IBM’s Watson Designed The Worst Burrito I’ve Ever Had

And I’ve had a lot of burritos.

It’s the worst burrito I’ve ever had.


I don’t know another way to say it. I’m staring at my plate in disbelief. Could burritos be bad? Yes, yes I’d just learned. But that’s not the biggest shocker. The biggest shocker is that this recipe was largely designed by Watson, IBM’s best artificial intelligence–one that had already fed me one of the most uniquely delicious BBQ sauces I’d ever eaten.

I thought through the recipe in my head again. I’d cheated a little, but not enough to ruin a good thing. What went wrong?

The Book
Chef Watson–IBM’s Jeopardy-robot-gone-algorithmic-recipe engine–has released his first cookbook. It’s a collection of boozy drinks and haute recipes, developed by Watson’s flavor engine, in conjunction with the human chefs from the Institute of Culinary Education (ICE)–a respected NYC school known for its intense six- to 12-month curricula–who helped turn strange ingredient lists into palatable plated dishes.

It’s a beautifully photographed book, full of challenging recipes that require modernist techniques like sous vide and spherification. But the cookbook’s most intriguing recipe is also its simplest to both cook and source: an Austrian Chocolate Burrito.

As a polyamorous lover of the entire burrito medium, I had to try it. I was a little skeptical looking at the recipe. Its ingredients resembled a citrusy chocolate truffle as much as any burrito I knew. But this was Watson, after all. Watson beat Ken Jennings in Jeopardy. Watson could probably beat Kasparov in chess. Surely, Watson could design a burrito recipe better than some self-taught home cook.

The Recipe
Chef Watson creates unexpected, but delicious food pairings by nature. (Read more how he does it here.) That’s in fact the software’s value–to suggest ideas we wouldn’t think of on our own (which is why IBM sees Chef Watson as a precursor to a Watson engine that could help doctors diagnose illnesses, or investors spot the next hot stock).


And yet. This burrito recipe is a strange beast. The ground beef isn’t an odd ingredient, nor is the chocolate if you’re familiar with traditional Oaxacan mole. But it contains almost no flavors you’d associate with either Mexican or Tex-Mex cuisine. No peppers, cumin, onion, garlic, or herbs of any sort.

IBM & Institute of Culinary Education

Instead, you zest a whole orange into a skillet of ground beef with a pinch of cinnamon. You puree some plain edamame as a sort of salsa. And you reduce apricot puree, vanilla bean, and dark chocolate to create a sauce. Then you top off this mixture with some cotija cheese. Well that’s what the recipe said. Then I noticed an asterisk. Apparently Watson had originally suggested cheese curds, while the human chef in charge of this recipe, ICE Creative Director Michael Laiskonis, had opted for a more traditional cotija cheese.

I figured, when one of the smartest AI engines on the planet tells you to eat more cheese curds, you eat more cheese curds. So I opted for cheese curds.

The Cooking
Maybe I should have seen the heartbreak coming. After mixing myself Watson’s Japanese Wasabi Cocktail as a little appetizer, I found it to be one of the least palatable cocktails I’d ever mixed. I would definitely never serve it to someone. A combination of wasabi powder, sake, and lime, it had a fine flavor, but the proportions were way off. With half a lime of juice to 2 ounces of sake, and no sugar, it was the most puckering drink I’ve ever had.

IBM & Institute of Culinary Education

So as I prepared to roll $32 worth of groceries into a warm tortilla shell, I doubled my pour of sake, and soldiered on following General Watson to burrito town.


I’ll save you the pangs of the cooking stuff. But in full disclosure, I couldn’t find fresh apricots in the barren wasteland of the Midwest, where I live, so I had to use an apricot jam.

The Eating
The first thing that hits my tongue is that flour tortilla. I wait for that kick–the announcement that the burrito filling has arrived–and it never comes. I swish my tongue through textures that honestly aren’t all that pleasant. Ground beef. Pulverized edamame. It’s just a bunch of little balls of quasi-dryness in my mouth. As for orangey beef, it’s neither bad nor good. It’s just not enough when your brain is expecting smokey peppers and hoping for a little heat. I could have used more cinnamon, too.

The bit of apricot-chocolate sauce I get in the next bite is exciting if only because it’s a sort of a lubricant. That’s the best part of the burrito. Even though my version is probably sweeter and less nuanced than it’s supposed to be, because my apricot puree is from a jar, I can begin to see the logic at play. There’s a good idea here, a rich, fatty, fruitiness with a flavor brighter than some dark mole. It could work in a better burrito though, again, maybe with some pepper added.

Who’s To Blame?
My meal wasn’t good. In fact, it was pretty bad. If I’d made this for my family, I’d apologize and suggest we order a pizza.

The drink I executed perfectly. The burrito I made, let’s say 85% right.

As I scan through the recipes again, ensuring that I didn’t miss some crucial step or ingredient, I can’t help but wonder, was my meal Watson’s mistake, or was it his human handler in each recipe–Chef Caporale for the drink, and Chef Laiskonis for the burrito?


Chef Watson is a platform that’s been evolving over time. In later iterations, he suggests preparation styles and amounts of each ingredient to use–you know, the normal stuff you expect in a recipe. In the recipes I made, Watson was still spitting out a list of compatible ingredients that the human chefs turned into dishes.

This is, of course, the relationship IBM wants to promote with the future of AI. The computer parses big data and serves up ideas. The human takes them to the next level. But in this scenario, who do you blame when things go wrong?

When I consider how I’d make Watson’s list of burrito ingredients my own way, I think I’d create a true edamame apricot salsa, and dust it with orange zest. Maybe I’d even melt the cheese with this fruit mix inside to create a sweet sort of fondue. Then I’d flavor the beef with cocoa powder. Maybe then, I’d have something pretty good. But it would still beg for some ancho peppers and maybe some cilantro. As far as I’m concerned, both the robot and the human screwed up this burrito.

Kenyan Brussels SproutsIBM & Institute of Culinary Education

Even still, I like the Chef Watson cookbook. In fact, I’d recommend you buy it the next time you know of a foodie in need of a gift. I’m convinced there are some delicious recipes in this book. (In particular the Kenyan Brussels Sprouts, flavored with ginger and cardamom, look tempting to try.) But more than that, I think it will be a fascinating cultural artifact down the line. Because one day, maybe not so long from now, the robots will learn to make better burritos than humans. And when that time comes, that one afternoon when we ate an algorithmically lousy burrito will feel like the good old days.

Buy the book here.


About the author

Mark Wilson is a senior writer at Fast Company who has written about design, technology, and culture for almost 15 years. His work has appeared at Gizmodo, Kotaku, PopMech, PopSci, Esquire, American Photo and Lucky Peach