“IBM’s AI-Driven Taste Technology: Revolutionizing Flavor Identification and Product Development”
IBM gained fame when its artificial intelligence supercomputer Watson triumphed over human contestants on the quiz show Jeopardy! more than a decade ago. Now, this 111-year-old tech giant is tackling a new challenge: Can it enable computers to have a sense of taste? Four years ago, IBM researchers began exploring this idea, starting with a tasteless liquid that is often difficult for humans to differentiate: water. After the computer learned to recognize various water samples with slight compositional differences due to mineral content, it competed against human tasters to see who could better identify previously encountered samples. The computer emerged victorious. “The AI system outperformed our human tasters in distinguishing four different types of mineral water,” stated Patrick Ruch, the lead researcher on IBM’s AI-driven e-tongue technology known as Hypertaste.
This circular device conducts its “taste test” by generating a digitized chemical profile of the liquid it samples. This “fingerprint” is then compared against a database of other liquids using artificial intelligence — a process that takes under a minute — to find a match. While Hypertaste is still a few years away from mainstream commercial application, IBM is collaborating with industry partners to explore various uses, including working with food and beverage companies to identify and predict flavors, and to quickly recognize coffee, soft drinks, and other products that appeal to consumers.
Ruch emphasized that Hypertaste is not intended to replace human experts but to alleviate some of the more tedious tasks, such as repeated product taste testing, ensuring consistent quality across batches. It can also help verify authenticity, such as detecting counterfeit wines or whiskies, and assist in product innovation by uncovering new flavor combinations. Future possibilities for this technology might include detecting contaminants in drinking water, tracking raw materials through the supply chain, and identifying foodborne illnesses. “The more users engage with it and contribute data, the more effective the technology will become,” Ruch noted.
Artificial intelligence is increasingly being adopted by food companies. With vast amounts of data and the pressure to innovate and accelerate product launches, companies are moving beyond traditional R&D and testing methods. For instance, McCormick & Co. partnered with IBM three years ago to analyze data more efficiently, identifying compatible ingredients or suitable substitutes. McCormick has already launched eight products developed with AI assistance.
This technology allows “our product developers to explore flavor possibilities more swiftly and efficiently, learning and predicting new combinations based on sensory science, consumer preferences, and flavor profiles,” the company explained in an email. Conagra Brands, known for products like Healthy Choice and Reddi-wip, also utilizes AI-enabled platforms to pinpoint consumer preferences and bring trendy products to market in significantly less time. In 2020, Brightseed and dairy giant Danone North America announced a partnership to use AI for uncovering hidden nutrients in soybeans.
The average human tongue has around 8,000 taste buds, which help identify the five primary tastes: sweet, sour, salty, bitter, and umami. However, the experience of tasting food is heavily influenced by the sense of smell. Unsurprisingly, these two senses are closely linked. For instance, while taste buds identify sweetness, the nose determines the specific sweetness related to a fruit, like a strawberry or a grape. Aromyx, a biotechnology firm, estimated that while the average human nose can discern a trillion different odors, it often struggles to differentiate between smells and tastes. To address this, Aromyx has turned to artificial intelligence.
The company has replicated receptors found in the nose and tongue. Scientists apply a sample, such as a drop of coffee, to these cloned receptors before measuring their response, effectively mimicking the brain’s reaction when a person sips coffee. This process, repeated millions of times, allows Aromyx to build a database for external R&D teams to reference while testing coffee. This data can help consumer packaged goods (CPG) companies predict whether consumers will enjoy similar coffees, how reactions may vary by demographics or location, and the reasons behind these preferences — for instance, if a coffee has a floral note or a particular ingredient.
“Now, we can quantify this information and make targeted recommendations about what products people will prefer, why they will like them, and how to enhance flavors or aromas for specific groups,” stated Josh Silverman, CEO of Aromyx. The database can identify flavor profiles and suggest necessary adjustments to ingredients to achieve the desired taste. Silverman noted that one of the most common applications of his platform is in reducing sugar content. Many CPGs are reformulating products to decrease sugar while maintaining original flavors, and Aromyx’s database can pinpoint which chemicals need alteration once sugar is removed to restore the original taste.
Silverman likened this process to color mixing at a hardware store, where the desired shade is universal. Each location uses a machine to measure customer preferences and determine the precise amounts of each dye needed to create that specific color by controlling the proportions of red, green, and blue wavelengths reflected in the paint. “Instead of random trial-and-error methods, food companies now have a clear direction,” he explained. “They know precisely which ingredients to adjust and by how much, allowing them to intelligently navigate toward the final solution. This insight is incredibly valuable.”
This approach can save millions in development costs while accelerating product launches — a significant advantage in a highly competitive food market where consumer trends shift rapidly. Each year, approximately 30,000 new food and beverage products hit the market, yet, despite extensive research and testing that spans months or even years, most fail. Nielsen reports that 85% of products in the CPG sector typically fall short within two years, prompting companies to seek innovative strategies to gain any competitive edge.
“Large companies are increasingly aware of the financial losses linked to failed product launches,” Silverman remarked. “They realize their internal data often lacks accuracy. If we can provide them with reliable data on customer preferences without the need for full-scale launches and market testing, that is immensely beneficial. It has proven to be a compelling selling proposition.” Aromyx’s revenue has doubled every year for the past three years, and the company secured $10 million in its first Series A funding round last summer to boost production. Business has surged to the point that Aromyx is facing a six-month backlog for new clients.
Silverman refrained from disclosing which food and beverage companies Aromyx collaborates with, aiming to maintain their competitive advantages. Many of Aromyx’s clients are reluctant to publicly acknowledge their partnership, fearing that revealing this information might encourage competitors to follow suit. “They often express, ‘We don’t want to inform our rivals that we are utilizing this technology, as they might adopt it too. We want to keep you under the radar,'” he shared, highlighting conversations with clients. “Those we work with definitely see this as a significant competitive advantage.”
Incorporating advanced technologies like artificial intelligence can also help businesses understand consumer interest in products enriched with nutrients such as calcium citrate, magnesium, and zinc, enhancing their marketability while ensuring they meet consumer demand.
