AI vs. Famous Chef Recipes

Letzte Aktualisierung:
8. June 2024

AI vs chefs cookoff

Once upon a time, there were a limited number of options for food inspiration. Besides the meals you enjoyed while dining out or the recipes you gleaned from a cookbook, you may not have had another go-to resource for what to cook for dinner on any given night. 

Of course, the internet has completely changed the #FoodPorn landscape. Not only have people turned taking pictures of their food into a social media rite of passage or can a near-endless catalog of new recipes be found online, but “foodies” and food bloggers have become as influential as professional chefs for some people. 

But now there’s a new contender in the food inspiration category, and it’s completely digital, and it’s name is GPT-3. We wanted to see if this third-generation deep-learning language model could produce delicious recipes, perhaps as delicious as some of the most famous chefs. OpenAI’s GPT-3 is the first one-shot learning language model, which means it trains itself using information found anywhere on the internet to make an incredibly sophisticated model of English language. This allows it to creatively extrapolate on nearly any natural language prompt with responses that feel human-generated. You can learn more about GPT-3 here

For this experiment, we chose some of the most popular recipes from the most notable chefs around the world, from Julia Child to David Chang. Then we simply gave GPT-3 the name of each famous dish, whether that was “braised beef,” “tomato sauce,” or “honey soy-glazed vegetables” and asked it to generate a recipe for us. This allowed GPT-3 to create a similar recipe using its learning language model. It’s important to note that GPT-3 was not given the name of Gordon Ramsay or any of his recipes, but seemed to have an affinity towards Ramsay probably due to the TV chef’s prolific recipes. We went a step further and had a group of volunteers cook up both the chef recipes alongside their AI counterparts. All ratings of each recipe were decided by our cooks, who weighed in on each recipe’s taste, creativity, and difficulty. Can AI create a better beef bourguignon than Julia Child or a superior schnitzel to Wolfgang Puck’s? Read on to see what we discovered from this chef vs. AI showdown. 

So how does this experiment work? To understand how well GTP-3 was able to create new recipes based on these famous chef’s meals, we randomly assigned the recipes to our volunteer chefs and had them cook each recipe side by side to compare taste as well as take notes on the recipe process. 

To help showcase where the AI-generated recipes succeeded (and possibly where they may have fallen short), we asked our volunteer chefs to score both the randomly assigned recipe and the original famous chef creations on a scale of one to five in three categories: taste, creativity, and difficulty of cooking. 

Combined, we found the AI-generated recipes were technically easy to execute, but also lacked in the taste-factor. Still, our volunteer chefs rated the creativity of the AI-generated recipes compared to the chef recipes nearly the same on average (3.2 and 3.5, respectively), without knowing how they were created. As you’ll see in the individual matchups below, GTP-3 came up with some pretty interesting recipes, and even once managed to generate a meal that outshined a famous chef’s meal

Ratings of Recipes

Our volunteers rated each recipe, the famous chef’s as well as AI’s version of their recipe, on a scale of 1-5. Here are the overall ratings for these three categories:

Scale: 1 = Not at all, 3 = Somewhat, 5 = Extremely

Recipe Taste:

Chef Recipes – 4.5

AI-Generated Recipes – 2.8

Recipe Creativity:

Chef Recipes – 3.5

AI-Generated Recipes – 3.2

Recipe Difficulty:

Chef Recipes – 2.8

AI-Generated Recipes – 2.5

Braised Beef Battle

Beef bourguignon, or simply “braised beef,”  has a long and controversial history, but there’s no denying the role Julia Child played in bringing the dish back to life and centering it as a standard in French cuisine. Dating as far back as 1878, this layered dish of meat (marinated in red wine) and vegetables was made popular in the U.S. in 1961 by Child and voted as the national dish of France in 2017. In order to generate this recipe, we prompted GPT_3 with “recipe for braised beef with onions and mushrooms” in order to generate something as close to Child’s braised beef, for comparison. Below is a graphic of the two cooking methods side-by-side: 

In the battle between the classic Julia Child dish and the AI-created equivalent of beef bourguignon, the AI interpretation might have been easier to compile. According to our volunteer cooks, the AI recipe (scoring a 3 out of 5 for difficulty) lost points for both creativity and taste. It also resembled a steak with toppings, compared to the stewed complexity of Child’s dish. 

According to our volunteer cook, the AI-inspired beef bourguignon captured the fundamental aspects of the meal, but missed on the more intricate details, leading to subpar taste and a lack of creativity (3 out of 5 stars each) compared to Julia Child’s recipe. 

Notable differences included:

  • The amount of wine included in the dish (a full bottle from Child vs. one cup from the AI).
  • Omission of some of the more flavorful ingredients in AI’s recipe, including thyme, tomato paste, and bacon. 
  • AI’s recipe did not mention whether the beef had to be chopped or not, in contrast to Child’s recipe in which the meat is cubed. 

Our volunteer cook’s note: “Overall, the AI recipe tasted better than expected, but it lacked the flavor profile of Julia Child’s recipe.”

Tomato Sauce Taste-Off

Marcella Hazan is known as an Italian cuisine legend, and one of her most inspired recipes is a tomato sauce heralded as being simple but extremely effective. A cornerstone in many Italian dishes, there’s no finer tribute to her culinary prowess than this potentially effortless tomato base. We asked GPT-3 to simply generate a recipe for “tomato sauce.” But here is what our volunteer cooks had to say about Marcella Hazan’s tomato sauce recipe compared to AI’s version.

The Marcella Hazan and AI-inspired recipes for tomato sauce may have looked nearly identical in the precook phase, but the finished products differed substantially. Our volunteer chef found that the AI-created recipe was both more creative (4 out of 5 stars) and more difficult (3 stars), but severely lacking in taste. As they described, the ingredients were present, but the AI program wasn’t able to reproduce the execution of Hazan’s sauce. 

Here were some notable differences between Hazan’s recipe and AI’s rendition:

  • AI’s recipe was much more confusing to follow and lacked attention to details.
  • AI’s inclusion of raw garlic and instructions to blend them into tomatoes resulted in an overpowering sauce. 
  • Hazan’s sauce did not require roasting prior to simmering. AI’s on the other hand called for roasting the tomatoes and blending them, with no simmering required. 

Our cook’s tip: don’t waste your tomatoes on AI’s raw, garlicky recipe unless you have a need for repelling vampires. “Hazan’s recipe was simple and outstanding … I would have been happy with that sauce at any high-end Italian restaurant.”

Who’s Schnitzel Is Better?

You don’t need to be a foodie, or even enjoy cooking, to know who Wolfgang Puck is. A former “Top Chef” judge and owner of dozens of restaurants all around the world, Puck has been known to proclaim the importance of simplicity in creating some of his most signature dishes. And while Puck may be inspired by Asian cuisine, we asked the GPT-3 AI program to generate a “recipe for pork schnitzel” so we could compare the final results. 

As you can see from the prep phase, the Wolfgang Puck recipe called for pounding the meat into a thin pancake prior to cooking, a step the AI recipe forwent. Instead, the AI recipe directed cooks to dip an entire pork chop into batter, without any meat preparation beforehand. While both recipes were relatively simple (both scored a 2 out of 3 for creativity), the Wolfgang Puck recipe was slightly more complicated in execution, but well worth the result in flavor.

Here were some notable differences between the recipes:

  • Puck’s requirement to pound the meat out made for quick and consistent cooking, and also produced a juicer schnitzel, with a crispier exterior than AI’s version.
  • Because the pork in AI’s recipe was notably thicker (did not mention pounding out the meat at all), the pork required a longer cooking time, which led to a more overcooked crust. 
  • AI’s recipe also gave us the lazy option of skipping any bread crumbs, which would have led to a very unpleasant texture for schnitzel. 

Our volunteer cook had this to say about their cooking experience: “Surprisingly, the final product of the AI recipe was still tasty, but I would not skip bread crumbs when it comes to schnitzel, as it allows. Puck’s recipe was very delicious and easy to cook, we’ll be making it again!”

All’s Fair in Love and Lasagna

If you’re a fan of Rachael Ray, you should know her Mexican lasagna is one of her most highly-rated recipes. We simply asked GPT-3 for a recipe for “Mexican lasagna,” for this prompt, and surprisingly, it seemed to take a lot of cues from Ray’s recipe, including the incorporation of tortillas for its “lasagna” layers. But when we asked our volunteer cook to create both Rachael Ray’s Mexican lasagna recipe alongside AI’s version, the final results couldn’t have been further apart in taste and consistency. 

Though the recipe for AI’s version of this lasagna contained some delicious ingredients similar to Rachael Ray’s version, there were critical ingredients missing that made the consistency less than appealing, straddling between a lasagna and a soup. Not only was the AI recipe more complicated (scoring a 3 out of 5 stars for difficulty), it earned the lowest score out of any of our six dishes, just 1 star for overall taste. 

Here were some notes our cooks took during their experience:

  • AI’s lasagna was missing chicken, which was included in Ray’s recipe. It did include a lot of chicken stock, however, which probably led to it’s soupy consistency. 
  • The level of tortilla prep in the AI recipe felt unnecessary. 
  • Rachael Ray’s recipe was easier to follow and was a bit more simple to make. 

Despite issues with the ingredients in AI’s recipe, our cooks said this recipe could have some potential: “I actually would have liked the combination of flavors in the recipe more (it included avocado and sour cream) if the filling were thicker and more consistent.”

Saucy Competition

You might know David Chang from his podcast, his Netflix show, possibly his new memoir, or any number of his restaurants across the U.S. Known for Asian cuisine and his famous Noodle Bar and Momofuko, David Chang has become a modern industry standard. So how did AI stack up to the Chang’s honey- and soy-glazed vegetables recipe? We asked it for a “honey- and soy-glazed vegetables recipe” and it delivered the honey and soy, but not the vegetables. 

As evidenced by the precook photo, the AI-created recipe left out a crucial detail, omitting the vegetables that needed to be coated in the sauce. Our chef improvised by adding cauliflower, but that wasn’t enough to elevate this dish to Chang’s unique flavors and ingredients. Chang’s recipe might be more complex (scoring a 3 out of 5 stars for difficulty), but it was undoubtedly more creative and delicious. 

Here were some differences noted by our cook: 

  • David Chang’s recipe called for using unique ingredients (including molasses to marinate the mushrooms) that helped the flavors of this dish shine. 
  • AI omitted vegetables from its ingredients list. 
  • AI also mentioned making vegetables “up to five days in advance,” and that just seemed gross. 
  • AI’s sauce was very sweet and one-dimensional. Chang’s recipe had a lot of complexity in both flavor and texture, from the molasses and rice cracker topping on the mushrooms to the use of turnips and swiss chard.

Our cook had some final notes on the experiment: “Chang’s recipe was really unique. I’ve never used molasses on mushrooms before, but it produced a very deep flavor that amplified the umami of the mushrooms in a very different way.” 

Chicken Fight in the Kitchen

Possibly better known as the Barefoot Contessa, Ina Garten’s numerous cookbooks and TV shows have earned her legions of fans and no small number of famous dishes. Self-taught, but still one of the biggest names in the food and culinary scene, Ina Garten has helped amateur chefs cook like professionals without having the full repertoire of skills or training. One of her most popular dishes, skillet-roasted lemon chicken, is a perfect and delicious recipe for beginner cooks, so we were curious about how AI would reinvent a “recipe for skillet lemon chicken.”

This recipe was the only one in our experiment in which AI’s version tasted better than the chef’s version! Though AI’s skillet lemon chicken recipe may have been slightly more confusing to follow (our chef had to read through it several times before beginning to cook), the end product might just have been worth the extra effort. 

Here are some notes our volunteer cook took during the experiment:

  • AI’s version was less presentable on the plate.
  • AI’s recipe earned higher marks for creativity (4 out of 5 stars).
  • AI’s chicken tasted better than Ina Garten’s chicken.
  • Ina Garten’s dish was more presentable. 

In closing, our cook noted: “AI’s recipe was not as clearly written as the chef recipe but was more tasty!”

Finding Food That Makes You Happy 

Cooking isn’t always a science or, at least, not one that artificial intelligence has mastered just yet. Though, it’s not impossible to put together a great recipe that can rival a famous chef’s – even a computer program has the capacity to put together the occasional tasty recipe. Among the six AI-created meals our chefs assembled, two earned high ratings (either 4 or 5 stars) for taste, and one even outscored the professional recipe it was inspired by. Overall, the AI recipes were more simple than their chef counterparts, pairing down the number of ingredients and occasionally omitting entire steps in the preparation. Occasionally, the AI recipes included unsafe cooking practices, including placing a casserole dish on a hot stove burner, so we’d recommend checking the temperature of your meats and looking up any questionable cooking directions before you undertake any AI-generated recipe experimentation.

Regardless of where your food inspiration comes from, no one should have to suffer from the effect of reflux. At Refluxgate, our mission is to help empower you to take control of your health by providing insight into all of the available treatment options in one easy-to-use location. By sharing information from the latest research, as well as interviews with experts in the field, we specialize in silent reflux, while also providing information on classic reflux symptoms (including heartburn). Read more by visiting us online at Refluxgate.com today.

Fair Use Statement

Feeling foodie inspired by some of the meals we put together? Share the feeling (and our findings) with your readers for any noncommercial use with the inclusion of a link back to this page in your story as credit to our team for their work on this report.

About the author 

Gerrit Sonnabend

Gerrit is a German data scientist & medical publisher. His formal education is in qualitative research. He had severe reflux himself. Read more about him here.