307. A.I. Can't Bake

27-07-2023 • 9 min

You’ve probably heard that, in a blind taste test, even experts can’t tell between white and red wine. Even if this were true – and it’s not – it wouldn’t matter.

I was in Rome last month, visiting some Raphael paintings to research my next book, and stopped by the Sistine Chapel.

I’ve spent a good amount of time studying what Michelangelo painted on that ceiling. There are lots of high-resolution images on Wikipedia.

But seeing a picture is nothing like the experience of seeing the Sistine Chapel. You’ve invested thousands of dollars and spent fifteen hours on planes. You’re jet-lagged and your feet ache from walking 20,000 steps. You’re hot.

When you enter, guards order you to keep moving, so you won’t block the door. They corral you to the center, and you can finally look up.

When you hear wine experts can’t tell between white and red wine, you imagine the following: Professional sommeliers are blindfolded, and directed to taste two wines. They then make an informed guess which is white, and which is red. In this imaginary scenario, they get it right half the time – as well as if they had flipped a coin.

If it were true wine experts couldn’t tell between white and red wine, the implication would be that the experience of tasting wine is separate from other aspects of the wine. That the color, the shape of the glass, the bottle, the label, and even the price of the wine are all insignificant. That they all distract from the only thing that matters: the taste of the wine.

There’s some psychophysiological trigger that gets pulled when you tilt your head back. Maybe it stimulates your pituitary gland. When you have your head back and are taking in the images on the Sistine Chapel ceiling, you feel vulnerable. (You literally are vulnerable. You can’t see what’s going on around you. You’d be easy to physically attack.)

What you see is overwhelming. As you try to focus your attention on some detail, some other portion of the imagery calls out and redirects your attention. This happens again and again.

After a while, your neck needs a rest, and you return your gaze to eye-level. And this is almost as cool as the ceiling: You see other people with their heads back, their eyes wide, mouths agape, hands on hearts, tears in eyes. You hear languages and see faces from all over the world. You realize they all, too, have invested thousands of dollars and spent fifteen hours on planes. They, too, are jet-lagged and hot and have walked 20,000 steps.

You can look at pictures of the Sistine Chapel ceiling on the internet. You can experience it in VR. In many ways, this is better than going to the Sistine Chapel. You can take as much time as you want, and look as close as you want. You don’t have to spend thousands of dollars and fifteen hours on a plane, take time off work, or even crane back your neck.

But seeing the Sistine Chapel ceiling on the internet or even VR is only better than seeing it in person, in the way that a spoonful of granulated sugar when you’re starving is better than a hypothetical burger in another iteration of the multiverse.

We’ve seen an explosion of AI capabilities in recent months. That has a lot of people worried about what it means to be a creator. Why do we need humans to write, for example, if ChatGPT can write?

The reason ChatGPT’s writing is impressive is the same reason there’s still a place for things created by humans.

Anyone old enough to have been on the internet in the heyday of America Online in the 1990s will remember this: When you were in a chat room, most the conversations were about being in a chat room: How long have you been on the internet? Isn’t the internet cool? What other chat rooms do you like? Part of the appeal of the question “ASL?” – Age, Sex, Location? – was marveling over the fact you were chatting in real-time with a stranger several states away.

Or maybe you remember when Uber or Lyft first came to your town. For the first year or two, likely every conversation you had with a driver was about how long they had been driving, about how quickly the service had grown in your town, which is better – Uber or Lyft?, or which nearby cities got which services first.

The first few months ChatGPT was out, it was seemingly the only thing anyone on the internet talked about. But it wasn’t because ChatGPT’s writing was amazing. ChatGPT is a bad writer’s idea of a good writer. It was because of the story: Wow, my computer is writing!

Now that much of the novelty of ChatGPT has worn off, many of us are falling into the Trough of Disillusionment on the Gartner Hype Cycle. We’re realizing ChatGPT is like a talking dog: It’s impressive the dog can appear to talk, but it’s not talking – it’s just saying the words it’s been taught. ChatGPT is very useful in some situations, but not as many as we had originally hoped.

What made us talk about the internet while on the internet, talk about Uber while in Ubers, and talk about ChatGPT while chatting with ChatGPT was the story. Once the story behind the internet or Uber wore off, we started to appreciate them for their own utility.

Part of what’s cool about seeing the Sistine Chapel ceiling in VR is that – we’re seeing it in VR. But even if that weren’t impressive, what would still be impressive about the paintings would be more than just that they’re amazing paintings. It’s incredible to us a human could paint such a massive expanse. We think about the stories and myths of Michelangelo, up on that scaffolding, painting in isolation. Part of our appreciation of the Sistine Chapel ceiling lies outside the ceiling itself. While marveling at it, we can’t help but think of Michelangelo’s other masterpieces, such as the David or the Pietà.

Lloyd Richards spent fourteen years writing Stone Maidens, and had almost no sales for decades. Suddenly, he sold 65,000 copies in a month. He was interviewed on the TODAY show, and got a book deal with a major publisher.

How did he do it? His daughter made a TikTok account. The first video showed Lloyd at his desk, and explained what a good dad he was, how hard he had worked on Stone Maidens, and how great it would be if he made some sales. Then the #BookTok community did the rest.

Stone Maidens is apparently a good book. But it’s no better today than it was all those years it didn’t sell. Most the comments on Lloyd’s TikTok account – which now has over 400,000 followers – aren’t about what a great book Stone Maidens is. They’re about how Lloyd seems like such a nice guy, or how excited each commenter is to have contributed to his success.

The study that started the myth that wine experts can’t taste the difference between white and red wine didn’t show that. The participants in the study literally weren’t allowed to describe the two wines the same way – they couldn’t use the same word for one as the other. It wasn’t blindfolded – it was a white wine versus the same wine, dyed red. The study wasn’t about taste at all: Participants weren’t allowed to taste the wine – they were only allowed to smell. And wine experts? That depends on your definition of “expert”. They were undergraduate students, studying wine. They knew more than most of us, but were far from the top echelon of wine professionals. Most damning for this myth was that the same study casually mentions doing an informal blind test: The success rate of their participants in distinguishing the taste of white versus red wine: 70%.

That this myth is false shouldn’t detract from the point that even if it were true, it wouldn’t matter. What the authors of this study found was not that wine enthusiasts couldn’t tell between white and red wine, but that the appearance of a wine as white or red shaped their perceptions of the smell of the wine.

Once you bake a cake, you can’t turn it back into flour, sugar, butter, and eggs. You can’t extract the taste of a wine from the color, the bottle, your mental image of where the grapes were grown and how the wine was made, or even the occasion for which you bought the wine. Something made by an AI can be awesome, either because it’s really good at doing what it’s supposed to, or because you appreciate it was made by an AI. Something made by a human is often awesome because of the story of the human who made it, and the story you as a human live as you interact with it.

If you want to be relevant in the age of AI, learn how to bake your story into the product. Because AI can’t bake.

Image: Figures on a Beach by Louis Marcoussis