There’s been a lot of talk about artificial intelligence’s potential to change the food industry, from a digital smell and taste tester to an emotionally savvy system that asks people about their dining experience. However, much of the hype has resulted in little more than pie-in-the-sky ideas that are either prohibitively expensive or occupy a niche area.
But a partnership between IBM and McCormick may have cracked the code by pioneering the use of AI in food development – a project that has been four years in the making.
The product-creating platform was developed through IBM’s experience in AI and machine learning and McCormick’s 40 years of sensory science and taste data, which included past formulas and millions of data points related to consumer taste preferences and palates.
It works by building algorithms that generate new recipe suggestions and takes just minutes to create dozens of formulations for both protein and vegetable-based meals.
The use of this tech isn’t just talk either. McCormick expects to launch its first AI-enabled products into US retail by late spring, comprising a set of one-dish recipe mix flavours like Tuscan chicken, bourbon pork tenderloin and New Orleans sausage.
McCormick has plans to scale its technology by 2021 to create a global lab across its 20 sites in 14 countries.It claims this is a new era of innovation and wants the technology to usher in the R&D of the 21st century.
“We are witnessing history,” McCormick’s chief science officer, Dr Hamed Faridi, told Food Spark’s sister site Food Navigator.
“We wanted to create a learning system that can learn from the experiences of everybody around the world and give feedback to everybody around the world. There is no way we could do this... without a learning machine that can look at the data, give feedback, learn from it, improve it and come back with a new suggestion.”
Dr Faridi added that the company had decades of documented sensory and taste data and were early adopters of sensory methodology, but the information always remained in silos. “We also have very strong roots in chef and culinary [development]. We wanted all these things to come together in pursuit of what is the next flavour.”
The technology is not only a taste adviser, but also a time saver. The platform could help speed up innovation by three times, allowing the company to streamline the product development process and launch into markets at a faster rate to respond to consumer trends.
”A developer always starts with a base formula and builds on it. Then sensory will test it and come back with suggestions,”Dr Faridi explained.“Somewhere between 50 and 150 iterations is the norm. We have found with this that we can cut the process by two-thirds. We can make it much faster. I think two-thirds is conservative: four or five years down the road will be much better.”
While the richness of McCormick’s data was fundamental to the success of using AI in this setting, Dr Robin Lougee, research scientist at IBM, said her team was also interested in how the technology could aid in creativity.
“Product developers sometimes have biases that the machine does not. We want to be able to help the products developers explore their set ideas to come up with the next iconic products,”she commented.
“Thanks to the wealth of data that McCormick has, the computer can look across a pool of data that the human would be unable to read and reason over. The AI system can find things about product developers may not have thought of... It can take you outside your biases and comfort zones to look at things you might not have realised.”
Exceeding human limitations is also something Dr Faridi believes is possible for the new AI system.
“When we develop any product, we have a hedonic target of 6.5 or 7. So the developer – when they get the sensory evaluation to 7 – accepts that it is ready to go full commercialisation. The machine has no boundary on this. It’s always looking for making it better and making suggestions,” he said.
The issue of high product churn would be addressed as well, Dr Faridi added, solving one of the biggest challenges facing NPD: how to increase staying power.