When Matias Muchnick started NotCo in 2015, food innovation was a slow-moving process.
“Food R&D was three guys in lab coats, doing trial and error in a developmental kitchen,” said Muchnick in a recent interview with The Spoon. “Reading research papers from 1980 about using soy to replace animal-based ingredients. That was it. So whenever you have an industry that has a very obsolete technology, then a lot of bad things happen.”
He and his co-founders wanted to create new plant-based food products, but they wanted to do it in a new way that didn’t rely on antiquated methodologies. Eventually, they started wondering if using technology like artificial intelligence could help them make better decisions and help create new types of food faster.
They decided yes and started building an extensive database of information about all the components that create the taste and experience of food.
“Your machine learning will always be directly proportionate to the amount of data and the dimensions of data that you collect,” said Muchnick. “So from the very beginning, understanding what data was relevant for the objective that we were trying to do, which was replacing animals with plants, was important to us.”
Their database was comprised of chemical and biological attributes that made food what it is and the attributes that impacted the human perception of taste, texture, smell, and color. The goal, said Muchnick, was to create a large enough database of information to use their AI (which would eventually be called Guiseppe) to create a whole line of new plant-based food products.
“We wanted to build a general-purpose artificial intelligence,” he said. “Not an algorithm that is only great at doing mayo, or a burger, or yogurt. The things you’d like from a burger are very different from what you’d like from yogurt, so (we wanted to know) how we could get a real understanding of the human brain to create an algorithm that would attack all of the categories of products. That was super important from the get-go.”
Seven years later, his company is an alt-protein unicorn: the company is growing very fast in the North American alt milk category, just started a joint venture with a food giant, and this week debuted a new alt-meat product at the natural foods show in Los Angeles.
So how does the platform that made all that possible work?
Muchnick gives an example of how the process would work if Giuseppe were used to, say, make a new kind of cheese.
“The algorithm comes up with this crazy amount of recipes and processes attached to each ingredient that we put into it,” said Muchnick. From there, they would take recipes and take them to the “AI test kitchen,” where a group of fifteen chefs try the product out, make the recipe, and then have it evaluated by a trained panel.
“The trained panel gives feedback to the algorithm. Maybe the formulation was good or bad, we feedback the algorithm with the good things and the bad things. So we feedback the algorithm with the many dimensions of the sensory aspects.”
Muchnick says its this continous loop where AI-generated concepts, recipes, and processes are tested in a kitchen, critiqued with feedback, which is then fed back into Guiseppe, which helps NotCo’s AI get better and better.
“You get an algorithm that is working on improving every single day with 1000 recipes.”
But it’s not just recipes getting better, but the optimization of processes around which they run experiments. Muchnick gives the example of a project on frothing plant-based milk. Instead of spending months experimenting with different ways to achieve it, Muchnick says it will help show faster routes to success to help deliver results in a week.
“Instead of starting from scratch with every food formulation you want to create, or any expression you want to create, the AI is telling the food scientist to go through different routes. The algorithm is optimizing every single set of experiments.”
And its this process and the success ultimately drew in Kraft Heinz to consider working with NotCo.
“Kraft Heinz said, guys, you do food products in a quality we’ve never seen before, at a pace we’ve never seen before, and with an agility and an execution that we haven’t seen before,'” said Muchnick. “‘How do you guys do it, and how can we partner up?'”
The answer to that question would eventually be a joint venture.
“They were like, ‘Why don’t we bring superior plant-based products with the familiarity of our brands and with your know-how of executing amazing R&D based products?'”
“And,” said Muchnick, “we’re like, ‘Yeah, I mean, that makes a ton of sense.'”
If you’d like to hear my full conversation with Muchnick about how they are using AI to accelerate food development, just click play below.
