Noa Weiss is an expert in the field of artificial intelligence and machine learning. Having worked with data for the last 13 years, starting ‘even before it was called data science’, AI has become a ‘true passion’, she told FoodNavigator.
Weiss is also a vegan, and a ‘big believer’ that the food industry needs to shift away from its dependence on animals towards plant-based nutrition. “In recent years, I’ve been thinking: How could I use my own skills to help the alternative protein industry…move towards that?”
The result is GreenProtein AI: a non-profit using AI and machine learning to help plant-based manufacturers using extrusion technology improve their texture, in a bid to ‘revolutionise’ the industry.
The ‘pain point’ of extrusion technology
Extrusion technology is used by most plant-based meat makers to transform native ingredient biopolymers (inputs) into a continuous semi-solid (output).
Extrusion technology in detail
According to R.C.E. Guy in the Encyclopedia of Food Sciences and Nutrition, to complete the extrusion process, a screw system within a barrel conveys mass (a combination of dry ingredients, water, and/or oil) through a die (small opening). The mass is exposed to a combination of parameters, including heat, mechanical energy, pressure, and moisture.
The continuous fluid passes through the extruder according to its specific parameters, exits a die, and is cut into various lengths. Post-extrusion, the output is semi-solid and retains moisture. Wet Textured Vegetable Protein (TVP) output will go through marination, coating, and/or cooling steps. To create dry TVP, the product will pass directly from the extruder onto a conveyor, and then into a dryer.
Since the technology is already being used at scale, Weiss is backing extrusion-based manufacturing infrastructure to produce the capacity required for a plant-based transition.
Of course, not all plant-based manufacturers use extrusion tech. Redefine Meat is one example of a company leveraging additive manufacturing technology to mimic meat with plants. But Weiss is not convinced such technologies can so easily scale.
That is not to say that extruders are perfect, and Weiss revealed plant-based manufacturers often have a ‘love-hate’ relationship with the technology. Protein interaction with extrusion technology can be unpredictable. “On the one hand it gives them texture, and on the other it can be very unpredictable: the tiniest changes can cause fluctuations in the formula.”
The trial-and-error process required can prove costly not only in time but also money: renting (start-ups are more likely to rent, than purchase their own equipment in the early days) extruders is expensive. “It seems to be a huge pain point…and somewhere that machine learning could help.”
A non-profit built on ‘trust and collaboration’
Weiss’ decision to build a non-profit, rather than for-profit, organisation came from a desire to create a network based on trust and collaboration for industry, we were told. Although based in Israel, GreenProtein AI sits within US-based not-for-profit Food Systems Innovation (FSI), which is dedicated to creating a more ‘sustainable, humane, and equitable’ global food system.
In order to work with GreenProtein AI, plant-based manufacturers are required to provide comprehensive measurements and data from their extrusion trials for AI model development. They should also be committed to the feedback process, meaning that when collaborators use the AI models, they provide feedback on performance and outcomes, and engage in iterative improvements.
Using classical machine learning rather than large language models means that less data is required in comparison, but GreenProtein AI still wants to source the highest quantity of quality data it can. “The more data we get, the better we’ll be able to make the algorithm.”
Weiss is aware that compared to the research institutes GreenProtein AI is collaborating with, commercial companies are more likely to want to keep their data confidential. But to make use of the GreenProtein AI platform, businesses will be required to share data. Being a non-profit helps makes it clear to such companies that GreenProtein AI will not share their data for profit.
Indeed, GreenProtein AI has committed to ‘strict confidentiality’ of shared data and IP and all data are aggregated and stripped of any identifiers, eliminating the possibility of associating specific data points with individual facilities or collaborators. Secure access is also ensured for data protection.
Benefits for collaborators amid a plant-based meat slowdown
For collaborators both big and small, GreenProtein AI is offering enhanced product quality and consistency: AI-driven insights aim to improve the texture, taste, and sensory attributes of their plant-based meat products.
In optimising extrusion processes, the non-profit also aims to enable scalability without compromising quality, reducing waste, and enhancing product efficiency.
Amid a plant-based meat slowdown, whereby consumers are cutting back on plant-based meat and even big-name brands such as Beyond Meat are taking a hit, Weiss expects such a platform could ‘really make a huge difference’.
But even if the plant-based meat industry was enjoying healthy growth, Weiss would still be encouraging the use of AI and machine learning within the sector. “The better your product is, the lower the price is, the more people you’re going to acquire as consumers. Even before COVID, when the market was [experiencing stronger growth], plant-based meat still wasn’t reaching the masses it needs to.
“Research shows you need to have the mouthfeel as good as animal-based meat and the price at least at price parity…and we’re not there yet. We hope to help the industry achieve that.”