By Javier Surasky
This post is dedicated to Chef Mariano Bargas
Introduction
Artificial intelligence is beginning to intervene in the creation of flavor itself, turning part of culinary creativity tied to social and cultural frameworks into a process that can be calculated, repeated, and industrialized.
This is not about robots in the kitchen or apps that suggest recipes, which are the least interesting aspects of the problem. It is about AI intervening in an activity we usually reserve for “the human”: the creation of flavors, dishes, and experiences.
For a long
time, culinary innovation was understood through a fairly recognizable figure:
the chef as author, the original recipe, the “signature” of a restaurant. That
image still exists, but it is beginning to shift. Today, the dish that reaches
the table may be the result of an interaction between people and computational
models capable of generating variations, proposing unlikely combinations, and
predicting preferences based on datasets.
Human
creativity is shifting toward forms of co-creation mediated by logics of
optimization, standardization, and scalability, historically associated with
industrial processes and now applied to an activity marked by sensory
experience, culture, and situated judgment.
From recipe to system
Using
technology is not the same as co-creating with it: a recipe search engine, or a
digital cookbook, saves time, but it does not “create” in the strict sense.
Something
different happens when a system proposes ideas, alternatives, and
recommendations while a dish is being prepared, and as a result technology
becomes directly integrated into the creative process.
Research on
human-AI co-creativity defines these processes as collaborations in which at
least one human and at least one AI system intentionally contribute to
generating novel and valuable outcomes. Put simply, this means that artificial
intelligence begins to operate as a counterpart with initiative of its own.
In
gastronomy, this intervention is especially complex because preparing a dish
means bringing together ingredients, techniques, timing, textures, and the
expectations of both the person cooking and the person eating, as well as
cultural frameworks that affect the meaning of the act of eating. First
problem: what AI can optimize statistically does not always coincide with what
a diner values as an experience, or with cultural patterns of food preparation.
Evidence
from other fields shows that “effective collaboration depends on careful role
allocation, and a high level of user control usually leads to greater
satisfaction, trust, and a sense of authorship over creative outcomes” (Singh
et al., 2025:28). In cooking, where authorship and sensory judgment are part of
professional identity, that is especially relevant.
This raises
a question that goes beyond the chef: should the person who is going to consume
a dish know that it was co-created with AI before deciding whether they want to
eat it? More than that: is that information part of the dish’s identity?
Optimization and exploration
AI introduces two productive aesthetics into the kitchen: optimization and exploration.
- The first seeks repeatability, error reduction, and the best average combination based on the available ingredients, applying a clearly industrial logic: standardizing results, ensuring stability, and enabling large-scale reproduction. When flavor stability is a commercial asset, this orientation becomes central.
- The second is aesthetic and operates differently, because culinary creativity is also nourished by the improbable, by cultural intersections, and by risky innovation. This is a terrain where AI can recombine ingredients and patterns in counterintuitive ways, but also close off paths when its data reward what has already been successfully tested.
In cooking, the combinatorial space is enormous, creating a permanent tension among preserving a tested flavor, proposing a novel one, designing new food pairings, and generating pleasantness.
In Life, on the Line, Grant Achatz describes his culinary training as shaped by discipline, repetition, collective learning, and material demands. His illness and temporary loss of taste break with any romantic idea of cooking as an expression of talent. For Achatz, cooking is work, system, and a form of vital resistance.
Flavor as data
AI has
become influential in the design of pairings and flavor profiles based on
compounds. The idea of a “flavor network” assumes that ingredients and aromatic
compounds can be connected to one another. Two ingredients, for example, are
linked if they share at least one compound, and the strength of that link
depends on how many compounds they share.
Ahn et al.
(2011) analyze tens of thousands of recipes and show profound cultural
differences: “North American and Western European cuisines exhibit a
statistically significant tendency toward recipes whose ingredients share
flavor compounds. In contrast, East Asian and Southern European cuisines avoid
recipes whose ingredients share flavor compounds” (Ahn et al., 2011:3).
This
approach creates a quantifiable language for discussing culinary affinities
while also showing that there is no universal formula for good pairing:
different cultures organize the chemical affinity between ingredients in
different ways.
NotCo and the industrialization of taste
NotCo makes
it possible to observe this transformation in the food market. The company uses
AI to design plant-based alternatives to animal products. Its system, called
“Giuseppe,” works with modules that identify market trends and map aromatic
compounds to find equivalences or complementarities between ingredients.
The company created an algorithm designed to learn vast combinations of plants in order to replicate the flavor of animal products. Flavor becomes an object of calculation for the computational system, oriented toward its design and reproduction through new combinations. This means that AI actively takes part in deciding what is produced and how, industrializing culinary creativity by breaking ingredients down into chemical data linked to preferences in consumption patterns.
The
FlavorGraph proposed by Park et al. (2021) deepens this logic by integrating
ingredients, chemical compounds, and co-occurrence relationships in recipes in
order to build vector representations of pairings, complementary pairs, and
novel combinations.
Conclusions: creativity, data, and culinary production
Food
requires tasting, fine-tuning, cultural context, and subjective perception. Lee
(2023) warns that AI can defeat world chess champions because that game allows
for rapid and complete evaluation. By contrast, “food cannot really be cooked,
tasted, and evaluated by a computer in the same way chess games can.”
Still, AI
has already transformed culinary activity.
Culinary
creativity was never fully individual or spontaneous. The novelty of AI lies in
algorithmically formalizing that systematic dimension, decoupling it from its
social context, and reorganizing it under standardized and scalable criteria.
Does a
“digital cuisine” exist, then, in the same way that a “French cuisine” or a
“Mediterranean cuisine” exists? Not yet.
A culinary
school requires rules, genealogy, and recognition. But digital cuisine can be
understood as a mode of production and exploration whose distinctive mark is
the industrialization of culinary innovation through data, predictions, and
optimization.
That mode
of production opens up concrete possibilities in areas as different as new
combinations and waste reduction, but it leaves a difficult question: what
happens when the data guiding that creativity do not adequately represent food
cultures, traditions, culinary memories, and situated forms of knowledge that
give meaning to what we eat beyond the dish itself?
That
question will be the starting point of a future post.
References
Achatz, G. y Kokonas, N. (2011).
Life, on the line: A chef’s
story of chasing greatness, facing death, and redefining the way we eat. Gotham Books.
Ahn, Y.-Y.; Ahnert, S.E.; Bagrow, J.P. y
Barabási, A-L. (2011). Flavor network and the principles of food pairing. Scientific Reports, 1, 196.
https://www.nature.com/articles/srep00196
Lee, K. (2023, 5 de mayo). Battling food AI bias and misinformation. The Future Market.
https://thefuturemarket.substack.com/p/battling-food-ai-bias-and-misinformation
Park, D.; Kim, K.; Kim, S.; Spranger, M. y
Kang, J. (2021). FlavorGraph: a large-scale food-chemical graph for generating
food representations and recommending food pairings. Scientific Reports
(11:931). https://www.nature.com/articles/s41598-020-79422-8
Singh, S.; Hindriks, K.; Heylen, D. y Baraka,
K. (2025). A Systematic Review of Human-AI Co-Creativity. ACM / ArXiv. https://arxiv.org/pdf/2506.21333