Artificial Intelligence is On the Menu for Food Researchers

Thirty-five percent of all global greenhouse gas emissions can be attributed to the food sector, according to the authors of a recent Nature Food paper. Alternative protein sources could help feed the growing population while addressing climate goals, but developing new foods that have the right taste, texture, and stability can be challenging.

Designing food materials is a complex and lengthy process. On a technical level, interactions must be balanced among proteins, carbohydrates, lipids, and processing conditions. The materials need to be affordable, easy to reproduce at scale, nutritionally valuable, shelf-stable, culturally appropriate, and of course delicious.

This is where AI comes in. Alongside a multi-disciplinary team led by MIT and Stanford University, co-author and Stern Family Distinguished Professor David Kaplan explored how artificial intelligence could improve food development with a focus on human health and sustainability. The paper, “Artificial Intelligence for Food Innovation” examines how AI could help shift food development from trial-and-error to data-driven discovery.

How AI fits into the food development process

Food scientists translate people’s food preferences—usually abstract qualities like “crunchy” “crispy” or “spicy”—into concrete physical and chemical properties that can be replicated. AI could help establish connections between the molecular composition of ingredients and how they perform as food. Based on chemical structure, physics-based machine learning and data-driven models could predict sensory outcomes and food texture and use this data to suggest new flavor compounds that align consumer preferences with sustainability.

Machine learning predictions could also help develop ingredients and recipes more efficiently. “Computational gastronomy has long shown that recipes and flavor chemistry form structured networks that we can mine for predictive patterns,” the researchers write. Rather than relying on a lengthy trial-and-error process, researchers could use AI to identify and test the most promising flavor compounds first.

As deep reasoning capabilities evolve, the authors envision a future with self-driving labs that design and test combinations in a self-learning closed loop system. These systems would factor in constraints such as cost, supply chain, allergen control, and clean label requirements. “By embedding sustainability constraints, cultural diversity, and personalized health data, these systems can design foods that meet functional, nutritional, and societal goals,” the researchers write. Although they imagine a future with more automation, they acknowledge that human validation is still an essential part of the process to ensure the highest quality results.

Tufts brings expertise in future foods

The paper was a collaboration among researchers from Tufts, MIT, Johns Hopkins University, Stanford University, Imperial College London, the University of Toronto, Southern Denmark University, the University of California Davis, the University of Leeds, and an AI company called NotCo. As Director of the Tufts University Center for Cellular Agriculture, Kaplan contributed Tufts’ expertise in developing high-quality proteins and food sustainability to the project.

There is still a lot of work to be done before AI-driven food innovation becomes a widespread reality, but Kaplan and others see a promising path forward for future foods to be developed more efficiently and sustainably with the help of AI.  

Read more about Tufts Center for Cellular Agriculture.