The maker of Oreo cookies is designing new flavors with AI

The adoption of artificial intelligence in the food industry has reached a new milestone as Mondelez International, the maker of Oreos and other popular snacks, reveals its use of machine learning for flavor development.

The innovation strategy: Mondelez has implemented a sophisticated machine learning tool that draws parallels to pharmaceutical drug discovery algorithms, moving beyond traditional flavor development methods.

• The AI system has been utilized in more than 70 products across the company’s portfolio, including the development of Gluten Free Golden Oreos and an updated Chips Ahoy recipe
• Unlike generative AI platforms such as ChatGPT, this specialized tool focuses on optimizing specific sensory characteristics like “burnt,” “egg-flavored,” and “oily” notes
• The system also analyzes flavor factors including “buttery,” “in-mouth saltiness,” and “vanilla intensity”
• Recent research indicates that despite an inability to taste, AI systems can perform surprisingly well in differentiating between different flavors, and in some cases even better than humans

Development and refinement: The machine learning algorithm, created in partnership with software consultant Fourkind, has evolved significantly since its inception in 2019.

• Early versions of the AI produced impractical suggestions, such as cookies with excessive baking soda content due to the ingredient’s low cost
• The company integrated human “brand stewards” to oversee and refine the AI’s recommendations
• Historical recipe and ingredient data form the foundation of the AI’s training dataset

Human oversight and testing: Despite the AI’s capabilities, Mondelez maintains a robust human validation process to ensure product quality.

• Professional taste testers evaluate all AI-generated flavor combinations
• Products undergo extensive internal and external consumer testing before release
• R&D manager Kevin Wallenstein emphasizes the rigorous nature of the tasting process, noting the challenges of repeated flavor testing

Technical limitations: The AI system operates within clear constraints, given its inability to actually taste or smell.

• The tool relies on data patterns and optimization rather than sensory experience
• Human expertise remains essential for validating the AI’s suggestions
• The system functions more as a sophisticated recipe optimization tool than a replacement for human taste expertise

Looking ahead: While Mondelez’s AI-driven approach to flavor development represents an innovative use of machine learning in food science, it also highlights the continued importance of human judgment in product development, suggesting a future where AI augments rather than replaces traditional food innovation methods.