Last year, as an experiment, I baked a cake from a recipe generated by ChatGPT and Bard. It went better than expected. That’s because many of the recipes didn’t look like they were created by an AI at all.
The recipe I got looked eerily similar to recipes I’d seen on food blogs and Instagram. They were slightly modified, and Bard (now called Gemini) was the only one who bothered to link to the Sally’s Baking Addiction recipe and provide the source. In some ways, this is not a new problem.I’ve seen too many copies Non– AI recipes all over the internet. But with the advent of AI, the ethics of recipe copying have become more difficult. Its sheer scale threatens to create a world where a human recipe developer is shut out by semi-random competition generated by his AI. But although there are signs that this world is coming, we are not yet in it.
Even without AI involved, recipe similarity is a very sensitive issue in the food world. Some influencers find themselves under fire when other food bloggers or cookbook authors claim they stole their recipes. new york times We have reported several cases of alleged recipe copying, including a 2021 incident in which author Sharon Wee accused British chef Elizabeth Hay of copying texts and recipes from her cookbook. Plus, there’s been some online hate for popular food blogger Half Baked Harvest. He has been accused multiple times of copying two other popular food bloggers.
But if a food creator feels someone has stolen their work, there are few legal recourses. A simple list of ingredients and instructions is treated as an idea that cannot be copyrighted. Many recipes have elements of oral tradition. Many are passed down through families. Some include long personal introductions; teeth The biggest backlash against recipe “plagiarism” is often the daily hate threads on Foodsnark Reddit, where the world of recipes is primarily ruled by etiquette rather than law.
And just like chords in music, there are only so many ingredients that can be combined to create a passable recipe. Take the puff pastry. How many permutations of sugar, butter, and flour can you write down before someone claims you’re just repeating the same recipe?
Large language models like those that power ChatGPT and Gemini can take these permutations, parse them faster than humans, and come up with fairly reliable recipes very quickly. As a result, finding recipes for specific meals is often touted as a great use for chatbots. On the other hand, of course, AI tools cannot actually prepare or eat food. They don’t actually “know” if a recipe will work or not. It’s just that it fits into a pattern of recipes that work.
Abi Balingit, a cookbook author and recipe developer who runs the blog The Dusky Kitchen, said she doesn’t think about AI when creating recipes. But she worries it will affect where food writers and developers can feature their work.
“There’s a gradient between what’s good and what’s not. AI doesn’t make recipe development worse, because there’s no guarantee that what it comes up with will work,” Balingit says. “But given the ephemeral and unstable nature of media, I worry that someday publications will rely on AI rather than recipe developers and chefs.”
As many companies have discovered the hard way, generative AI still sometimes hallucinates and makes up things that are physically impossible. Grocery delivery platform Instacart has partnered with OpenAI to run his ChatGPT on recipe images. The results ranged from hot dogs with tomato filling to a salmon Caesar salad that somehow created a hybrid of lemon and lettuce. The proportions were off— washington post As pointed out, the size of the steak in the Instacart recipe can easily feed more people than you planned. buzzfeed We also announced an AI tool that recommends Tasty-branded recipes.
Balingit added that when people read someone’s recipe or watch them make a dish, they develop a certain level of trust. We have the expertise to have cooked and tasted the food.
This explains why I immediately felt the need to revisit the recipe from the chatbot. AI models can still hallucinate and grossly misjudge how ingredient amounts affect taste. For example, Google’s chatbot inexplicably doubled the eggs, making the cake moist but at the same time thick and sticky, which was not to my liking.
Balingit argues that one of the benefits of being a recipe creator is the human connection.her cookbook Mayumu is filled with dishes that recreate his imaginative journey through his parents’ migration from the Philippines to the United States, his childhood in California, and his current life in New York. She said AI has no cultural or nostalgic ties to food and meals, which are historically personal things that people share with others.
I felt the same way. I never cared who first came up with the idea of boiling chicken in ginger soup, but it’s tinola, which remains my favorite Filipino dish. When I first started learning how to cook (thanks to my dad kicking me out of the kitchen as a nuisance), I searched the internet for “authentic” recipes. I decided to follow the work of people who had a strong personal connection to cooking and who inspired me to create the dishes they wrote about with passion.
ChatGPT and Bard can generate functional recipes. I know that because I followed them. But as someone who baked those cakes, I knew it was sobering and common. My editor, Addie Robertson, likened one to a cake mix in a box, while another reminded me of a sad cafeteria cake. Indeed, that’s spot on. Yes, it’s chocolate cake. But there’s more to the cake than that.
