Researcher Spotlight: Aaron Hertzmann on perception, creativity, and the reasons AI can’t replace artists

June 3, 2022

Tags: AI & Machine Learning, Careers, Researcher Spotlights

Principal Scientist Aaron Hertzmann wants to understand more about the connections between human visual perception, art, and photography. In his research, he’s considered everything from how line drawings work, to the limitations of linear perspective, to the evolving relationship between art and artificial intelligence.

We talked to Hertzmann—who is also an ACM Fellow and an IEEE Fellow—about his current research interests, the meaning of art in an age of AI, and the future of creative technologies.

Can you tell us about what you’re working on at Adobe Research?

I’m interested in how ideas from computer graphics and vision can help us understand art and perception in photography, and how these understandings can help us create better tools for making art.

Here’s an example: Our received wisdom tells us that linear perspective, as invented in the Renaissance, is the correct way to do perspective. So almost all of our photography is founded on linear perspective—all of our smartphone cameras and SLRs use it.

But when we look at how artists actually work, they often don’t use linear perspective. So I’m interested in how we can incorporate ideas from what artists actually do—the ways they create better images—into our cameras.

You’ve also done research about the differences between how people see, paint, and draw things and how they appear in photos. Can you tell us about that?

The differences between photographs and other types of art compared to what we see in the real world can be hard to see. When you take a picture, you usually think, ‘this is how the world was.’ But if you compare with the thing you’ve photographed, you might start to wonder, ‘Is that mountain as big in the picture as it seems in real life?’ And you might realize that it’s smaller. Or the moon might look tiny in a picture, but it feels big in the real world.

I’ve had this experience a bunch of times. I’ll draw a picture of a bridge or a building in the distance and I’ll compare it with a photograph and my bridge or building is much bigger in the drawing—that’s the way it felt to me in the world, but the photograph made the object in the distance very small. A number of researchers have written about this effect. We’ve published a paper on editing images based on this idea, and there’s a lot more to be done. A similar thing happens with color. The sunset looks so bright and colorful, and the photograph looks good, too, but it’s in a different way. The smartphone camera creates such compelling illusions, but they’re not reality, and artists often make different choices to capture different aspects of how the world looks.

The phone has one set of choices built into it, but I think there’s a lot more we can do to represent other choices people could make, and to represent the thing you want to express about the scene.

Artwork and photograph by Aaron Hertzmann

You’ve said AI can’t replace artists. Can you explain?

The low-level reason that AI won’t replace artists is because it’s really just about tools. Each new thing seems super cool at first. After you play with it, you see that it becomes predictable. Once it’s been around for a while, you get used to it. It becomes a tool you can use in new ways.

I think the newest technologies follow the same general trend we’ve seen over time. To make a realistic picture 200 years ago, you had to pick every color yourself. A hundred years ago, you just had to point the camera and choose exposure settings, but you still had to find the thing you wanted to photograph and do a lot of darkroom work. We’re moving towards being able to describe an idea to create a picture. But it’ll only be effective as art when the artist puts a lot of skill and effort into it in some way. What that looks like remains to be seen.

As we get new tools, the older art forms don’t die. They evolve.

I’ve also started to think about art as a social behavior, which is why it will always require humans. We create art to express our social relationships with other people, to affect them, to give gifts, to express our affiliations and tribal status, things like that. In all these cases, we care about art because it came from a person.

What would you say to someone who’s considering joining Adobe Research?

I took an unusual path to Adobe Research. I’d been a professor for 10 years and I was at the point where I’d done everything I was going to do there, so coming to Adobe Research was a good change. I love my role here—I love exploring all of these ideas. For someone considering Adobe Research, or any new position, I’d say that it’s important to think about how the values and goals of an organization align with the things you care about. Research freedom and who your coworkers are is very important, too.

Related Posts