Stanford's AI designs burgers that beat Big Mac while slashing environmental impact

Generative artificial intelligence creates delicious, sustainable, and nutritious burgers - npj Science of Food

Beyond validation, we integrate environmental sustainability and nutritional quality as additional criteria to select from a large ensemble of generated recipes. We quantify environmental impact using land use, greenhouse gas emissions, eutrophication potential, and scarcity-weighted water use30, and assess nutritional quality using established profiling frameworks, including the healthy eating index31. By sampling broadly and selecting recipes that jointly satisfy palatability, sustainability, and nutrition, we show that generative artificial intelligence can identify burgers that substantially reduce environmental impact and improve nutritional quality without abandoning cultural familiarity. Sensory validation with more than 100 participants confirms that the model correctly captures human taste preferences and creates burger recipes that match or exceed the sensory appeal of the classic Big Mac®. Generative AI successfully creates burger recipes We first validate the generative AI model by testing whether it reproduces key statistical properties of 2216 human-designed burger recipes while generating novel combinations (Fig. 1). The model architecture combines a multinomial diffusion model for ingredient selection with a score-based generative model for ingredient quantification, which together generate complete burger recipes defined by 146 ingredients and their quantities (Fig. 1a). Comparison of generated samples with the training data shows close agreement across multiple marginal and higher-order statistics: In particular, the model reproduces the distributions of ingredient quantities (Fig. 1b) and ingredient popularity, defined as the probability of ingredient occurrence across recipes (Fig. 1c), which demonstrates that it learns both how often ingredients appear and in what amounts. The model also captures higher-order structure, including strong positive and negative correlations between ingredient pairs commonly observed in real recipes (Fig. 1d), and accurately matches the distribution of recipe length, measured by the total number of ingredients per burger (Fig. 1e). After establishing statistical fidelity, we generate one million burger recipes and map their palatability, environmental, and nutritional scores to reveal the structure of the generated design space (Fig. 1f). Recipes with high palatability scores cluster in regions associated with popular, conventional ingredient combinations with lower nutritional and medium environmental scores, whereas recipes with low palatability scores occupy regions associated with rare or unconventional combinations, consistent with human culinary preferences. Together, these results demonstrate that the generative model learns the underlying distribution of real human-designed burger recipes and enables systematic exploration of trade-offs between palatability, nutrition, and environmental impact. Generative AI rediscovers classic burgers and creates novel designs We next assess whether the generative AI model can both rediscover canonical burger recipes and generate novel, appealing alternatives (Fig. 2). We quantify similarity between generated samples and a reference recipe using a substantial difference score, where SDS = 0 indicates a match in ingredients and quantities, and SDS > 0 measures increasing novelty (Fig. 2a). From random samples, the model successfully rediscovers the classic Big Mac®, both in correct ingredients and weights, although the Big Mac® was never part of the initial training data (Fig. 2a). Across ten independent randomizations, rediscovering the Big Mac® requires on average 7.3 million samples which demonstrates that exact replication of recipes is a low-probability event under the learned distribution (Fig. 2e). Beyond rediscovery, the model generates new burgers with varying degrees of novelty, illustrated by two representative recipes, the Delicious Burger 1 with SDS = 3 and the Delicious Burger 2 with SDS = 6, which exhibit progressively more distinct ingredient profiles while retaining familiar burger structure (Fig. 2c, d). Sensory evaluation indicates that the two delicious burgers achieve consumer ratings that are comparable to, and in some cases exceed, those of the classic Big Mac® (Figs. 2f, g): Delicious Burger 1 received significantly higher ratings than the Big Mac® for flavor (5.8 ± 1.3 vs. 5.4 ± 1.5) and Delicious Burger 2 for overall liking (5.7 ± 1.2 vs. 5.3 ± 1.5) and flavor (5.8 ± 1.3 vs. 5.4 ± 1.5), whereas both texture ratings did not differ significantly from the Big Mac® (n = 101, p < 0.05). Participants more frequently described the Delicious Burger 1 as meaty (67% vs. 42%), moist (60% vs. 32%), and fatty (40% vs 12%) than the Big Mac®, and the Delicious Burger 2 as meaty (66% vs 42%) and smoky (47% vs. 4%) than the Big Mac® (n = 101; paired comparisons, p < 0.05) Together, these examples validate the generative AI model by demonstrating its ability to internalize canonical burger recipes and generate novel, delicious, and palatable designs. Generative AI creates sustainable burgers We next evaluate whether the generative AI model can identify and generate burger recipes with reduced environmental impact while maintaining consumer acceptance (Fig. 3). We quantify sustainability using an environmental impact score that aggregates ingredient-level contributions from land use, eutrophication potential, scarcity-weighted water use, and greenhouse gas emissions (Fig. 3a). Analysis of the training data shows that environmental impact scores vary substantially across recipes of different primary protein sources, with lamb- and beef-based recipes exhibiting systematically higher impacts than poultry- and mushroom-based recipes (Fig. 3b). Sampling one million recipes from the model enables identification of sustainable burger candidates, illustrated by two representative examples, Sustainable Burger 1 and Sustainable Burger 2, which differ in ingredient composition and dominant protein source (Fig. 3c, d). Sustainable Burger 1, a mushroom-based formulation, achieves an environmental impact score of 0.06, more than one order of magnitude lower the Big Mac® with 0.93, whereas Sustainable Burger 2, a mushroom-beef blend, with 1.02 ranks comparable to the Big Mac®(Fig. 3e). Consumer feedback indicates that Sustainable Burger 1 scores modestly below the Big Mac® in overall liking, flavor, and texture, whereas Sustainable Burger 2 performs on par with the Big Mac® in these categories. Sensory evaluation indicates that the two sustainable burgers achieve consumer ratings that differ in systematic ways from those of the classic Big Mac® (Fig. 3f, g): Sustainable Burger 1 received significantly lower ratings than the Big Mac® for overall liking (4.8 ± 1.8 vs. 5.3 ± 1.5), flavor (5.0 ± 1.9 vs. 5.4 ± 1.5), and texture (4.5 ± 1.9 vs. 5.2 ± 1.5), whereas ratings for Sustainable Burger 2 did not differ significantly from the Big Mac® across these attributes (n = 101; p < 0.05). Participants more frequently described the Sustainable Burger 1 as earthy (63% vs. 2%), strong (37% vs. 14%), moist (53% vs. 32%), and soft (50% vs. 25%) than the Big Mac®, and the Sustainable Burger 2 as smoky (55% vs. 4%), moist (49% vs. 32%), and fatty (30% vs. 12%) (n = 101; paired comparisons, p < 0.05 for all reported attributes). Together, these results validate that the generative AI model successfully navigates the trade-off between sustainability and palatability and discovers burgers with markedly reduced environmental impact without compromising taste. Generative AI creates nutritious and personalized burgers We next evaluate whether the generative AI model can identify burger recipes optimized for nutritional quality and adapt them to individual dietary needs (Fig. 4). We quantify nutrition using the healthy eating index, which aggregates contributions from food groups to promote, fatty acid composition, and nutrients to limit (Fig. 4a). Analysis of the training data reveals substantial variation in healthy eating index across recipes with different primary protein sources, with bean- and mushroom-based recipes exhibiting systematically higher nutritional scores than beef- and lamb-based recipes (Fig. 4b). Sampling one million recipes enables identification of a representative Nutritious Burger with a high nutritional score, which occupies a favorable region of the nutritional-environmental design space compared to the other AI generated burgers (Fig. 4c, d). The Nutritious Burger, a bean-based formulation, achieves a healthy eating index of 63.12, nearly twice as high as the Big Mac® with 33.71, while also reducing its environmental impact score by a factor of six (Fig. 4c). Relative to the Big Mac®, the Nutritious Burger shows improved alignment with dietary guidelines across multiple healthy eating index components, including increased contributions from vegetables, whole grains, and plant protein, alongside reduced refined grains, sodium, and saturated fat (Fig. 4e). Consumer feedback reveals a clear reduction in hedonic ratings for the Nutritious Burger relative to the Big Mac® (Fig. 4f): The Nutritious Burger received significantly lower ratings than the Big Mac® for overall liking (3.8 ± 1.7 vs. 5.3 ± 1.5), flavor (4.0 ± 1.8 vs. 5.4 ± 1.5), and texture (3.7 ± 1.8 vs. 5.2 ± 1.5) (n = 101; p < 0.05). Participants more frequently described the Nutritious Burger as earthy (55% vs. 2%), bland (43% vs. 21%), dry (51% vs. 18%), soft (50% vs. 25%), and grainy (42% vs. 7%), and less savory (27% vs. 53%) than the Big Mac® (n = 101; paired comparisons, p < 0.05 for all reported attributes). Beyond population-level optimization, the AI model also generates personalized burger recipes tailored to individual nutritional requirements. We demonstrate this feature by producing personalized recipes for a highly active 15-year-old male and a moderately active 70-year-old female, which differ in ingredient composition and quantities in accordance with age- and activity-specific dietary needs (Fig. 4g). Finally, a direct comparison across all six burgers highlights systematic trade-offs between ingredient count, novelty, environmental impact, and nutrition, and places the AI generated burgers in distinct regions of this multi-objective design space (Fig. 4h). In this design space, Delicious Burger 1 and Delicious Burger 2 score best in overall liking, flavor, and texture, whereas Sustainable Burger 1 and Nutritious Burger score best in nutrition and environment (Fig. 4i). Together, these results validate that the generative AI model can optimize for nutritional quality at both the population and individual levels while maintaining sensory acceptance.

AI designs the ideal burger for taste, health, and planet

As a proof of concept for AI's broader design capabilities, they believe the same generative design framework could have implications in other fields, like pharmaceuticals, materials, biomolecules, and other complex systems with huge design spaces. Stanford researcher Ellen Kuhl estimates that there are some 10 potential burger recipes in the world. And with BurgerAI, a new tool developed in her lab, artificial intelligence can now design the best one for you based on your age, taste, nutritional need, and even your sustainability goal. But BurgerAI's ability to suggest a great-tasting, nutritionally complex, sustainably produced burger is only part of the story. More broadly, this innovation heralds a shift for AI itself: moving AI from prediction to design. "Most AI systems are trained to predict what already exists. We wanted AI to invent what should exist next," explained Kuhl, a professor of mechanical engineering in the School of Engineering who now directs Stanford Bio-X, an interdisciplinary life sciences institute that brings together researchers across medicine, engineering, and the natural sciences. "BurgerAI does not ask, 'What burger is most likely?' It asks, 'What burger best satisfies these important and complex objectives?'" Food in focus Food is the next big thing in the biosciences, Kuhl said, a focus that combines elements of human experience and culture, health and nutrition, and environmental impact, which are topics that inspire multidisciplinary researchers across the schools of medicine, engineering, sustainability, humanities and sustainability, and beyond. "Food choices are some of the most consequential decisions humans make every day," said Vahidullah Tac, a Schmidt Science postdoctoral fellow in Kuhl's lab. "Food was an easy motivator. With one arrow, you can hit two targets - planetary health and personal health. It's a great and impactful research area." As such, food proved an ideal test bed for Bio-X. Kuhl's team has just published two papers on BurgerAI, of which Tac is the first author. The first paper introduces BurgerAI. The second paper reveals that the same mathematical principles that drive BurgerAI also underpin diffusion-based generative AI more broadly and create connections to technical fields such as materials design, physics, and engineering. "For centuries, food design has been a matter of intuition, experience, and trial and error," Kuhl added. "We are beginning to show that AI can transform food design into a quantitative science with applications in other important fields." Taste-tested Using 2,216 burger recipes from Food.com as a data source, BurgerAI learns patterns in ingredient combinations and quantities and then generates new burger recipes from scratch. The AI then matches those characterizations against human flavor and textural preference profiles. The results are entirely novel recipes optimized for deliciousness, sustainability, and nutrition, and personalized based on gender, age, and physical activity. The ultimate test was not computational but culinary. The researchers served five professionally prepared, AI-designed burgers to more than 100 diners in a blinded taste test at a San Francisco restaurant. In a side-by-side comparison to a popular fast-food burger, BurgerAI's two variations of its Delicious Burger scored the same or better in overall liking, flavor, and texture. Its Mushroom Burger reduced environmental impact by more than an order of magnitude, and its Bean Burger achieved roughly twice the nutritional score of the fast-food burger. "AI did not just generate plausible burger recipes - it created burgers that real people enjoy," Kuhl said. "That may sound simple, but it means the model learned what makes food appealing to the human palate and was able to navigate a design space with near-infinite possible burger combinations to find real-world solutions." Beyond burgers Tac was genuinely surprised by how well the sustainable burgers performed. "We expected some trade-off between sustainability and consumer acceptance," he said. "But we found a burger with dramatically lower environmental impact could still compete with one of the world's most successful burgers." For Tac and Kuhl, BurgerAI is not really about burgers. It is a proof of concept for AI's broader design capabilities. The same generative design framework could have implications in other consequential fields - pharmaceuticals, materials, biomolecules, and other complex systems with huge design spaces. As with food, which requires a balance of taste, nutrition, cost, and sustainability, many of society's biggest challenges must balance competing objectives. If AI can help navigate trade-offs in recipe design, Kuhl said, it could also help discover new medicines, engineer advanced materials, and create more sustainable products. "The burger is just the beginning," Kuhl assured. "We see food as a model system for a much larger vision: AI as a partner in scientific and engineering discovery."

Stanford's AI-designed burgers could advance healthier, more sustainable food

There are, by one researcher's estimate, around 10 to the power of 43 possible burger recipes in the world. That's a number so large it's effectively meaningless, and exactly the kind of challenge BurgerAI was designed to solve. This new tool from Stanford University can design the optimal burger for a specific person's age, taste preferences, nutritional needs, and environmental values. The burger, though, is almost beside the point. The research was published in two papers by a team at Stanford Bio-X, the university's interdisciplinary life sciences institute, led by Professor Ellen Kuhl of the School of Engineering, who now directs Bio-X. First author on both papers is Vahidullah Tac, a postdoctoral fellow in Kuhl's lab. The first paper introduces BurgerAI and its culinary results. The second reveals something broader: the mathematical principles behind it connect to generative AI, materials design, physics, and engineering in ways that extend well beyond food. From prediction to design Most AI systems are trained to predict - to identify patterns in existing data and extrapolate from them. BurgerAI was built to do something different. "We wanted AI to invent what should exist next," Kuhl said. "BurgerAI does not ask, 'What burger is most likely?' It asks, 'What burger best satisfies these important and complex objectives?'" That distinction - prediction versus design - is the conceptual shift the research is pointing toward. Prediction is retrospective, while design is generative. It doesn't ask what has been done; it asks what should be built, and that capability, the team argues, has implications well beyond the kitchen. How the system works The system trained on 2,216 burger recipes from Food.com, learning patterns in ingredient combinations and quantities. It then generates entirely new recipes from scratch. These are not variations on existing ones, but novel combinations optimized simultaneously for taste, nutrition, sustainability, and personal characteristics, including gender, age, and physical activity level. "Food choices are some of the most consequential decisions humans make every day," Tac said. "Food was an easy motivator. With one arrow, you can hit two targets - planetary health and personal health." Food, Kuhl argues, is an ideal test environment for this kind of AI precisely because it involves competing objectives that are genuinely difficult to satisfy simultaneously. A burger can be delicious or nutritious or sustainable - but can it be all three at once, without obvious compromise? That tension mirrors the trade-offs embedded in many of the larger problems AI might eventually help solve. The taste test Computational results are one thing. The real test was whether people actually wanted to eat what the AI designed. The researchers had five BurgerAI recipes professionally prepared and served to more than 100 diners in a blinded taste test at a San Francisco restaurant alongside a popular fast-food burger used as a benchmark. The results held up. BurgerAI's two variations of its Delicious Burger scored the same or better than the fast-food benchmark in overall liking, flavor, and texture. Its Mushroom Burger reduced environmental impact by more than an order of magnitude compared to a conventional beef burger. Its Bean Burger achieved roughly twice the nutritional score of the fast-food option. "AI did not just generate plausible burger recipes - it created burgers that real people enjoy," Kuhl said. "That may sound simple, but it means the model learned what makes food appealing to the human palate and was able to navigate a design space with near-infinite possible burger combinations to find real-world solutions." "We expected some trade-off between sustainability and consumer acceptance," Tac added. "But we found a burger with dramatically lower environmental impact could still compete with one of the world's most successful burgers." The bigger ambition Kuhl is clear that burgers were chosen as a test case, not as the destination. Food is a tractable, human-scale system with a large and accessible design space, real human feedback in the form of taste preferences, and measurable outcomes for both nutrition and environmental impact. It's a good place to prove that generative AI design works. But the same framework, the researchers argue, could be applied to problems with far higher stakes. Drug discovery involves navigating vast molecular design spaces in search of compounds that are simultaneously effective, safe, and manufacturable. Materials science involves finding combinations of elements and structures that satisfy competing physical and chemical requirements. Synthetic biology involves designing organisms or biological systems that achieve specific functional goals. "The burger is just the beginning," Kuhl said. "We see food as a model system for a much larger vision: AI as a partner in scientific and engineering discovery." "For centuries, food design has been a matter of intuition, experience, and trial and error. We are beginning to show that AI can transform food design into a quantitative science with applications in other important fields," she concluded. The first study is published in the journal npj Science of Food and the second one in Computer Methods in Applied Mechanics and Engineering. -- - Like what you read? Subscribe to our newsletter for engaging articles, exclusive content, and the latest updates. Check us out on EarthSnap, a free app brought to you by Eric Ralls and Earth.com.

Stanford scientists built an AI that can design healthier, greener burgers

The new system balances nutrition, taste, cost, and environmental impact to create better recipes. Artificial intelligence has already helped write code, discover drugs, and generate videos. Now, it's trying to make a better burger. Researchers at Stanford University have unveiled BurgerAI, a new AI system that designs burger recipes by balancing taste, nutrition, sustainability, and cost. The surprising part? In blind taste tests, diners liked some of the AI-created burgers just as much as, and in some cases more than, a popular fast-food burger. BurgerAI is designed to invent recipes, not copy them According to Stanford, BurgerAI was trained using more than 2,200 burger recipes to understand how different ingredients interact. Rather than predicting which existing burger someone might like, the model generates entirely new recipes based on factors such as age, nutritional needs, personal taste, and even sustainability goals. Recommended Videos The researchers say there are an estimated 1043 possible burger combinations, making it an ideal challenge for AI-driven design. To see whether the AI actually worked, the team prepared five BurgerAI recipes and served them to more than 100 diners in a blinded taste test. Two of the AI-designed burgers matched or outperformed a popular fast-food burger in overall liking, flavor, and texture. Even more impressively, one sustainable mushroom-based recipe delivered a significantly lower environmental footprint without sacrificing consumer acceptance. Lead researcher Ellen Kuhl says that's exactly the point. Instead of asking, "What burger is most likely?" BurgerAI asks, "What burger best satisfies these competing objectives?" In other words, the AI isn't simply predicting outcomes. It's inventing entirely new ones. Interestingly, this isn't really about burgers The funny thing is that BurgerAI isn't meant to revolutionize fast food. The burger simply serves as a proof of concept. The researchers believe the same AI framework could eventually help design everything from new medicines and biomaterials to sustainable manufacturing processes, where engineers must constantly balance competing goals rather than optimize for just one outcome. That's what makes this research so interesting. Most generative AI models today focus on creating content that resembles what already exists. BurgerAI takes a different approach by generating solutions that have never existed before and then validating them in the real world. However, the burger is just the beginning. If AI can successfully navigate trade-offs between taste, health, cost, and sustainability, it may eventually help solve far more consequential engineering problems than what's on the dinner menu.