In 2014, a Russian computer program named Eugene Goostman became the first machine to pass the Turing Test, a benchmark for artificially intelligent machines.

Mathematician Alan Turing first proposed the test, also known as the imitation game, in 1950. If a human and a computer are having a conversation, and a judge reading a transcript of the dialogue can’t tell who is who, then Turing thought it would be fair to say that the computer is able to exhibit an “artificial intelligence.”

Goostman won the Turing Test by imitating an arrogant 13-year-old Ukrainian with a limited grasp of English, so we’re probably still safe from a robot uprising.

Nevertheless, programs like Eugene Goostman are becoming increasingly common in our daily lives. You’ve probably spent a few minutes talking to one the last time you dialed a customer service hotline, or asked for help from an online chat assistant. In place of having tedious conversations with underpaid humans, the cutting edge of computer science is letting us have tedious conversations with unpaid software.

If “artificial intelligence” is the goal, though, we have to admit there was never anything very intelligent about our conversations with human telemarketers and teenage boys. So programmers have set their sights on a more sophisticated trait of intelligence: creativity. Can computers make music? Can computers write? Can computers cook?

That last question is gamely taken on in “Cognitive Cooking with Chef Watson,” a collaboration between IBM and the Institute for Culinary Education (ICE), a cooking school in New York City.

You might remember Watson as the computer that won a few matches of “Jeopardy!” a few years ago to much fanfare. The gimmick for this cookbook is laid out in a brief introduction: “We humans are limited by what we already know, combinations of ingredients that we inherited through various culinary cultures.” Trillions of possible combinations exist: many of our favorites have come about by accidents of history (such as when tomatoes were imported from the New World to Italy in the 16th century, or when French colonialists brought baguettes to Vietnam). Perhaps a computer could be used to find more?

Watson’s programmers fed it vast databases of existing recipes plus chemical profiles and nutritional information for thousands of ingredients. From this data, Watson could come up with odd new combinations that humans might find tasty. Here’s one such sample output from the book:


These kinds of combinations, in turn, were fine-tuned into actual recipes by chefs from the Institute for Culinary Education. For the ingredients listed above, for instance, chefs improvised an ambitious recipe for a Cuban-inspired seafood bouillabaisse with gelled saffron broth.

The human chefs from ICE clearly did most of the work putting this book together, and this is one shortcoming of the recipes in “Cognitive Cooking”: In an apparent effort to accommodate all the unusual ingredient combinations produced from the computer, the human chefs making these recipes often lean on trendy commercial-kitchen techniques – like foams and sous-vide – that lie beyond the abilities or ambitions of most home cooks. I was unwilling to refine pearls of cucumber juice with locust bean gum just because Watson thought it might make an interesting garnish.

A more serious shortcoming of “Cognitive Cooking,” though, is that its recipes just don’t taste very good.

For instance: the book’s recipe for “Austrian chocolate burritos” suggested browning ground beef with some orange zest, cinnamon and dark chocolate, then serving it in grilled tortillas with queso fresco, mashed edamame and a cocoa-apricot puree.

It was certainly unique. But it was also cloyingly sweet and left me mildly nauseated.

I had a similar experience with the book’s recipe for “Kenyan Brussels Sprouts,” which ruined one of my favorite vegetables with an overpowering dose of ginger and cardamom. These recipes did indeed taste interesting for the first couple of bites. But once the novelty wore off, I was left confronting the fact that I didn’t really want to finish eating them.

They reminded me of the algorithmically-generated ads you’ll see at the bottom of articles on and other websites: They might look enticing, but if you actually click on them, you usually end up regretting it (another similarity: just as those online ads try really hard to sell us diet pills or “one weird trick to boost your credit score,” “Cognitive Cooking” is trying really hard to convince its readers that IBM is still capable of producing interesting technology).

In these early days of artificial intelligence, when computers are still stumbling to interact with humans in a convincing way, it’s interesting to look at the ways the computers fail. Those traits of intelligence, emotion, curiosity and creativity that are the most difficult to boil down into a computer program are the traits where our real humanity lies.

Watson’s attempt at making recipes provides a reminder that a good meal isn’t merely a combination of interesting flavors – it’s a craft, a social occasion, a bonding experience, a repository of traditions and memories, an act of gratitude.

The combinations of ingredients we’ve inherited from families and culture might not actually be a limitation of our meals, but something that enhances our experiences. In the parlance of computer science, it’s a feature of our cuisine, not a bug.

Someday, computers might learn to imitate these more complicated aspects of human experience. If they do, we’ll waste more time clicking on manipulative Internet ads, and we’ll have customer service bots that actually care when we get frustrated with them.

But even in a world of creative and emotionally complex computer programs, Brussels sprouts will still taste best roasted with just a little bit of salt and olive oil.