AI and the Future of Flavour

Ever thought of pairing oysters and kiwi? How about caviar on your white chocolate? These pairings sound odd, but artificial intelligence (AI) can help us discover some surprisingly delicious combinations.

AI finds what we actually like

Food manufacturers want to know the next brilliant flavor, but what will it be? Market research company Nielson says that 85 per cent of new fast-moving consumer goods – including items like soft drinks and packaged food - fail within two years.¹ 

Some companies are using AI to try to predict the market. Singapore startup Ai Palette integrates online data from social media, recipes and menus to identify upcoming trends as they emerge locally.² Fast food franchise McDonald's uses AI to maximise sales through digital drive-through menus according to time of day, weather, and traffic.³ US researchers have even made an AI to read Amazon reviews of food for hints that it is unsafe and will be recalled by the FDA.⁴

AI and food pairing

Other companies are pursuing AI-inspired gastronomy to create new fusions of flavours based on the science behind how our smell and taste receptors are linked. It’s widely reported that about 80 per cent of food taste is down to smell. Food can taste repugnant to people who have lost their sense of smell. If you pinch your nose, a strawberry does not taste sweet and coffee tastes bitter. What we taste comes down to both the smell of the food before we eat and the aromatic molecules released in the mouth.⁵ 

A food’s aroma can be profiled using a technique called gas chromatography. The company Foodpairing have trained an AI to suggest the most appealing flavours based on shared aroma profiles. For instance, gas chromatography of a strawberry shows it has fruity, cheesy, green and roasted aromas, so Foodpairing’s AI suggests pairing strawberry with cheesy parmesan. It may sound unconventional, but it works.⁶

Future of food tasting

A new food product needs extensive taste-testing, a process traditionally carried out by humans. Newer tech, like an ‘electronic tongue’ made of sensors that can identify the molecular components in food, can provide an alternative to human taste testers. It is used in food testing, and it can rapidly check if honey is pure, identify the grapes used in a vintage of wine, and taste unlimited rounds of spicy food.⁷ But it can’t identify the mouthfeel, the smell, the colour – all of the elements that make human tasting so important.

Discover how colour can affect how we perceive the taste of food

Yet, unlike machines, humans can be inconsistent. They are swayed by colour – tasters given white wine dyed red believed it was genuine red wine.⁸ Taste buds can be quickly exhausted, particularly by spicy food. I’ve never eaten at a Michelin restaurant – is my tasting palate less refined? 

AI flavour banks

Companies like Analytical Flavour Systems (AFS) aim to take taste-testing into the 21st Century. AFS uses data from professional tasters but also collects data via a free app called Gastrograph. Users of Gastrograph can score flavours on carefully selected criteria, such as ‘marine’, ‘gamey’ and ‘floral’, and make specific comparisons to known flavours. There are 600 - 1000 different variables. But Gastrograph can also make predictions and work out the best questions to ask.

AI suggests unusual recipes, but are they really tasty?
IBM had launched AI-powered ‘Chef Watson’ in 2014, a collaboration with partners from Bon Appétit and the Institute of Culinary Education (ICE). Chef Watson could suggest unusual recipes based on a few ingredients – Belgian bacon pudding was a hit, curry powder ice cream less so.¹⁰ 

Think back to the strangest foreign dish you’ve eaten. You might struggle to describe its flavour; you might not even know the name of it. Gastrograph could work out what it is even if you can only describe it in terms you’re familiar with. The makers say it could even suggest how to change a flavour to suit a different culture.⁹

Having a bank of reliable ‘flavour profiles’ can also help monitor food products with a long development time, such as brewing beer, maturing whisky, or fermenting pickles.