AI can help chefs in the kitchen, but you had better double-check its work

Chef Ian Ramirez, founder of Mad Honey Culinary Studio, is known for his creativity on the plate as a college chef in the past and lately with his own company. For a few years now, he’s been experimenting with how artificial intelligence (AI) might be able to boost human intelligence (HI) when working on recipes, menu cycles and more.

“A lot of people have been making fun of AI in the past year,” Ramirez said during his appearance at the Menu Directions conference, presented by Foodservice Director, showing examples of the weird, uncanny fingers AI can conjure, and its reaction to the prompt “salmon jumping in a river,” which showed fully cooked salmon fillets cheerfully swimming upstream. Another prompt, “How many rocks should I eat?” led to AI’s recommendation of “everyone should eat one small rock per day.” Or, when asking AI for advice on helping cheese stick to a pizza, the answer was “non-toxic glue.” In another example, AI said that cooking spaghetti sauce with gasoline will make it extra spicy. Wow.

While there’s no doubt some silliness has occurred, but jokes aside, for lots of fields, professionals are asking themselves how they will co-exist with AI as it gets smarter and learns more by the day. Many human professionals are feeling a bit uneasy, and that’s understandable.

“There’s a big corporate resistance to AI adoption and there’s reasons for that,” Ramirez said. “Data security is one of those reasons. That’s why your corporation may not want you using this. Another big one is, you don’t know if what you’re using may be someone else’s intellectual property. There are so many unknowns right now, and that’s why corporate America doesn’t want you using it.”

Many of us, as consumers, are using AI in some form or another, without even realizing it, Ramirez said, citing online ordering, optimizing food delivery times and drive-thru technology. “My wife is in marketing, and she gave me an example: If you pick a shade of lipstick from Sephora, it can give you a matching dress from another site,” he said.

For chefs and managers, AI is already helping with things like inventory, procurement and streamlining business through workflow tracking, such as measuring the steps employees are taking in a kitchen throughout the day to determine wasted steps.

“It’s pretty interesting and really great, but how invasive is that and how comfortable we feel with that, I don’t know,” Ramirez said. “But monitoring a kitchen is something we all do. How many steps are being taken, how do we shorten this and how do we make this easier for people?”

He pointed out another way to use AI is plugging in screenshots of complex math equations and charts with the prompt, “Put this in layman’s terms for me.”

As far as robot chefs doing the actual cooking, Ramirez predicts, “I don’t think they’ll ever completely replace us, but they’re getting better and more effective.”

Planning several-week cycles is a keystone of onsite chefs’ work, and Ramirez has been tinkering with AI to help with that often laborious and time-consuming process.

“This is a pain point I’ve always had as a chef and it takes a lot of time,” he said, showing an image of a room with wall-to-wall sticky notes to build a five-week menu cycle. “What we were trying to do is create a nice menu mix of those items. We had to re-do it over and over.”

So Ramirez’s next questions were: “How does AI augment our potential? How does AI augment our HI?”

Fine-tuning prompts that will get effective AI responses is human’s work and it’s not always easy, he found. “With AI, we engineered hundreds of prompts over and over again. The system kept breaking. We said, ‘I want a five-week menu cycle for six stations in a Midwest college town. I put all these guardrails in place and it would just break. So, we had to play around with that. We kind of landed on taking small bites, creating prompts for one station, one week at a time.”

And humans must still keep a watch on AI when planning menus in this way. For example, on a menu for a Mexican concept, AI came up with “Taco Thursday,” rather than “Taco Tuesday.” For the same concept, Ramirez prompted for entrees, but black bean soup and queso fundido appeared, which are more of side dishes or starters.

The same goes for recipes. An example Ramirez found when he asked AI to modify a recipe into a low-sodium version, AI simply said, “Use less salt.”

Basically, when creating recipes using AI, “you have to double-check because AI lies,” Ramirez said.

When it comes to nutritional and allergen information, an even closer watch is required for AI. “You can ask it to provide that information, but often AI is very wrong,” Ramirez said. “You and also use it to generate images for menu items, but these images are pulled from thousands of photos from the Internet, sometimes a combination of thousands of people’s recipes and photos, and they’re ok, but they’re obviously fake.”

Ramirez shared his screen, and with a suggestion from the audience, looked for paella recipes and photos. At first, the photos looked good, but slightly “off,” something of an “uncanny valley” situation, where it’s close to the real thing, but unnerving because of subtle signals and cues our human subconscious can pick up on.

One success story is the fast-casual chain Velvet Taco, which used AI to use existing pantry items to create five unique taco builds, which worked well. “They call it ‘The AI Taco,’ and it’s been a pretty neat marketing plan,” Ramirez said.

Overall, AI is something chefs should continue playing with in different ways. “What AI first started coming about, it was one of those things, where, I’m a chef and I have a creative process,” he said. “I thought creative jobs are safe. We’re safe. It’ll just be things for doctors and lawyers, more technical. But it’s now creating all this beautiful stuff and we never saw it coming. It’s scary.”