Test kitchen and blog post by Eileen Reiner.
For the past week I have been tinkering in the kitchen to put chat GPT to the test and test cook a variety of AI generated recipes as part of the Center’s MVP X FFF project.
The work explores if and how digital tools and AI can aid in the creation of viable and desirable recipes from novel ingredient combinations. It features the MVP x FFF food computer that harmonizes the rhythms and flavors of alternative protein products (MVPs) and food forest produce (FFFs). The computer remixes MVPs and FFFs based on user preferences, creating novel and delicious recipes that are nutritionally-complete and ecologically-minded.
I was curious to see if the current prompts we had fed Chat GPT were working, and if the AI had already developed the culinary expertise and creativity to create great dishes, or if we needed to prompt and support it some more…
Dish 1: “Cluck ‘n Chestnut Surprise Bites: Savory Candyland Delight!”
Playing with the food computer menu options, I selected red berries and chestnuts as my food forrest foods in season, and the vegetarian butchers ‘what the cluck’ chicken pieces as my alternative protein. I was curious to see how chat GPT would taste pair the three and how many other surprising and complimenting ingredients it would combine them with.
I chose a mixed salad for my typology – which seemed to give ample room and opportunity for an array of different flavours, and I chose it to be ‘in the style of’ savoury pick-n—mix finger food to bring a new experiential element to how the dish is consumed.
The first few salad recipes chat GPT gave me sounded tasty but pretty safe – perfect for the suspicious eater, but I felt like I wanted to be surprised and inspired. So I gave Chat GPT a little nudge by adding ‘something surprising and fun’ to the prompt which seemed to do the trick! an array of options from wonton candy land salad parcels, to feta and raisin sprinkled Caesars arose. Finally I landed on “Cluck ‘n Chestnut Surprise Bites: Savory Candyland Delight!”
I managed to find everything in my local supermarket, and set to work. The salad was easy and simple to put together (mainly chopping) However, I did allow myself a little creative autonomy during the process where I felt the recipe needed tweaking. Chat GPT hadn’t told me how to prep the Kale. I didn’t really fancy ribbons of raw kale in my salad so I massaged it with olive oil and salt and put it in the air fryer to add a satisfying umami crisp to the affair.
Although advertised as such, I was still surprised by the similarity of what the cluck chicken strips to chicken. fried up much the same as chicken breast, about 5 mins on each side – until they were nicely golden brown, they possessed a very similar textual experience.
The tricky bit was assembling the parcels. I feel like this is heavily dependant on the type of salad leaves you have. Large, soft, yet not too fragile seems the way to go. In retrospect I think baked or boiled cabbage leaves could have been the one here.
The result was pretty much as promised: Surprising savoury candyland bites. The the tart ‘pop’ of the berries worked really well to cut through the grease of the’ what the cluck’ chicken and the soft, sweet nutty texture of the chestnuts. The mint leaves and snap of the watery lettuce really freshened it up, as the crispy fried kale added a salty twist.
This particular dish was what I was hoping the role of chat GPT would deliver. Something that was rooted in an understanding of the existing salad format and taste paring data – yet offering something unexpected and beyond my cultural imagination. Having the specific list of seasonal food forrest ingredients, and Alternative proteins to chose from also gave me some interesting boundaries in which to play with new taste combinations.
So was this just a lucky dish, or does is Chat GPT generally a success in the kitchen? I thought Dish 2 would be a good opportunity to find out.
Dish 2: “Red Currant Vegan Ice Cream Cone: Crispy Delight Edition”
This time I wanted to the try red berries with something sweet. Combining them with oatly oatgurt and Veggs (egg substitute), I put Chat GPT to the culinary test with some vegan pastry baking, and homemade ice cream.
I was pretty dubious about this dish, as vegan pastry is an art form. Having thousands of years of experience around the way that egg behaves – vegan baking is relatively new. Since my mum is a chef that has experimented in that field I have seen that it takes trial and error, especially around the structural integrity of things – which was causing me concern around the ice cream cone – for which chat GPT simply instructed me to bake a pancake and wrap it around a cone shape until it was cool.
I found all the ingredients at my local supermarket – with one substitution – ‘oggs’ instead of ‘veggs’ (which seemed very similar).
The ice cream itself was pretty successful! The Oatgurt was creamy and rich – a perfect base to mix the vanilla extract and maple syrup. The red berries were fresh and tart which complimented the sweet syrup nicely. I would have probably gone for nuts over pretzels, but the salty element was exciting.
The cone however was extremely tricky and time consuming. Although relatively easy to get into the right shape it didnt hold it’s structural integrity when the support came out. I experimented with various thicknesses of the pancake at various temperature, for varying time-frames, and even tried putting it in the freezer once set. Vegan baking is quite a fine art that it seems in this case chat gpt still needs to finesse. Back to the prompt board with that one I think!
I really enjoyed the creative process of pairing the seasonal forrest food flavours with the list of alternative proteins and was inspired by the new flavour combinations – even those I was weary of! Chat GPT was fun to work with in this arena and encouraged me to try new things, although It definitely has its oversights! These seemed to be more around the cooking and structural integrity of things rather than the flavour pairing itself. Although I found Chat GPT to be pretty impressive in creative writing, I feel it still needs a little more support and lots of clear prompts in the kitchen for now.
This project is developed as part of the Hungry EcoCities S+T+ARTS Residency which has received funding from the European Union’s Horizon Europe research and innovation programme under grant agreement 101069990.
