Can Artifical Intelligence Beat Mom's Home Cooking?

It's the holiday season: even if it could you wouldn't dare tell her. JL

Geoffrey Fowler reports in the Wall Street Journal:

Can it replace Mom’s home cooking? No. AI still hasn’t mastered bacon, for one. But dining (it) gave me a glimpse of a future where we don’t need to be chefs to make restaurant-grade meals.
My dinner was made by an oven that has a mind of its own.
And this artificial intelligence serves up a pretty tasty baked salmon. Slide a tray of raw fillets into the June oven, then its camera IDs what’s on your menu. The Wi-Fi-equipped June chirps, and an icon of salmon shows up on its touch-screen door. After you insert a probe into the fish, you close the door, then pour yourself a Chardonnay, because the June will cook the salmon to your liking.
Or, at least how a computer thinks you might like it. The June—a robotic Mrs. Cleaver—is humankind’s oldest appliance reinvented for the machine-learning age. For better and worse, appliances are now going online to learn how to do some of the cooking themselves. Can it replace Mom’s home cooking? No. AI still hasn’t mastered bacon, for one. But dining with a June for a month gave me a glimpse of a future where we don’t need to be chefs to make restaurant-grade meals. Created by June Life Inc., a startup that includes former Apple engineers, the oven is audacious for many reasons, including its price: $1,500. It’s the Maserati of toaster ovens, with instant-on carbon-fiber heating elements and a dual-fan convection mechanism for faster and more-precise cooking. But it costs two to three times as much as a fancy “dumb” countertop oven.
As one of the first smart ovens (but by no means the last), the June is already sending foodies into a fuss. Some home chefs de cuisine who wouldn’t balk at the price still might not want a machine ordering them around. At launch, the June knows only about 50 recipes, and they aren’t always especially smart. You can also cook your own recipes, of course.
Yet living with the June, I came to appreciate it in a more practical context. Do I want an oven with a camera, Wi-Fi and the ability to send alerts to my watch? Not particularly. But do I want an oven that keeps me from screwing up what I’m cooking? Now you’ve got my attention. I made some outstanding dishes in the June, from roast broccoli to a whole chicken, while I was busy doing other stuff.
The appeal may be generational. More young professionals subscribe to meal-kit services like Blue Apron, which deliver ready-to-cook ingredients and simple recipes. Why? They’re overworked and grew up in a home where their parents were overworked, too, and might not have passed on cooking skills.
What if the oven knew what was in the Blue Apron box? The June can’t do that today, but you can see where this is heading.
To understand what the June can and can’t do, my editor Wilson and I cooked more than four dozen dishes with two June ovens. We were satisfied with about three quarters of them.
Salmon is one of the June’s best dishes, and one that I’ve been known to screw up on my own. When cooking, the June’s computer monitors temperature in the oven and inside the fish, via that probe. A readout in the app lets you monitor the progress and warns you shortly before it is done. Its camera lets you peek inside the oven: Baking salmon, live-streamed and available for posterity in the June app. (This certainly redefines food porn.)
The question is: How does the June oven define “done”? Professional chefs trained its software to identify a temperature and cooking pace that turned out best in their test kitchen. We found it pretty much ideal across multiple salmon meals. We also got restaurant-quality broccoli, Brussels sprouts, potato wedges, chocolate chip cookies, pork chops and roast chicken, without opening a cookbook.
With steak or lamb, the June transitions between roast and broil on its own, but it won’t flip the meat for you. June’s software can also factor in hotter oven temperatures when you’re cooking multiple recipes back to back—like several dozen chocolate-chip cookies.
The software has its downsides. While not onerous, the June’s touch-screen menus can involve more taps than your traditional oven. There are still bugs: While trying to cook lamb chops on Saturday night, the program kept crashing. June says it’s fixing this—and has already fixed other bugs we noticed. And nothing will enrage hungry humans more than making them wait for lunch because you’re updating the oven’s firmware. (That really happened.)
It also didn’t take long to find the point where cooking science overwhelms the June’s AI. When I made chocolate cookies with a 6-year-old, the June’s automatic recipe did fine at first. But when my little helper decided to go with giant gobs of cookie dough, the oven didn’t know to adjust its recipe. We got gooey cookies.
The June has room to improve. It collects more data than it uses to cook. There’s a scale built into the June’s feet: What about using that to determine how much water is lost through evaporation?
Then there’s that camera built into the hood. Right now, it recognizes raw ingredients. What if it detected changes in shape and color to determine doneness? That would have helped with cookies and steaks, which aren’t really done until they’re brown.
Every time you cook, the June collects data about the dish; but to really improve, the June’s AI needs your feedback. You already rate your Uber driver, why not your lamb chops?
The June’s makers say that’s all in the works.
But will this oven ever really learn what your family likes? Nothing underscores the June’s challenges like bacon. No two Americans agree on what the ideal strip looks like. The June turned out mine a wee bit too crispy for my taste, but it burned a few of Wilson’s batches. He started yanking the pan out early to avoid turning the strips to charcoal. Even if the June adds crispness and thickness settings, the programs can’t keep up with the wild inconsistencies in the bacon world.
To truly master cooking, the June would have to grasp personal taste and culture, too. To make everyone happy, the June may need thousands or millions of recipes—and possibly a questionnaire to nail down exactly what you liked about your mom’s muffins. Perhaps it should also make it easy to simply input Mom’s muffin recipe.
These are serious hurdles. But the June doesn’t need to solve them all at once. Amazon’s Echo, the talking speaker bot, was extremely limited when it first arrived, yet two years later has learned enough to be a Christmas hit. The June’s biggest leap was getting a dumb oven onto the internet. Now that it’s here, it’s worth seeing how it grows.
For me, I’ll know it’s time to buy an intelligent oven when one learns to make an incredible peach pie, a dish I’ve been working to perfect for two decades. I may get jealous when a robot oven beats me at my own game, but given the progress of AI, I’m not betting against it.