backgrounds,” he said. “One of my first objectives was: what if I could make each of those 500 as good as the best one?”
Dr. Faridi also wanted to tap the collective wisdom of the 500 and make it available to all. Finally, he wanted the process of product development to be faster and better.
IBM and McCormick’s collaboration was officially announced this February. Combining IBM Research’s AI expertise with McCormick’s deep sensory science and taste data (including decades of past product formulas and millions of data points on consumer taste preferences and palettes), its first AI-enabled platform “ONE” launches later this year.
“Creating food products is much more challenging than just creating a recipe at home,” Goodwin said. Amounts need to be exact, as do the ingredients.
More than 400 types of garlic exist, he explained, and every kind of variation—from country of origin to powdered grain size—can make a difference. Finding the right one can be a challenge.
“You may want multiple garlics,” he said. “One for taste, one for appearance. Maybe even different sizes. One type may hit your tongue quickly, another will give you the back flavor. It’s a very complex task.”
Flavorists and product developers have created hundreds of thousands of formulas across the decades. For a human to read through them all is a near impossibility—but not for a computer. That’s where AI comes in, searching for patterns that led to previous flavor success, and suggesting different flavor combinations that a human may not.
“People have their own style,” Goodwin said. “What they’ve been successful with in the past tends to be what they lean on in the future.”
Computers don’t have that same bias.
“If I’m a developer, I may already have a favorite flavor,” Faridi said. “If I’m developing a pizza seasoning, I’m thinking Italian, Mediterranean flavors. I wouldn’t think about adding cumin.”
But the system just might, proposing pairings a human may not because of past experiences. “The system makes a proposal and sometimes it doesn’t work. But sometimes it does.”
And since the system is a learning machine, it incorporates feedback from developers into its next proposals.
We learn five things from the system data, according to Goodwin: pairings of flavors; functional substitutes; a definition of what it means to be a particular product (barbecue sauce vs. marinade); flavor space (a sweet dessert shouldn’t get a yellow onion); and how to predict success based on various factors.
Speed, too, is of the essence, and where the technology can help. A significant portion of product development in the flavor solution world is responding to competitive requests. Being first to respond with an innovative product or flavor is a strong competitive advantage, Goodwin said.
“One of the challenges people at my level have is how to help companies develop products that have greater staying power in the market,” Faridi said. Some products may stick around for just a few years, while others—like Oreos, introduced in 1912—have obvious staying power.
“Finding the next Oreo can be like finding a needle in a haystack,” Faridi says. “Finding a formula with very broad appeal is a challenge.”
He hopes this new platform can help his team discover new icons.
“We want to be at the forefront of flavor,” Faridi said, which is especially important in today’s highly segmented and varied market. Plus, younger generations want to experiment, he’s found. That, combined with e-commerce, is another great fit for the new platform.
“We can rapidly screen the data, analyze it, and give new suggestions,” he said. He’s excited for the future possibilities, too: “We’ll be in the discovery phase for several years. And what we’re seeing today is just the beginning.”