The latest news, trends, analysis, interviews and podcasts from the global food and beverage industry

Food System Innovations (FSI) has launched the Food Intelligence Lab, a new interdisciplinary initiative designed to build the AI infrastructure needed to accelerate research and development across the sustainable protein sector.

Backed by a $2 million grant from the Bezos Earth Fund awarded last year, the lab will develop open-source datasets, machine learning models and benchmarking tools that aim to shorten product development timelines and improve the sensory performance of plant-based and other sustainable protein products.

The announcement comes as alternative protein companies continue to face slowing consumer adoption, driven in part by persistent concerns around taste and texture despite years of investment in formulation improvements.

Anna Thomas, director of machine learning at the Food Intelligence Lab and a computer scientist at Stanford University, said: "AI is already transforming fields like drug and materials discovery, but food still lacks the shared infrastructure needed to fully unlock the potential of AI in this space. We're building tools to help food scientists iterate faster and create truly exceptional sustainable protein products."

Unlike sectors such as pharmaceuticals, where extensive public datasets have accelerated AI development, food formulation remains constrained by fragmented data, proprietary research and expensive experimental cycles.

The Food Intelligence Lab aims to address that gap by creating large-scale, open datasets that combine sensory evaluations with instrumental measurements such as texture profile analysis, pH and shear testing. These resources will underpin AI models capable of predicting consumer-relevant attributes, including taste and texture, before products undergo physical testing.

The initiative will also collaborate with food companies, academic researchers and non-profits to translate those models into commercial R&D workflows.

FSI highlighted an early collaboration with Proxy Foods AI, in which the teams developed an optimisation system known as Expert-Guided Bayesian Optimisation (EGBO).

According to the organisation, the AI system improved the sensory performance of a plant-based Greek-style yogurt by 29% in just 10 formulation iterations completed over five days. The optimised formulation matched an animal-based benchmark on three of four key sensory attributes – consistency, creaminess and tanginess.

FSI also reported that EGBO outperformed a professional food scientist working under the same time constraints, achieving a higher optimisation score while arriving at a stronger formulation more quickly.

Panos Kostopoulos, founder and CEO of Proxy Foods AI, said: "Food scientists shouldn't have to spend months on trial-and-error to get texture, mouthfeel, flavour, and aftertaste right."

Kostopoulos continued: "Partnering with FSI's Food Intelligence Lab to open-source these tools is how we accelerate those breakthroughs and ultimately change how we feed the planet for the better."

Beyond formulation optimisation, the lab is also developing AI tools to predict sensory outcomes, an area that could reduce the industry's reliance on costly and time-consuming consumer taste panels.

Researchers recently introduced TasteBench, an open benchmark and Kaggle competition that evaluates AI models on their ability to predict how closely sustainable protein products resemble their animal-based counterparts. According to FSI, the strongest AI model currently performs at roughly the level of the median human sensory panellist.

The organisation believes combining scientific literature, experimental data, foundation models and human expertise will eventually enable AI systems to recommend the next best formulation experiment, significantly reducing development timelines.

The launch reflects a broader trend toward applying artificial intelligence across food product development, as manufacturers seek to improve formulation efficiency while lowering development costs.

For the alternative protein sector, where consumer acceptance remains closely tied to sensory quality, more accurate predictive tools could help companies bring better-performing products to market faster while reducing reliance on costly trial-and-error experimentation.

By making its datasets, benchmarks and models openly available, FSI also hopes to lower barriers for start-ups, academic researchers and established manufacturers, encouraging greater collaboration across the sustainable protein ecosystem.

The Food Intelligence Lab's research, including papers on EGBO and TasteBench, will be presented at the AI for Scientific Discovery Workshop during the 2026 International Conference on Machine Learning.