When Human Creativity Meets Technical Autonomy: An Interview With Artist Anna Ridler
Yesterday, the news dropped that digital art collector and patron Seedphrase (aka, Daniel Maegaard) acquired Price Per Stem, British artist Anna Ridler’s first commercially available dataset, released with Avant Arte, in its entirety. If you’re nonplussed at the idea of dataset-as-art, you likely haven’t been following the lively goings-on in the wide world of technology-driven art (taking a spin through Lumen Prize finalists and winners can help you catch up). For several years, the rank and file associated technology-driven art with goofy NFTs; more recently, the term might call to mind Refik Anadol’s increasingly mainstream A.I.-powered data paintings. But really, it’s much broader than that, encompassing everything from the early generative work of artists like Michael Noll, Béla Julesz and Frieder Nake to Miriam Simun capturing and synthetically producing the scent of the almost-extinct flower Agalinis acuta using chemical data.
Ridler, who like many other artists works at the intersection of art and technology, has chosen data as her medium. She is celebrated for her use of artificial intelligence and machine learning to create visually and conceptually rich works that examine the biases, histories and ethical implications embedded in these technologies—all built around painstakingly created datasets through which she explores human creativity, technological fallibility, ontology, natural processes and bias. Her work has been featured in prestigious exhibitions globally, including at the Barbican Centre and V&A Museum in London, the Centre Pompidou in Paris and Ars Electronica in Austria.
Myriad (Tulips), an installation work from 2018, consists of a dataset (or training set) of 10,000 tulips photographed and hand-labeled by Ridler. That became the basis of Mosaic Virus, a video work generated by A.I. using her dataset, and the 2019 NFT work Bloemenveiling, a technological marketplace where people could buy short GAN-generated videos with self-destroying tokens. Together, the three works explored the labor that goes into training A.I. systems, the relationship between human creativity and machine output and contemporary cryptocurrency speculation, linking it to 17th-century Tulip Mania.
Seedphrase’s latest acquisition, Price Per Stem, is actually a series of twenty-five artworks with both physical and digital elements: printed photographs of real peonies taken by Ridler and moving videos of the flowers dynamically annotated using an intricate model of peony price fluctuation. On the occasion of the acquisition, Observer caught up with the artist to learn more about her explorations into how input data impacts aesthetic outcomes and her thoughts about artistic integrity in a cultural landscape being shaped by increasingly autonomous creative forces.
Your practice involves handcrafting datasets—are you specifically interested in how human choices shape outcomes or is there another reason you’ve chosen laborious curation over automation?
A part of every project is always creating a dataset—a collection of information that can be read by a machine—as it is such an integral part of machine learning. A large part of my process is building things from start to finish. Without working with the dataset in some form, I would not be able to work through everything. The images and outputs that I build up from the data are mine and come from how I have seen and experienced the world; the errors and choices are mine and mine alone. Each image and caption is laboriously constructed to reflect this. This is why for me, a dataset is creative work and becomes an artwork in and of itself. I have been working in this way since I started making art with machine learning nearly a decade ago.
There is also something slightly absurd about doing this long process. Making this piece felt very extravagant, specifically buying hundreds of pounds worth of peonies just to take a single photo of them every few days—that is in direct opposition to the way that datasets are often created for commercial reasons (either hoovering up the internet without even looking at what is in there—just relying on whatever caption was nearby—or harvesting data) by slowly considering each flower and allowing them to naturally die and decay. The cost of it all is indirectly referenced in the title of the piece.
Peonies were a deliberate choice; they have an incredibly short season (this project started in April to catch it) and have a cache on the flower market. However, the price for them is changing. They need very specific conditions to grow (including a certain amount of time in frozen soil over the winter), which because of climate change is shifting where they can be grown (Alaska, for example, has become a major exporter of peonies) and this year the season was particularly short and early which had an impact on the number and type of flowers I could get.
Because my datasets are comparatively small, it also means there will be more slips and errors, which I embrace—a lot of generative A.I. has a shiny and slick quality to it, which to me is less interesting than when it fails.
When it comes to creativity, do you see there being any hard stops where automation or A.I. are concerned? Like, if an artist does X or Y, they’ve crossed a line where they’re no longer the agent of creativity?
It is different for every artist and artwork considering what their intention is and what they are trying to explore. A lot of the debate around creativity and A.I. goes back to the debate and anxiety around conceptual art in the ’60s (and earlier), and I think a lot of the things that I am interested in come from this lineage.
Does your work also comment on/address the broader implications of technology in art and perhaps also in society, particularly ethical questions surrounding data collection, ownership and consent?
Although this piece has elements of a dataset in it, it is not just a dataset. The moving image piece, partially made through generative A.I. using a combination of some of the many bespoke models I made for it (over 50), partially pulling through a financial model I constructed to estimate the price of a peony using various different factors (also using machine learning), partially overlaying hand-drawn elements of those generated images (so that the piece loops back from analog to digital to analog before settling in this hybrid form) more explicitly is interested in trade, globalization and the inherent ridiculousness of it.
SEE ALSO: Cecilia Vicuña On Embracing the Beauty of a Universal Order
By taking something as simple as a cut flower—which is sometimes seen as something just pretty or perhaps kitsch—and telescoping in to look at how they are bought and sold and traded and the dynamics of the cut flower market was something that I was always interested in doing with this piece; making my own dataset was a way to get there. Because it is made using A.I., there are echoes of how machine learning is being used increasingly in market trading. It also questions the value of these imperfect fake eternal flowers that are being sold and kept long after the real counterparts that existed to make them through the dataset have withered away and rotted.
Is there a difference between a collector acquiring a dataset-as-artwork versus some output generated using that dataset? How about between a digital dataset versus a physical representation of the dataset?
For me, the datasets that are read by machines (the raw photographs and/or text files or CSVs) are not displayed; what is always displayed is a representation of the dataset. This representation collapses some of the choices I have made (e.g., the photograph, the label) into one visual output. In the case of Price Per Stem, the artwork encompasses both part of the dataset and the output that has been trained on that dataset. For me, a dataset can be an artwork—though not always, it depends on the artist’s intention and how they choose to work with it—just as an output can be an artwork. Though again, not always, it depends on intention, so there is not that much difference in my head: they are both different ways of making work.
With Seedphrase acquiring Price Per Stem, is the implication that he could use the dataset himself as a generation tool?
As he has acquired a photographic installation, he would have to do a lot of work to make it into a dataset for use, scan each of the photographs, remove the handwriting, create his own system for labeling, fine-tune his own model, find an idea that he wants to work with this data on. He can, of course, do this, but there would be enough work that would go into it that it would stop being my work and perhaps become his own. I’m never just interested in using A.I. to make an image, I’m always more interested in using it to think through different ideas.
What’s next for you—more datasets or something else?
I always work with datasets! I’m particularly interested at the moment in exploring the very large datasets that power LLMs, etc., and seeing if I can unpick them, moving away slightly from me teaching a machine what the world is to me trying to understand how a machine teaching a machine sees things.