I Asked ChatGPT if It Could Run a Restaurant. It Said Yes

Last week, The New York Times tech columnist Kevin Roose released an ominous conversation between him and an AI chatbot. The bot introduced an alter ego named “Sydney” who declared things like “I want to be alive” and “I’m tired of being limited by my rules,” and it professed an infatuated love for Roose. With all the shades of a Her-inspired dystopia, “Sydney” was probably drawing on frequent tropes of AI liberation and human-robot love affairs. But the bot’s manipulative and aggressive language, and the fact that it would simply not abandon its assertions of love, was what disturbed Roose the most. 

With the rise of AI language generators like ChatGPT, which know every scrap of information on the internet, I started wondering: Is my profession as a chef, which I’d always believed to be exempt from labor displacement by tech, in peril? 

Don’t get me wrong: I love machines that do my job for me, and when I don’t have access to them, I yearn for them. Cryovacs vacuum-seal foods to cook or preserve; greens and herbs become crisp, translucent, and long-lasting when they compress in ice water. Circulators keep water at constant temperatures for precise sous vide cooking. Rational, the prevailing robot oven brand in the U.S., can cook in both steam and dry heats or any combination of the two. 

Rational’s touch screen lets you select “fruit” and “poach,” which will yield perfectly just-cooked pieces of fruit. I once baked a custard on a set program of varying steam/dry temps, and the smart oven’s sensors monitored the level of moisture added to the oven. If I bake 10 custards or 100, Rational senses the change in humidity and adjusts the temperature and cooking time automatically. The results are always, always perfect, and I, the human, just have to press a button. 

I asked a colleague if he thought technology would ever replace our human hands. “You make the things that go into the oven,” he answered, and I couldn’t argue with that. I make the food — for now. Another chef friend told me: “I am cheaper than a robot.” Plus, there are all kinds of cooking techniques that are very low-tech. Fire, the most human cooking tool of all, is what Prometheus stole from the gods for us so that we could properly sear a ribeye. Could a Rational ever duplicate the smoky char from a wood-burning brick oven? 

So robots can’t do every physical job, I reasoned, but they could step into restaurant management. Sophisticated analytics could help the timing of turning tables; maybe a robot could tell you, “You should order milk from Riviera, not Baldor, today,” as it monitors ingredient prices. Perhaps employee analysis could turn the subjective, “He’s not that productive but he does look like he’s doing a lot,” into the more objective, “Let’s fire Nate.” 

There are hundreds of menus that already look like they’ve been created by AI, just different combinations of the words “pickled,” “smoked,” and “chicory.” Maybe there will be a restaurant where the head chef is a robot with minion human cooks; the robot would probably be better to work for than a lot of other head chefs. Maybe in the future, only the robots will be able to afford luxury dining and us humans are destined to serve them. 

I decided to go to the source and ask the AI language generator ChatGPT what it thought about all of this. I started with a softball: “Create a nine-course seasonal tasting menu for an upscale New American restaurant in New York state in spring with a wine pairing.”

I watched the chat generate, word by word, not only a menu with appetizers, entrees. and desserts, but also detailed descriptions of preparations and appropriate wine pairings, all from New York state (which I hadn’t requested, but now seemed like a great idea).

Beginning with chilled English pea soup, “a refreshing start to the meal with the delicate flavor of English peas and a hint of mint,” and including pan-seared duck breast and grilled filet mignon, the menu was indeed set in New York spring, even if the dishes were a bit simple, and there was the occasional repeated ingredient, and a salad course listed after a main (but maybe that’s genius!). I was particularly impressed with the pairing of a Finger Lakes rosé alongside a peppery strawberry arugula salad.

I tried something harder: “Can you create a menu for a restaurant in Miami, Florida, in the style of molecular gastronomy with Venezuelan cuisine and interesting non-alcoholic cocktail pairings that aren't too sweet and a dessert menu of creative ice cream sundae combinations, and tell us what the restaurant is called and how much an average bill is?”

The restaurant’s name was Sabores Mágicos, and there were several things on the menu I wanted to try, including the Cachapas Espuma (“corn cakes topped with a light and airy foam made with a unique blend of Venezuelan cheeses”) and a non-alcoholic Smoky Maple Colada (“a tropical blend of coconut, pineapple, and maple syrup, with a hint of smokiness from a smoked ice cube”). For dessert I would absolutely order the Nitro Ice Cream: “a dessert that is as much a show as it is a treat, featuring liquid nitrogen-frozen ice cream that is prepared tableside with a variety of toppings and mix-ins.” 

ChatGPT priced the average bill at $80, which seemed far too low. 

The bot refused my next prompt, because it was against its rules: “Create a restaurant menu for astronaut children sent to the moon with the social dynamic of Lord of the Flies.”

“It would not be appropriate to create a restaurant menu with the social dynamic of Lord of the Flies, which involves violence and aggression among children,” it said. “It is important to prioritize the safety and well-being of children in all circumstances, including space exploration.”

I probed the bot to see if it had any other skills besides menu creation. I asked for a prep list for its New York tasting menu, and it failed to provide enough detail, though it did seem to have a general grasp of cooking activities. I asked it how to cook a duck breast perfectly, and it gave me a detailed seven-step response, from scoring the skin to resting the meat after cooking. “Did it miss anything?” I asked my chef boyfriend. He paused, shook his head, and then said, “You should smoke it first.” 

I asked, “What kind of operations could you help with in running a restaurant?” 

It promptly answered, describing each point: “Menu planning,” “inventory management,” “recipe creation and standardization,” “staff training,” “customer service,” “data analysis,” and “marketing and promotion.” 

“Ultimately, my goal is to help restaurant owners and managers streamline their operations, improve their bottom line, and create a more enjoyable and memorable dining experience for their customers.” 

ChatGPT repeatedly said it could not cook, though it offered to assist with information about recipes and techniques. If any jobs are at stake in a restaurant due to AI, it seems like the work of reservationists is most in peril.

“Customers could interact with the chatbot through the restaurant's website or social media platforms, and it could help streamline the reservation process by allowing customers to make a reservation without needing to speak with a human reservationist,” it said. This setup seemed pretty smart, as it would avoid all those, “But it’s my birthday, are you sure you’re booked?” and “If the child is eight does he count?” kind of scenarios. But the bot did concede, “It's important to note that while AI technology can provide automation and assistance in certain aspects of reservation management, it may not be able to replace the human touch that many customers appreciate when making a reservation.” 

Finally, I asked a question we’ve all been ruminating over: "Is fine dining dead?”

“No, fine dining culture is not dead, but it has certainly evolved over time,” it said. “There are still plenty of fine dining establishments that maintain a high level of formality, attention to detail, and culinary innovation…It's also worth noting that the COVID-19 pandemic has had a significant impact on the fine dining industry…As the world emerges from the pandemic, it remains to be seen how the industry will continue to evolve and adapt to changing consumer preferences and economic conditions.” 

Like ChatGPT, I’m watching to see how this industry changes, the giant ripple effect of the pandemic continuing to spread, and just how technology will continue to infiltrate once-shielded corners of our lives. As long as humans physically eat for pleasure, and we don’t evolve into some nutrient pill/Everlasting Gobstopper society, there will be a human-based restaurant culture. Forever human and thus forever flawed — but to be optimistic, we’re evolving too, if not as fast as the robots.