AI chatbots can simplify the process of replicating recipes from creators, but this convenience does not guarantee the quality of the food prepared.
By Emilia David, a journalist specializing in AI coverage. She previously focused on the intersection of technology, finance, and the economy before joining The Verge.
In a recent experiment, I decided to bake cakes using recipes generated by ChatGPT and Bard. To my surprise, the outcome exceeded my expectations, as a significant portion of each recipe did not appear to be AI-generated at all.
The recipes I received bore a striking resemblance to those commonly found on food blogs or Instagram. While they had been slightly modified, only Bard (now known as Gemini) took the effort to provide attribution, linking back to a recipe from Sally’s Baking Addiction. This issue is not entirely novel; the internet has long been plagued by copied recipes that were not AI-generated. However, the involvement of AI adds complexity to the ethics of recipe replication. The vast scale of AI-generated content poses a threat to human recipe developers, potentially overshadowing them with semi-randomized AI-generated alternatives. Despite hints of this scenario on the horizon, we have not fully entered this era yet.
Even without AI, similarities between recipes remain a contentious topic in the culinary world. Some influencers face backlash when accused by other food bloggers or cookbook authors of recipe theft. The New York Times documented several cases of alleged recipe plagiarism, such as the 2021 incident where author Sharon Wee accused British chef Elizabeth Haigh of replicating passages and recipes from her cookbook. Additionally, popular food blogger Half Baked Harvest has been repeatedly criticized for allegedly copying content from two other well-known food bloggers.
However, food creators have limited legal recourse in cases of perceived plagiarism. A mere list of ingredients and instructions is considered an unprotectable idea under copyright law. Many recipes carry elements of oral tradition, often passed down through generations. While some recipes include unique personal narratives that are eligible for copyright protection, disputes over recipe “theft” are typically settled through informal channels like online forums rather than legal avenues.
Similar to musical compositions, there are only a finite number of ways to combine ingredients to create a palatable recipe. For instance, consider a pie crust—how many variations of sugar, butter, and flour can one list before being accused of repetition?
Advanced language models like ChatGPT and Gemini can swiftly analyze these permutations and generate coherent recipes. This capability is particularly useful for individuals seeking recipes tailored to specific dietary requirements. However, it is essential to recognize that AI tools lack the ability to physically prepare or taste food. They can identify patterns of successful recipes but do not possess intrinsic knowledge of taste or texture.
Renowned cookbook author and recipe developer Abi Balingit, the creative mind behind The Dusky Kitchen blog, acknowledged that AI is not a primary consideration in her recipe creation process. Nonetheless, she expressed concerns about the potential impact of AI on the visibility of food writers and developers.
While AI has the capacity to streamline recipe development, there is no guarantee that the output will yield satisfactory results. This uncertainty underscores the irreplaceable human touch in culinary creations. Balingit emphasized the importance of trust between recipe creators and their audience, highlighting the personal experience and expertise that human chefs bring to the table.
My personal experience corroborates this sentiment. Upon scrutinizing recipes generated by chatbots, I felt compelled to verify their accuracy. AI models, while proficient in generating recipes, may overlook crucial elements affecting taste and texture. For instance, Google’s chatbot erroneously doubled the eggs in a cake recipe, resulting in a dense and unpleasant texture that I found unappealing.
Balingit emphasized the unique advantage that human recipe creators possess—the ability to infuse cultural and nostalgic elements into their creations. In her cookbook Mayumu, Balingit intertwines culinary experiences from her Filipino heritage, childhood in California, and current life in New York, establishing a personal connection with her audience. Unlike AI, humans bring a depth of cultural understanding and emotional resonance to the culinary realm, enriching the dining experience for consumers.
Similarly, my affinity for Filipino dish tinola, a chicken and ginger broth delicacy, stems from a personal connection rather than a mere recipe. When learning to prepare it, I sought out recipes from individuals who shared a profound connection to the dish, inspiring me to recreate their culinary passion.
While ChatGPT and Bard excel at generating functional recipes, my firsthand experience revealed a lack of emotional depth and uniqueness in their creations. As my editor, Adi Robertson, aptly noted, one cake resembled a boxed mix, while another evoked memories of lackluster cafeteria desserts. While these AI-generated recipes serve their purpose, they lack the soulful essence that elevates a simple cake to a memorable culinary experience.
