How Unilever is harnessing AI to create better food

Artificial intelligence (AI) and machine learning has been around since the 1950s, but only now is the true extent of its potential beginning to be understood.

Throughout its almost 75-year existence, AI has become more sophisticated and user-friendly, with tools now being used by researchers, including the Unilever team, that give way to great possibilities.

Carla Hilhorst heads up R&D at the Unilever Nutrition Business Group, operating from the Food Innovation Centre in Wageningen, the Netherlands.

Known as the ‘Hive’, a team of around 500 employees working from the campus are able to regularly collaborate with students from neighbouring Wageningen University & Research, while also utilising Unilever’s suite of AI modelling tools to innovate.

In addition to product development for brands such as Hellman’s, Knorr and The Vegetarian Butcher, the group is involved in a number of forward-looking projects that hope to improve health and reduce emissions.

With these projects running concurrently and the urgency for businesses to become more sustainable only growing, the use of AI modelling and big data is taking on increasing importance.

“The use of AI has transformed the way we perform repetitive tasks, which can now be performed by models, allowing us to create new products​,” Hilhorst told Food Manufacture.

“We have been scaling digital technology for years and it is really helping us to work smarter in this area of product formulation.”​

Creating the right foundations for AI

With the introduction of platforms such as ChatGPT, most will be familiar with a certain degree of AI. However, the way in which Unilever is using it is somewhat different.

Its models are built on a foundation of proprietary data gleaned from years of research and experimentation, with intelligence focused on the texture and sensory profile of a raft of ingredients simultaneously informing the formulation of new products.

With AI, the research and testing phase for these new products can be expedited, but this is only worthwhile because of the data that informs the models.

“There is no AI if you do not have the right data foundation​,” Hilhorst explained.

“You need to have the right quantity of data and the right quality of data, which requires you to continuously evolve and build your data management systems and your modelling capabilities. Even if you have all of those digital tools, it is not good enough as you also need to ensure that your workforce possesses the digital skills and a digital-first mindset​.”

What is Unilever’s AI capable of?

The capability now possessed by employees of the Nutrition Business Group was not built over night though. Hilhorst admitted that when the team first used the technology 15 years ago it was “a bit scary”​, but that it is now used to empower human experts to perform their roles.

“If today we design a vegan mayonnaise, we would conduct the majority of our analysis using the model, predicting the shelf life without the need for extensive testing,” ​she added.​

“We also use our models to inform what the formulation of our products needs to look like in terms of taste and texture. The modelling techniques allow us to test the sensory profile we desire against 400 different formulation options almost instantly​.”

The data collected through this experimentation is then fed back into the Hive’s model, which further improves the tool.

“Continuously developing that suite of quality data enables us to use the AI to our advantage​,” Hilhorst said. “With any model, rubbish in gives rubbish out​.”

Using AI to develop alternative proteins

The development of alternative protein products has been a common trend in food manufacturing over the past 10 years, with plant-based burgers and hot dogs now widely available.

One Unilever brand in this space, The Vegetarian Butcher​​, is currently stocked by supermarkets around the UK including Tesco, Sainsbury's and Co-op.

Over recent months there has been several plant-based brands forced into administration, evidence of the difficult market conditions facing producers in the space, and it’s clear that consolidation is occurring across the segment.

Despite the headlines, Hilhorst is bullish on meat alternatives provided they get the right flavour.

“We know that if food does not taste nice, people will not buy it, regardless of how good it is for the planet or their health​,” she explained.

In a bid to achieve better tasting alternatives, Hilhorst and the nutrition group’s R&D team have conducted extensive research into fermented products, while also working with companies like Algenuity on microalgae-based alternatives.

The project with Algenuity involves the discovery of a Chlorella strain within microalgae that is composed of high-quality amino acids.

While microalgae tend to possess an unappealing taste and colour, this particular strain does not, therefore making it more suitable for use in alternative protein products.

“By selecting this specific Chlorella strain during the fermentation process, we can use our modelling capability to include it as a source of protein in meat-free products,” ​Hilhorst said.

In other words, the AI tool can identify the combination of ingredients required to create the product so that work can begin on testing.

Understanding the gut microbiome

One area that Hilhorst has been focused on recently is research into the gut microbiome, a part of the body that consists of all the microbes that live inside human intestines.

Hilhorst said that around “90% of your health and wellbeing” ​is linked to your gut microbiome in some way, and that a person’s diet can have a significant influence on the make-up of their microbiome.

With this in mind, the Unilever team at the Hive hope to explore this connection between diet and health, with the goal of improving wellbeing through the development of food products.

“Each gut microbiome is very individual, like a fingerprint, and this is where big data and AI can help us uncover some of its secrets​,” Hilhorst said.

“There is an ingredient called GABA which is known to have a positive effect on stress and cognition. Some manufacturers include GABA in supplements already, so we started to look into including it in our food products. We also know that some of the bacteria in your gut actually produce GABA, the same way that your body produces serotonin​.”

With this goal in place, Unilever has partnered with a research outfit called Holobiome, which successfully identified the bacteria in the human gut microbiome that produces GABA naturally.

“With Holobiome, we are using our model to test these bacteria against different ingredients to see which combinations will lead to the production of more GABA​,” Hilhorst said.

“We have now completed this screening and identified a few leads, before conducting an in vitro study that gave us a smaller selection. These will now be tested with [another] in vivo study and that will hopefully provide us with proof of concept. Once we identify what leads to GABA being released, we can incorporate that into products.”​

The human study will kick off at the end of 2023, with Hilhorst hopeful that the potential ingredient could be used like a prebiotic in the future.