People have long assumed that machines aren’t capable of being creative. Maybe some of this is ego: machines already have an easier time moving heavy things, they can beat us at chess or Go, and process massive data sets faster than us, so it’s natural to want to claim creativity as a uniquely human strength. Sure, a machine can solve a Rubik’s cube in less than one second, but they don’t create great works of art or make music that stands the test of time, right?
In fact, perhaps they do. Machines have now created musical melodies and generated artwork—and it’s getting harder to tell if art has been made by a human or not. Machines, it turns out, can be creative.
Computational creativity is a multidisciplinary endeavour located at the intersection of artificial intelligence, cognitive psychology, philosophy, and the arts. At Accenture Labs, we focus not on the creation of artwork but on products and formulations, such as those created by Fast Moving Consumer Good (FMCG) companies. From our perspective, computational creativity is the application of AI to enhance human creativity with assistive tools.
Using AI to enhance and augment the creative process means companies can innovate faster. Tools that use computational creativity to generate new product ideas or product adaptations will allow teams to tailor products to more specific market segments, and get customized products to market faster than before.
To develop a proof of concept, we chose a domain dear to everyone’s hearts: food. Many FMCG companies offer food products, and the space also has a wealth of publicly accessible data. What’s more, even with large cultural and regional variations, food is something that’s instantly familiar to all of us. All of these factors make food a great testbed for computational creativity.
AI tapas: flavor combinations suggested by the Labs proof-of-concept system
We held a design workshop at our Dock location, working with an AI system that suggests new ingredient and flavor combinations. On its face, suggesting new combinations of ingredients and flavors seems easy: Why not combine every possible pair, triple and so on? An automated tool could do this easily. Unfortunately, this would result in a huge number of potential combinations to sort through, and without any filtering, many of these combinations will likely be unpleasant (“Pleasantness” here refers to human-rated olfactory pleasantness, which comes from the flavor molecules in the ingredients of a recipe or the product). A suggestion to combine tomatoes with peanut butter, for example, is probably not very helpful.
Our AI system incorporates a knowledge graph—a structure that categorizes different pieces of data by the way they relate to each other—of recipes, ingredients, recipe region, flavor compounds, nutrients and health benefits. The system uses this knowledge graph to make combination suggestions against requests like novelty, surprise and pleasantness. A deep learning model predicts pleasantness, and people can interact with the system to explore additional ingredients and flavors: for example, you can ask what ingredient could be added to a chocolate bar that is pleasant and offers health benefits. Some combinations that have made it onto the menu include parmesan with dark chocolate; beetroot and lemongrass; and fudge and pumpkin.
We invited Robert Collender, co-owner of the Michelin-starred Mews restaurant to join our design workshop. Robert admitted when he arrived that he was skeptical about AI exhibiting the kind of creativity that chefs do when creating dishes. But when he realized that AI is designed to work with humans, not replace them in the creative process, he saw that the system could help culinary experts searching for new ways to deliver creativity.
“I can see how AI could be used in terms of staying on trend, giving you a good idea of where trends are moving, and also helping with new flavor combinations that might not necessarily be obvious,” said Collender. “[I can see] that you could get some real winners from what AI can give you, and also give you that element of surprise for the customer—a good surprise, rather than a bad one!”
We’ve since worked with the catering team at the Dock to create “AI Tapas” using flavor combinations from the prototype. These are now are on the menu for client workshops held on site, which we use as a “hands on” way to introduce clients to the concept of computational creativity, and explore with them how it can be applied to their business.
Looking forward, we’re adapting what we’ve learned from these early explorations for applications in product formulations across the FMCG landscape. One persistent challenge in this space is finding substitutions for ingredients quickly, in the case of supply chain delays or specific shortages; in addition to suggesting new combinations, our system supports searches for such potential ingredient substitutions.
Companies are also facing increasing pressure from consumers to make their products more targeted and personal. The “post-digital” world (as outlined in the 2019 Accenture Technology Vision) will be characterized by companies looking for the next competitive edge, moving to capture “moments” of opportunity. Doing so requires developing new, personalised products at speed that surprise and delight the consumer, and that’s just where computational creativity can play a central role: in the innovation or re-innovation of products to deliver creativity at scale.
To learn more about our Chef’s Kitchen demo or our work in computational creativity, contact Jeremiah Hayes.