Humans creating, AI curating
AI researcher Laura Herman on algorithmic curation and empowering creative humans
Social media influencer? That’s one of those jobs that’s being lost to AI.
AI curators are influencing not only what we see, but what creators create. What you see on Instagram, for example, might give you a certain impression of what the art world is looking at and thinking about today. As a creator, it might influence work that you’re making.
Maybe, going further, you might create something specifically intended to appeal to the AI curator, in order to move up in the feed, to maximize your likes and followers, maybe go viral. The attention economy, says Laura Herman, is shifting our cultural norms.
In her work and research, Herman is uniquely qualified to provide insight into AI and how it’s impacting the creative sector.
We met at London’s Whitechapel Gallery, where she recently completed a curating course. This informed her PhD research at Oxford, which compares human and algorithmic curation, “specifically focused,” she told me, “on what I call curation systems or recommendation systems — not generative models. Thinking about Instagram or TikTok or Google Images, and how they collate and display our visual culture, and how that impacts creatives and what and how they make.”
Branding the artist
In a recent talk, Herman describes Instagram as a brand-building platform. “And when you’re building a brand, what you want to do is to very clearly correlate the same type with your entity over and over and over again.
“But as an artist, what you’re supposed to do… is post different things, new things, unique things, ultimately creative things.” Platforms might punish you for that.”
She staged an exhibition to explore this, and it provided some stark evidence for this. She took a subset of images from the Metropolitan Museum of Art’s online database, uploaded them to Instagram and also gave them to an artist — she wanted to see how each would curate the images: What would the algorithm push up to the top, and what would the human prioritize?
The difference, in short: faces. Instagram’s algorithm seems biased in favor of them, whereas the artist preferred more abstract imagery. (This explains why I don’t have a million followers.)
Herman is quick to point out that this wasn’t a controlled experiment; she mainly used it to provoke conversations with visitors, artists and others. “This was the most interesting part to me, because they looked at the images chosen and said things like, ‘Ah yes, I always feel this pressure to post images with my face in them’. As you can see, there’s all kinds of faces in the curation.”
But then she did a computational analysis of the images. “We had all these reactions from humans,” she told me, “But how does a machine perceive and interpret these? That’s been helping us understand why Instagram chose their images.”
This helped solve a mystery in her data.
“The number one image [Instagram] chose to put at the top is this blurry image of a black-and-white tapestry. We had no idea why — we repeated it three times, and each time, that was the number one image.
“But when we ran a computer vision protocol on it, it thought that there were nine faces in the tapestry. It was seeing this pattern, and mis-perceiving it as a face. And we know that Instagram tends to prioritise faces.
“So we’re getting a peek into how the machine sees. And when we understand how the machine sees, we can understand a bit more about how the machine makes decisions.”
Crossing the streams
How Herman herself sees is even more interesting, because she has synaesthesia. “Synaesthesia is this automatic connection between two different parts of the brain essentially,” she says. “For me, I see time oriented spatially around me. It’s not a building, but a similar sort of spatial memory.
“The other thing I have is letters and numbers to colors, which is super useful for remembering things, because everything’s automatically color-coded. I also have music to taste, which is fun. And shapes to colors.”
“Wow, that must be confusing,” I suggest.
“I’ve never known anything differently. Essentially, with synaesthesia, the way all of the experiences work is that it’s formed when you first learn. I remember learning the alphabet, and the colors were already there. They’ve stayed the same. So when I was young and learned about time, this is how it was oriented, and now it’s stuck there.”
This led her to study neuropsychology. “The reason I got so fascinated with research is that there’s so many open questions and very few answers. But when I talk about time, some people are like, “Oh yes, of course time is like this spatial thing that I see oriented around me.” People talk about a Ferris wheel or a timeline, and the question is how much of that is entrained, versus innate.”
“We talk about time moving forward,” she goes on, “but in cultures where they use Arabic for example, there are different concepts about the forwardness of time. So there are definitely cultural components to it, but they’re really hard to disentangle.”
Generation generative
Given all her experience, we couldn’t not talk about generative AI and its impact on creative practice.
“Generative AI is not the beginning, nor is it the end,” she said. “It’s a new form of creative expression. It’s a new tool in the artist’s toolbox.”
This tool changes the role of the human creator, she says. Specifically, because AI can produce high-fidelity ideas right at the start of a workflow, the human creator (now curator) can direct and select, she says, instead of sketching or even producing.
And direction and selection, she points out, are also what human curators do. But having a sense of subjective sensibility — taste — “is far too subjective for a variable-based mathematical system like an AI to have.”
I realised this myself when I started hanging out in artists’ studios: here was a space where someone can control — they can put up on the walls the things that inspire them, that they want to see every day.
I ask her about the ‘Boring Apocalypse’ scenario: If everyone uses AI to generate more content, the people receiving it have to use AI to digest it, and it just creates more work for everyone.
“Right, and no one’s actually doing anything anymore,” she adds. “So much visual content is being displayed on algorithmic platforms, and if the generative tools are the ones making that content, it’s algorithms curating algorithmic art — with humans as passive consumers.
“If you combine generative images becoming hyper-real, with platforms like Instagram prioritising what is ‘real’ — also Google Images: the top result will be the image that looks the most like the thing you search for.”
“And because more generative AI content is flooding the internet,” I add “and then AI systems are being trained on that data, quality goes down.”
“Which might be a bit of a saving grace for creatives,” she says, “because the generative tools might self-deconstruct a little bit if they continue scraping the web.”
