Synthetic Recipes

A.I. Food on IRL Plates

It seems like there is a new breathless headline every day about artificial intelligence (A.I.) and how it will completely change our lives in the not-so-distant future. Every industry is scrambling to figure how they can both exploit the emerging technology and insulate themselves from becoming obsolete. Food is no different. 

Chefs, food writers, and food personalities all want to be seen as earlier adopters of new tech so they can make headlines, get interviews, and ultimately make some money. Whether or not the tech makes any sense doesn’t seem to matter that much. It happened first with NFTs. Chef Tom Coliccchio minted his CHFTY NFTs and got world-wide headlines. Sure, none of us can tell how these odd pizza drawings will really do anything that a membership card couldn’t but that doesn’t matter. Chefs are teaming up with tech bros to create fractional cookbooks where individual recipes are treated like NFTs and purchased individually to create a collection based on the tastes of the individual buying them. Gourmet NFT, who boasts Chef David Skinner as a founder, promises to take chefs “from the butcher block to block chain.” Whatever the fuck that means. But in this news cycle, it’s all about A.I.

a slice of pizza that delivers pizza? Isn’t that technically trafficking?

The New York Times published an article late last year about A.I. generated recipes under the headline “Can A.I. Write Recipes Better Than A Human? We Put It to the Ultimate Test.” Since I write recipes and talk about food for a living, this was forwarded to me by almost everyone I know. The article ultimately comes to conclusion that the recipes aren’t very good *yet*. Reporter Mallory Arnold came to a similar conclusion this year for Outside Magazine. There are new chatbots and A.I. systems coming out all the time. I see recipe generators backed by A.I. in my targeted ads on Instagram. Clearly, this isn’t going away so I had to check for myself. I used the recipe generator tool over at MealPractice to see if it could help me come up with some ideas for the new bbq I got as a wedding gift. Using their very limited categories, I chose my inputs: vegetarian, whole foods, BBQ. 

The program then had me select between 3 recipe titles to see which I would like it to generate for me: a portobello burger (kill me), BBQ jackfruit tacos (interesting but we’ve already done it), and smoky grilled tofu kebabs (skeptical). I chose the kebabs because as someone who’s been vegan for 2 decades, I know that most people don’t know how to make tofu delicious. Could A.I. be better than most omnivore chefs? No.

This recipe is a certified mess. The header claims that the tofu will have a crispy texture after it’s time on the grill. One look at the ingredients and you can tell there is nothing in there, let alone the grill, that is going to make that tofu crispy. You’re just going to end up with bland tofu basted in bbq sauce. I didn’t even bother testing it. Why waste food?

This is one of the major problems with A.I.: it’s only as good as the data it pulls from. If most of the internet doesn’t know how to make delicious tofu, the A.I. that trawls it for information won’t either. Cookbook author @niksharma over at

The Flavor Files

pointed out to the New York Times how this allows A.I. to have so many biases hidden in it content as it draws from predominately Western style recipes written in English. We aren’t going to get the best tasting food possible by doing that. Not even close.

an ai generated image of the recipe, feels like I’m having a stroke

These A.I. generated recipes don’t worry me as a recipe creator. Despite appearing like plausible recipes, the proportions are off and the flavors are fucked. But that’s where the real danger of A.I. in food media is: noninformation. These recipes aren’t bad because they were written by artificial intelligence, but because there’s not a lot of intelligence in these tools. Rubbing spices on tofu, throwing it on the grill, and expecting something crispy is dumb as hell. But it looks real. It reads like a real recipe, but it isn’t. It’s the representation of one. It’s synthetic. A.I. recipes are like the spam calls and phishing scams that clog up our modern lives. They’re noninformation that exhausts us as we look for something real, something useful. They’re words strung together in the style of a recipe, but those words bear no relation to their real-world representations. They aren’t created to be successful in a real kitchen. They’re created to look like a real recipe. But their existence, and coming ubiquity, will make finding trusted recipes online that much harder.

ai just churning out bullshit

Currently there are no requirements for recipes, or any writing, to disclose whether it was written by a person or some type of computing system. As our digital ecosystem gets increasingly filled up by this kind of fluff, we should consider how these synthetic recipes might impact our eating habits. Cooking is grounded in real life. People cook to feed themselves and their families, not as an intellectual exercise. Most cooks can’t glance over a recipe and gauge its likelihood for success. Most people are just trying to make dinner. If people new to cooking use synthetic recipes that keep failing them, they’re going to give up on cooking. Full stop. They aren’t going to take the time to learn how to distinguish between a real recipe and A.I. They’re going to order food. We are already losing cooking skills as each generation spends less and less time in the kitchen. We shouldn’t lose recipes too. Plus, people can’t afford to cook a whole Thanksgiving menu like The New York Times did just to see if the recipes will work. Dinner has to hit the table at 6:30 pm so you can’t just cook whatever a generator spits out. There has to be trust in the system. There’s a reason people like Ina Garten have such a die-hard fanbase: her recipes fucking work.

Artificial Intelligence absolutely has a place in food, just not currently in recipe development. The Artificial Intelligence Institute for Next Generation Food Systems (AIFS), founded in 2020, is a project between UC Davis, UC Berkeley, Cornell University, the University of Illinois, UC Agriculture and Natural Resources, and the U.S. Department of Agriculture’s Agricultural Research Service. They are doing some of the most cutting edge and thoughtful scholarship on real world applications of A.I. in our food system. One of their current projects is using deep learning, a subset of A.I., to identify the nutritive value of a meal and to make health predictions based on photos of food. That’s where A.I. is used best right now, at approximate knowledge. I’d love to be able to take a photo of my meal and have an app tell me if I’d had too much salt, or not enough fiber, in my diet any given day. But an approximation of a recipe for dinner? Naw, I’m hungry for real.

---

Thanks for joining us here in The Broiler Room. Do you use A.I. in your life or job? Cooked an A.I. recipe and like it? Let us know in the comments. If you want to support real recipe development and get recipes delivered right to your inbox every Sunday, become a paid supporter for just $5 a month. As a paid subscriber, you’ll get access to our chat forum where you can ask us questions about recipes, or whatever, plus a huge back catalog of recipes. Tomorrow’s recipe is Grilled Corn and Eggplant Soba Noodles that are perfect for summer.

We’d love to see you over there. Don’t forget to follow us on social too. Instagram: @badmanners @mdavisfood @shotbyholloway