A.I.: It’s What’s for Dinner

Cooks, chefs, and enthusiastic diners have long debated whether cooking is an art or a craft, a form of creative expression or the deft execution of learned skills. But whichever side of the fence they land on, for the most part, food people agree that the ability to produce delicious things requires a relatively high level of perception — the capacity to recognize when a pan of caramelizing onions reaches the perfect shade of auburn or to smell when roasting vegetables teeter across the line between underdone and just right. In short: Good cooking requires good instincts in the kitchen.

Or it used to at least. A new Silicon Valley startup called CloudChef is using software and artificial intelligence to digitize cooking — not recipes, but the actual process of preparing specific dishes. In the view of CloudChef CEO Nikhil Abraham, the company’s product aims to simulate the kind of kitchen instincts that can take talented cooks and chefs years to perfect. “Our problem statement was: We need to codify the intuition of the chef,” Abraham says.

In a way, the execution might not be as complicated as you’d think. Stepping inside the company’s Palo Alto kitchen, it’d be relatively easy to overlook the technological enhancements that differentiate this commercial kitchen from any other. Aside from a dry pantry where barcodes replace pieces of painter’s tape with ingredient names scrawled across them, and the addition of desktop computer-sized monitors next to the hulking industrial range, there’s not much that separates a CloudChef kitchen from any other. At least, not by looks alone.

CloudChef

Through a combination of software and hardware — as in basic kitchen equipment that’s been souped up with sensors and cameras — CloudChef can record someone cooking a dish one time, then turn around and produce what the company calls “a machine-readable recipe file.” The data capture not only includes video, but also information from infrared and thermal sensors, scales, and other equipment that allows the software to measure temperature and weight. The result is a set of digital cues delivered through the enhanced kitchen equipment that guide CloudChef kitchen staff, regardless of culinary skill, as they stand in front of a burner with a pan and recreate the dish.

Already Bay Area diners can order food from the CloudChef website or through third-party delivery apps, and someone who doesn’t know a scallion from a shallot will cook up a Michelin Guide-worthy plate of chicken pulao — or at least, that’s what the company advertises.

For example: On CloudChef’s website, a customer can order Vellore Chicken Curry, a dish from Bay Area chef Srijith Gopinathan made with coconut, chiles, and a fragrant spice mix. When ordered online for pickup at the CloudChef kitchen the entree costs $25, about 20 percent cheaper than the $31 the entree costs at Gopinathan’s Michelin Guide-listed Palo Alto restaurant Ettan. The person gathering up the ingredients in the company’s kitchen might have no understanding of the difference between turmeric and coriander, but that doesn’t matter because the CloudChef software instructs the cook to pull spices off a rack and weigh them out based on a system of barcodes, no reading, tasting, or knowledge of spices required.

A different person will then collect the ingredients and carry them over to the kitchen’s tech-enabled range and cook the dish by following instructions provided via video monitor. There’s no need to understand when the chicken achieves doneness or when the curry reaches the ideal thickness; the software does that using data from sensors that measure the thermal temperature of ingredients in the pan and the relative weight of the food as it cooks down. When the sauce reaches the appropriate thickness, a sound alerts the cook and a message appears on a screen telling the worker to remove the pan and food from the heat.

Unlike a traditional restaurant kitchen, which usually requires an experienced cook to work each station, a CloudChef kitchen only needs one skilled worker to operate: a kitchen manager who understands the computer interface. With a kitchen manager on-site to troubleshoot any potential technological hiccups, the rest of the staff can fulfill orders despite having never worked in a kitchen before. In fact, Abraham says most of the staff at the company’s Palo Alto kitchen are gig workers hired through companies like Instawork and Qwick. They often come into the kitchen with little to no cooking experience but can start preparing food in the CloudChef kitchen after five to 10 minutes of training.

In press materials, the company positions itself as “Spotify for food.” Abraham and co-founders Atish Aloor and Mohit Shah say CloudChef provides well-known chefs from around the world with the ability to bring their food to vast new audiences. “They already have brand recognition,” Abraham says. “But they don’t have distribution.”

CloudChef

The team envisions a world where CloudChef operates kitchens in markets all over the United States, from which customers can order off a menu stacked with top hits from some of the most recognizable chefs in the world. Abraham says he understands the infrastructure needed to make that a reality would require a massive amount of capital — and time. But he says because CloudChef’s software only requires adding some additional hardware to standard kitchen equipment, he’s hopeful they could accelerate the growth trajectory by piggybacking on existing commissary kitchens.

For now, the kitchen in Palo Alto offers dishes from Gopinathan, who earned two Michelin stars while cooking at Campton Place in San Francisco, and chef Thomas Zacharias, who cooked at three Michelin-star Le Bernardin before opening the Bombay Canteen in Mumbai. The founders, who are graduates of the Indian Institute of Technology, are launching with a focus on Indian cuisine but plan to expand their recipe repertoire as they grow, not unlike how Amazon initially focused on selling books before becoming the go-to online store for pretty much everything, Abraham says.

Culinary talent, who the company calls “Creators,” can request to record recipes through the CloudChef website. Abraham says the company rotates available recipes on a weekly basis, monitoring customer demand and dropping dishes that don’t seem to interest diners. In a similar fashion to how Spotify pays artists royalties every time someone streams a song, chefs get a percentage of the money each time a diner orders their dish. The royalties range from three to 15 percent per order, depending on the recipe creator’s “existing distribution,” in Abraham’s words. Chefs with little to no name recognition require more marketing effort on the part of the CloudChef team, he says, so they’ll see lower royalties. On the other hand, a big-name chef with the potential to draw customers to the website on their own would receive royalties on the higher end of the company’s spectrum.

CloudChef

What’s possibly even more intriguing is what CloudChef plans next: the ability for anyone to upload a recipe, have it processed by the company’s artificially intelligent software, and cooked by staff in one of the company’s kitchens. The company leverages OpenAI, the generative AI tool behind ChatGPT, to make it possible. In theory that means you could give the company your grandma’s recipe for pancit or borscht or marinara sauce and, thanks to artificial intelligence, it’ll discern what she meant by “cook till done” and make appropriate adjustments and substitutions based on the availability and relative quality of the ingredients at hand.

It’s not hard to envision how this particular confluence of food and technology could change the way we eat down the line. Ostensibly it could facilitate the ability for people to experience foods from parts of the world they may never have the chance to visit. And unlike when a chef returns from a faraway place and recreates a dish through the lens of their own experience, CloudChef allows anyone in its kitchen to exactly duplicate the recipe. On the other hand, it escalates the ongoing conversation around ownership of recipes to new heights.

At this point, however, those in the restaurant industry seem less worried about the ramifications around intellectual property rights and more skeptical that the technology has useful applications for chefs and restaurant operators at all. John Park, managing partner of San Francisco-based Brick x Brick Hospitality Group, runs a diverse portfolio of restaurants and bars in the city including Novela, Kaiyo Cow Hollow and Rooftop, and the new Cavaña. He’s also working on expansion plans for locations in other cities down the line. For Park, technology that might make it easier to train staff in satellite cities could be helpful — consistency, he says, is one of the biggest challenges for operators as they open additional stores. But with the bar and restaurant industry’s notoriously slim margins (the average restaurant profit margin falls between three to five percent) he’s doubtful restaurants could afford to implement CloudChef’s tech. Even if they could afford it, he’s still not certain untrained workers in an AI-enabled kitchen would be a good thing for diners. “I think it would fall short,” he says, referring to the quality of the food.

Chef David Barzelay of two-Michelin-starred Lazy Bear in the Mission District says he, too, can appreciate the potential benefits of having more data available in the kitchen. He’s intrigued by the idea of being able to refine a recipe through the use of data from thermal sensors and even artificial intelligence. “There’s nothing wrong with that,” he says. “Recipes would be better with a few more data points.” But at the same time, he’s not convinced that it’s really, truly possible for a computer — even an exceptionally smart one, even one that can pass the bar exam, which, for the record, Barzelay did too — to acquire the same level of culinary intuition as a talented and skilled human chef. The quality and flavor of fresh ingredients vary day to day, he points out, and while cooks and chefs always hope to achieve perfection in every dish they create, perfection isn’t a stationary target in the kitchen. “Perfect means it comes together with harmony and balance and synergy,” he says. “There’s a whole lot of art and skill and technique involved, and there’s no substitute for having done something enough times to develop muscle memory.”

Abraham says no one needs to worry about a dystopian kitchen full of robot cooks. The team believes in what it calls a “co-botic” cooking model, where humans lean heavily on technology but are still fundamental to the cooking process. Still, if CloudChef’s tech spreads more widely, it could usher in a future where just about anyone can be a cook, but the job looks terribly different than it does today. CloudChef’s success is predicated on the idea that siloing kitchen workers into specific tasks such as measuring ingredients or stirring sauce in a pan until a computer tells them to stop will lower the barrier to entry in cooking and make good food more accessible to more people. But looking at other companies that use a half-robot, half-human workforce — think, Amazon warehouses where robotic arms could soon replace humans who say they’re being treated like robots already — the argument seems dubious at best. Nevertheless, Abraham remains adamant the robots aren’t coming for your kitchen. At least not yet.

“If robots knew how to make food, we’d already all have really delicious food,” Abraham says.