How Mars is developing new ingredients with AI

Artificial intelligence is no longer a novelty – it’s part of everyday life. Large language models (LLMs) answer questions, craft social media posts, and streamline repetitive tasks for consumers worldwide. But in the food and beverage industry, AI’s potential goes far beyond that. Manufacturers are leveraging advanced tools to predict consumer trends, accelerate product development, and in the case of confectionery and snacking giant Mars, Inc., even discover entirely new ingredients for future formulations. According to Mars – the multinational behind Dolmio pasta sauce, Pedigree dog food, and of course, the Mars bar – the search for new ingredients never stops. That drive, explains Darren Logan, VP of research at the Mars Advanced Research Institute and Global Food Safety Center, is fuelled by its desire to boost health and sustainability through food. On the health side of things, Mars is looking to bioactive compounds in plants to expand its functional food offering. In sustainability, the company is interested in alternative protein sources with lower environmental impact. And if new ingredients can be identified that help mitigate potential challenges in plant-based, like the bitterness of alternative proteins, even better. Add fermentation to the equation, and the opportunity to develop new ingredients from the thousands of edible plants on the planet (just a fraction of which is used in food and drink today) expands dramatically. “When you combine plants with microbes through fermentation, you massively expand chemical diversity,” explains Logan. “This creates new compounds that wouldn’t exist otherwise.” So how does Mars go about identifying these new bioactives and ingredients? That’s where UC Davis spin-out PIPA comes in. Mars has been working with PIPA and its AI research assistant platform, LEAP, for several years with the aim of discovering a new suite of ingredients. Unlike an LLM, PIPA’s LEAP platform helps build a “knowledge graph” that in turn, helps Mars identify connections between ingredients, nutrition, certain microbial elements, and human health. It does this by combining publicly available scientific literature, databases, and omics data with Mars’ proprietary data. “PIPA owns the platform and technology, but we’ve been working with them to optimise and personalise our own version within Mars,” explains Logan. “That lets us integrate what we know as a company with what the rest of the world knows publicly, and use that for innovation, food science research, and increasingly, productivity and efficiency.” The real boon for Mars comes in saving time and money on endless experiments to identify the right connections. Of course, Mars will always run its own tests on bioactives before considering them for product development, but since the LEAP platform understands previously identified relationships, it can recommend which tests are most likely to yield a benefit. “We’ve tested this between random choices, expert choices, and AI-assisted choices,” says Logan. “What we see is that in some cases, AI significantly improves hit rate.” However, as we’ve come to learn, AI is not only used for good. When data is incorrectly handled, AI can present existential risk. AI tools used in business can create legal and ethical challenges, explains Logan. “One emerging risk is AI learning from AI-generated content, which can amplify errors.” But the VP is quick to stress LEAP is not an LLM trained on the internet at-large, and instead uses structured knowledge graphs learning from scientific literature only. “We’re also working on filters to identify AI-generated or low-confidence content,” reveals Logan. Mars is also using AI in packaging to simulate alternative materials and shapes and cut plastic use. In pet care, the multinational is using the technology to read radiographs much faster, and in advertising and design, AI is being used to generate packaging concepts in minutes. AI is also being used to help make new products safer. A toxicologist will also make the final decision, but leveraging AI helps save time. There is always a human “in the loop”, says Logan, stressing the importance of human-made decision-making. “Across all our AI work, Mars operates under a strong AI policy that encompasses security, ethics, legality and a ‘do no harm’ principle. No AI decision is made without human accountability.” As for Mars’ work with PIPA, Logan reveals new ingredient discoveries have already been made, but that it takes time for new ingredients to reach products. “New ingredients are in progress,” he says, and expects results to be revealed publicly within the next six months.