A New Era for Tastes and Scents: Protecting AI-Generated Recipes and Perfume
Browse this blog post
Related news and insights
Blog Post: 20 February 2024
Blog Post: 15 February 2024
News: 09 February 2024
News: 09 February 2024
Food and perfume are two simple pleasures in life that invoke two senses: taste and smell. As much as both are linked with personal preferences and memories, the AI revolution has found a way to make its (impersonal) mark in both spaces. AI’s role in creating new recipes for both food and fragrance begs the question: how can companies in these industries that have started employing AI best protect their intellectual property rights?
How is AI used to generate recipes and fragrances?
As early as 2011, IBM employed computational creativity in its AI-powered cooking app, Chef Watson, which selects or combines ingredients to create what the AI considers to have high levels of surprise, pleasantness, and synergy. Since then, several companies promoting AI-generated recipes have debuted, including DishGen, cookAIfood, and Let’s Foodie. Simply enter the desired ingredients, recipe ideas, cuisine types, or dietary preferences, and receive AI-crafted, tailor-made recipes.
Often the textbook illustration of trade secret, Coca-Cola (which boasts a famously secretive formula) has recently enlisted AI’s help to create new flavors and packaging designs. Visitors to the World of Coca-Cola in Atlanta, GA often remark at the sheer creativity behind the seemingly endless number of Coca-Cola’s adventurous flavors. With AI’s help, Coca-Cola continues experimenting with various limited-edition beverages having “mystery tastes” involving “vague, futuristic concepts and undisclosed flavors.”
On the scent side, IBM has partnered with Symrise to develop a scent algorithm “Philyra” for perfume products. Philyra has been analogized to an apprentice continuously learning from perfume masters, by studying existing fragrance formulas and analyzing those ingredients in relation to datasets including geography and customer age. Other companies are leveraging AI to create new fragrances: Givaudan uses an AI tool “Carto” to produce instant samples using a computationally determined formula, EveryHuman uses algorithmic perfumery to generate scent recipes based on a user’s responses to a questionnaire, and No Ordinary Scent uses AI to create scents based on images.
What can we learn from AI-assisted drug discovery?
There has been much focus on the use of machine learning and AI to assist drug discovery, and the underlying principles for protection AI-generated creations for drugs are also applicable to foods and fragrances. For each, AI augments or assists human creativity and innovation, or it can generate new and valuable creations all on its own. There remain challenges and open questions regarding how to best protect such AI-generated creations as well as the AI itself. Companies using AI for food- or fragrance-creation should consider protection strategies utilized by pharmaceutical companies using AI.
In the pharmaceutical industry, patents are essential in covering new drugs, formulations, compounds, molecules, processes for drug development and design, as well as data analysis and visualization techniques applying AI and machine learning training. Patent filings involving AI in drug discovery have steadily grown over the years. This is despite increasing concerns with AI-related inventions, including (1) that such inventions are not patent-eligible because they are too abstract, (2) that such inventions are challenging to describe with sufficient detail to meet disclosure requirements, (3) whether a person of ordinary skill in the art would have access to AI, which would seem to render many inventions obvious, and (4) complications regarding inventorship when the AI plays a significant role in the conception or reduction to practice of the invention.
How can you protect AI-generated recipes and fragrances?
For food and beverage and fragrance companies with segments employing machine learning and AI, there may therefore be limits from patent protection that other IP strategies would help to complement.1
The AI tool itself may find trade secret protection to be an effective strategy that is not limited in term like patents. Already, some pharmaceutical companies apply trade secret protections to certain categories of training data for AI and machine learning models, software code, data analysis processes, and other drug development and analysis-related confidential information. Similarly, traditional perfume and food and beverage companies have historically employed trade secrets to protect recipes. Those employing AI should similarly build robust trade secret frameworks internally to protect their training data for AI and machine learning models, software code, data analysis processes, and other confidential information related to the recipe development and analysis process. To maintain a trade secret, robust protection protocols should be periodically reviewed and reinforced.
Such AI tools require large amounts of data, which also needs protection. Data indicating which ingredients pair well together has always been important in creating commercially successful scents and flavors. Now with the usage of machine learning and AI, data has become even more critical as a core competitive advantage. The “garbage in, garbage out” principle (that nonsense input produces nonsense output) is essential to ensuring the AI model generates tasteful outputs. Thus, the input selection process (including how inputs are formatted and structured) and training of the model becomes critical. In terms of data protection, food and beverage and fragrance companies should consider ways to clearly define the purpose and scope of data use in relevant agreements, including data use clauses specifying how certain data can be used, for which purposes, and by whom.