Connect with us

Lifestyle

Can AI make better chocolate chip cookie recipes than humans? We taste tested 2

Published

on

Can AI make better chocolate chip cookie recipes than humans? We taste tested 2

These chocolate chip cookies from America’s Test Kitchen are yummy. But can they top a cookie created by AI?

‎/America’s Test Kitchen


hide caption

toggle caption

Advertisement

‎/America’s Test Kitchen

Can artificial intelligence make a tastier chocolate chip cookie recipe than a human being?

At the risk of upsetting millions of grandmothers everywhere, we set out to find an answer.

We recruited Dan Souza, chief content officer for America’s Test Kitchen, for our experiment. He matched the Test Kitchen’s Perfect Chocolate Chip Cookie recipe against recipes from two AI programs, ChatGPT and DishGen.

Advertisement

Why chocolate chip cookies?

“Cookies, when you make small changes to ingredients, we find that you get some pretty massive differences,” Souza told Morning Edition’s A Martinez. So, it’s kind of a fun one to use as a litmus test for how successful a recipe development could be.”

How did the experiment work?

Souza asked each AI program to come up with a chocolate chip cookie recipe. The results were nearly identical. Souza said there’s a good reason for that.

“What it gave me was a pretty traditional chocolate chip cookie recipe. If you look on the back of Toll House [chocolate chip] morsels, which is where most chocolate chip cookie recipes kind of originate, it was a pretty good mimic to that. You have your classic ingredients, you have your flour, you’ve got white sugar and a little bit of brown sugar, a couple of eggs,” Souza said. “What you find with these engines is they’re pulling from all over the place and so you get sort of an average output, and it looked like a really average cookie to me.”

One of these cookies is different from the other. The one on left was created from a recipe generated by ChatGPT. The cookie on the right came from a recipe app called DishGen.

One of these cookies is different from the other. The one on the left was created from a recipe generated by ChatGPT. The cookie on the right came from a recipe app called DishGen.

‎/America’s Test Kitchen

Advertisement


hide caption

toggle caption

‎/America’s Test Kitchen

Advertisement

Souza said both of the programs were fairly straightforward. He typed in prompts of what he was looking for and the results popped up quickly.

DishGen had a “modify” button if, for example, you wanted to change the recipe to make the cookies chewier. But there were some frustrations, as well.

“The craziest thing is I would do the search — I did it multiple times on different days — and I actually got completely different recipes. So, the same prompt but I had a different recipe, which I found like totally infuriating. If I had something that I liked and I wanted to make it again, I couldn’t.”

The taste test

In our view, the cookies from ChatGPT and DishGen were pretty good but a little boring. A variation on the ChatGPT recipe that was intended to make the cookies more chewy actually made them too chewy.

Advertisement

The Test Kitchen’s Perfect Chocolate Chip Cookie recipe lived up to its name. The cookies were crunchy on the edges and chewy in the middle. They had a nutty flavor with a hint of toffee.

Souza said the Test Kitchen recipe has many innovations that the AI engines could not pick up on.

“It’s seemingly simple, but there’s a lot going on there,” Souza said. “One of the things we’ve done is we browned the butter, which does two things. It adds tons of rich, nutty flavor. But liquefying the butter also means that you get a denser cookie, which eats chewier. So, it’s really, really valuable.”

Souza said the AI recipes have a lot of potential, but that the technology is not quite there yet.

“It’s missing the people part of it. So, if you’re tasting a chocolate chip cookie recipe that your mom has made forever, there’s a big emotional pull there. And that actually influences how you taste something. You know, we eat with our eyes, but we also eat with all of our emotions and kind of everything that we bring into the picture. You’re never going to get that from an AI.”

Advertisement

Reena Advani edited the radio version of this story. Obed Manuel edited the digital.

Continue Reading
Click to comment

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Lifestyle

‘Wait Wait’ for February 28. 2026: Live in Bloomington with Lilly King!

Published

on

‘Wait Wait’ for February 28. 2026: Live in Bloomington with Lilly King!

An underwater view shows US’ Lilly King competing in a heat of the women’s 200m breaststroke swimming event during the Paris 2024 Olympic Games at the Paris La Defense Arena in Nanterre, west of Paris, on July 31, 2024. (Photo by François-Xavier MARIT / AFP) (Photo by FRANCOIS-XAVIER MARIT/AFP via Getty Images)

François-Xavier Marit/Getty Images


hide caption

toggle caption

Advertisement

François-Xavier Marit/Getty Images

This week’s show was recorded in Bloomington, Indiana with host Peter Sagal, judge and scorekeeper Bill Kurtis, Not My Job guest Lilly King and panelists Alonzo Bodden, Josh Gondelman, and Faith Salie. Click the audio link above to hear the whole show.

Who’s Bill This Time

State of the Union is Hot; The Tribal Council Convenes Again; A Glow Up In the Doll Aisle

Advertisement

Panel Questions

The Toot Tracker

Bluff The Listener

Our panelists tell three stories about a travel hack in the news, only one of which is true.

Not My Job: Olympic Swimmer Lilly King answers our questions about Lil’ Kings

Advertisement

Olympic Swimmer Lilly King plays our game called, “Lilly King meet these Lil’ Kings” Three questions about short kings.

Panel Questions

Cleaning Out The Cabinet; Bedtime Stacking

Limericks

Bill Kurtis reads three news-related limericks: Getting Cozy With Cross Country Skiing; Pickleball’s New Competition; Bees Get Freaky

Advertisement

Lightning Fill In The Blank

All the news we couldn’t fit anywhere else

Predictions

Our panelists predict, after American Girls, what’ll be the next toy to get an update.

Advertisement
Continue Reading

Lifestyle

Zendaya and Tom Holland Are Married, Her Longtime Stylist Claims

Published

on

Zendaya and Tom Holland Are Married, Her Longtime Stylist Claims

Law Roach
Zendaya and Tom’s Wedding Already Happened …
Y’all Missed It!!!

Published

Advertisement

Advertisement

Advertisement
Continue Reading

Lifestyle

Bet on Anything, Everywhere, All at Once : Up First from NPR

Published

on

Bet on Anything, Everywhere, All at Once : Up First from NPR

Online prediction market platforms allow people to place bets on wide-ranging subjects such as sports, finance, politics and currents events.

Photo Illustration by Scott Olson/Getty Images


hide caption

toggle caption

Advertisement

Photo Illustration by Scott Olson/Getty Images

The rise of prediction markets means you can now bet on just about anything, right from your phone. Apps like Kalshi and Polymarket have grown exponentially in President Trump’s second term, as his administration has rolled back regulations designed to keep the industry in check. Billions of dollars have flooded in, and users are placing bets on everything from whether it will rain in Seattle today to whether the US will take over control of Greenland. Who’s winning big on these apps? And who is losing? NPR correspondent Bobby Allyn joins The Sunday Story to explain how these markets came to be and where they are going.

This episode was produced by Andrew Mambo. It was edited by Liana Simstrom and Brett Neely. Fact-checking by Barclay Walsh and Susie Cummings. It was engineered by Robert Rodriguez. 

We’d love to hear from you. Send us an email at TheSundayStory@npr.org.

Advertisement

Listen to Up First on Apple Podcasts and Spotify.

Continue Reading

Trending