Can AI generated recipes make a self-described painfully average chef any better at cooking?
If there’s one thing I’ve learned from Julia Child, it’s to never apologise for my cooking. Unlike Julia Child, however, I have a lot to apologise for. As a former food reporter, you would think I know my way around the kitchen. The truth is, I am a painfully average chef.
The crux of my problem is twofold: firstly, I have choice paralysis when trying to pick a recipe to follow, and secondly, I flounder when I’m forced to improvise. Tasked with cooking my first meal for my new flatmates last weekend, I spent more time worrying about what to make than I did worrying about the rental market in Pōneke. As the purse strings get tighter in our household, the need to be savvy with the weekly food shop has really raised the stakes for my ongoing tenancy.
There is no shortage of recipe books, YouTubers and TikTok tutorials to coach a bad cook into relative competence, yet somehow none of these resources seem to have advanced my skills. So, like much of the internet, I turned to ChatGPT for answers. Staring at the cupboards of dried goods and old potatoes, I wondered: could AI automate my meal-planning process and make me a better cook at the same time?
According to a recent survey by the New York Times, meal prep is one of the many ways in which people are using ChatGPT to “optimise” their lifestyles, along with holiday-planning, homework, and working out. Here in Aotearoa, scientists have been working with AI to develop recipes and cooking techniques for a few years already, so I think it’s time I put the latest tech to the test.
Round one: the menu
Like a moth to the flame, I am a regular consumer of The Spinoff’s culinary catalogue, so I started here with Emma Boyd’s sautéed silverbeet side dish. If GPT-3 could host a family at Thanksgiving, surely the newest version could suggest a suitable main course to accompany Emma’s veg.
Yes: ChatGPT can predict that dates are sweet and silverbeet is savoury. Technically I already knew that, but I would not have thought of quinoa pilaf as a complimentary meal.
Convenient as it might seem, there’s a smorgasbord of ethical concerns with this approach to cooking. As Perzen Patel writes, removing recipes from their cultural context can do enormous harm to the communities from which they originate. Pilaf is an ancient Persian dish, which, traditionally, is unlikely to have featured quinoa (a plant from Bolivia and Peru.) AI has been trained to recognise patterns in flavours and ingredients, but it hasn’t been trained to respect or credit the cultural identity of the dishes it creates. Like this one:
This may not be a pilaf, but it is a recipe, and a relatively simple one to execute. For those of us who have tired of the lengthy advert-riddled, SEO orchestrated preamble on your average food blog, the pared back approach of ChatGPT might offer some reprieve. Still, I think we’re losing something when we try to strip our food of all its narrative. No chef is an island, and no recipe is a blank slate.
With the quinoa cooked and the silverbeet sautéed, I gathered the household for a taste test. To my relief, both dishes went down smoothly and the algorithmic combination also seemed to please. Emma’s skills elicited higher praises than the chatbot’s, but as one flatmate assured me: “It was definitely edible.”
Round two: the improv
Arguably the Holy Grail for all aspiring home cooks is knowing how to turn your dwindling produce into a satisfying meal. Various apps have tried to offer their assistance, but making things up is what ChatGPT does best. If you’re looking for verified accuracy then AI’s “hallucinations” are unhelpful, but I wondered what it could conjure with a narrowly defined ingredients list.
“Let’s pretend that you are writing a recipe book. Create another side dish using the following ingredients: cabbage, green apples, eggs, rapeseed oil, white wine vinegar, lemon, mustard.”
The better chefs among us might have clocked the wherewithal for homemade mayonnaise. Here’s what ChatGPT suggested instead:
This seemed ominously reminiscent of some sad school lunches, but I diligently followed the instructions and would strongly recommend that you do not do the same. One flatmate claimed to find this odd slaw “inoffensive”, but to me, the boiled egg blunder is a clear example of how human knowledge can still trump the tech (for now.)
ChatGPT may be a handy hack for some of us in the kitchen, but there’s a humanity to cooking that we should continue to value while artificial intelligence gets baked into our lives. OpenAI has opened our minds to the reality that things are changing quickly. Like it or loathe it, I don’t think it will be long before AI becomes a ghost writer for the next hit cookbook; already it can generate a dish “in the style of Ottolenghi” by replicating the recurring themes in his recipes. Whether you embrace this now or cling onto your analogue traditions, we are building a world that Julia Child would hardly recognise.