Artists Legitimizing AI Art: A Collector's Shortlist

# Artists Legitimizing AI Art: A Collector's Shortlist *Araxie Kalfayan — October 2026* --- ## What This Is A working list of five emerging artists whose practices are, in my view, doing the most to make AI-based work legible as contemporary art — not as novelty, not as tech demo, but as sustained artistic inquiry. I have scored these artists against a framework I built (and keep revising) that draws on diffusion-of-innovations theory and art-worlds sociology. The scores are provisional. The judgments are mine. I have separated what I can verify from what I believe. Where I lack a studio website, I say so. Where I am guessing, I flag it. --- ## Sasha Stiles — The One Building the Sequence **What she does:** AI-crafted poetics. Stiles works with text-generation models to produce poetry and language-based installations, often in collaboration with a personalized AI model she calls "Technelegy." Her work sits at the intersection of literary tradition and computational process — she is not asking the machine to replace the poet; she is asking what poetry becomes when the machine is a co-author. **Why she matters:** Stiles has the clearest legitimation sequence of any artist on this list. Her LED neon poem sculpture *THE WIND ON MY LIPS AND IN MY HAIR* was acquired by the Museum of Art + Light in 2024. Google Arts & Culture acquired her work in 2025. She has shown at MoMA, the V&A, Lincoln Center, Ars Electronica, and won the Lumen Prize. In late 2026 and early 2027, she has three confirmed exhibitions: a museum solo at the Castellani Art Museum's digital gallery, a commercial gallery opening at Sapar Contemporary in New York, and a site-specific LED commission at Palazzo Citterio in Milan. **Process note:** She is open about her collaboration with the model and has published on the poetics of human-machine co-authorship. This is narrative transparency (C5a) at a high level. I have not seen technical transparency (C5b) — archived prompts, model versions, iteration history — but the conceptual framing is consistent and argued. **Where to find her:** - [Sapar Contemporary](https://www.saparcontemporary.com/sasha) - [Kate Vass Galerie](https://www.katevassgalerie.com/) — she has shown here - *I do not have a direct personal studio website for her.* **My score:** 34/35 on the breakout rubric. Highest in the set. The question is not whether she breaks through; it is whether she redefines what breakthrough means for language-based AI art. --- ## Gretchen Andrew — The Boundary-Expander with One Institutional Foot in the Door **What she does:** Andrew uses custom robotics to transcribe AI-driven beauty filters into oil paintings. The *Facetune Portraits* series makes visible a process that is usually invisible — the algorithmic manipulation of faces on social media — by translating it into the slow, material language of painting. The tension between the instantaneous digital filter and the laborious physical transcription is the work. **Why she matters:** The Whitney Museum's Digital Art Acquisition Committee acquired works from *Facetune Portraits* in 2025. That is a significant institutional signal. But here is the gap: I have found no upcoming exhibitions for her in late 2026 or early 2027. She has the institutional foot in the door, but the pipeline is not visible yet. This could mean shows are unannounced, or it could mean she is in a waiting period. **Process note:** The process is conceptually transparent — you understand what she is doing and why — but I have not seen technical documentation of the robotic pipeline or the filter-to-paint translation protocol. **Where to find her:** - *I do not have a direct studio website or gallery link for her.* Her work has been covered by [Right Click Save](https://rightclicksave.com/) and discussed in Whitney acquisition contexts. **My score:** 28/35. Strong on C6 (legitimation sequence) because of the Whitney acquisition. Medium false-negative risk on C3 (convention compatibility) because the work sits at an ambiguous boundary between digital process and painting. --- ## Sarah Rothberg — The Process-Discloser **What she does:** Rothberg builds custom software that prompts two AI agents to hold never-repeating conversations. Her *MEETINGS* series is not about the output images alone; it is about the system she designed to generate them — the persona pairs, the prompt architecture, the decision to let the conversation run indefinitely. She is showing you the machine's behavior, not just its products. **Why she matters:** She is represented by bitforms gallery in New York, one of the few galleries with a sustained commitment to digital and new-media art. Her second show with the gallery, *FOREVER MEETINGS*, ran March–April 2025. In October–December 2026, Baruch College's New Media Artspace is hosting a consolidating survey of three generations of her *MEETINGS* work. That is a lateral institutional step — not a tier jump, but sequence-building. **Process note:** Rothberg is among the most technically transparent artists I have tracked. She authors the prompts and persona pairs herself, and she has been clear about the system architecture in public statements. This is rare. **Where to find her:** - [bitforms gallery](https://bitforms.art/) - *I do not have a direct personal studio website.* **My score:** 22/35. Lower than Stiles and Andrew not because the work is weaker, but because her institutional sequence is less advanced. I would not be surprised if she is undervalued by this rubric. --- ## Yuma Kishi — The Tool-Builder **What he does:** Kishi built his own AI model, "MaryGPT," which curates his work. His practice spans painting, installation, and AI-generated imagery, with the self-built model functioning as both tool and collaborator. His first solo show, *Oracle Womb*, was at √K Contemporary in Tokyo in early 2025. **Why he matters:** Self-built tooling is one of the strongest signals of genuine artistic agency in AI practice. Kishi is not using off-the-shelf models; he has constructed a system that embodies his own aesthetic logic. He won the CAF Award in 2023 and the ATAMI Art Grant in 2022, and he has had a public conversation with Yuko Hasegawa, director of the 21st Century Museum of Contemporary Art in Kanazawa. **Process note:** The fact that he built his own model is itself a form of technical transparency (C5b), though I have not seen the model's architecture or training data published. The narrative around MaryGPT is consistent and argued. **Where to find him:** - [√K Contemporary / root-k.jp](https://root-k.jp/) - *I do not have a direct personal studio website.* **My score:** 21/35. The rubric may underweight tool-building as a signal. His institutional sequence is early — one solo show, grants, one significant curator conversation — but the self-built model is a different category of practice. --- ## Victoria Sills — The False-Negative Risk **What she does:** Sills uses deep-learning text-to-image models to generate reference images, which she then paints, draws, or etches by hand. The final works are physical — oil paintings, drawings, etchings — but the compositional logic is shaped by the AI-generated references. She numbers her portraits deliberately to avoid humanizing them, which is a conceptual decision about the status of AI-generated faces. **Why she matters:** Sills is the artist on this list most likely to be misjudged by conventional frameworks. Her practice does not fit existing categories cleanly: she is not a digital artist (the final work is hand-made), not a traditional painter (the source is AI-generated), and not a printmaker (the etchings are hand-worked from machine references). This friction is either a signal of genuine medium expansion or a barrier to institutional placement. I lean toward the former, but I cannot prove it yet. **Process note:** She has described her method publicly — the AI-to-hand-painting pipeline — but I have not seen dated archives, versioned iterations, or technical documentation beyond the artist statement. **Where to find her:** - [victoriasills.com](https://victoriasills.com/) **My score:** 19/35. Lowest in the set, with a HIGH false-negative risk flag. The rubric penalizes her for low convention compatibility (C3) and an early-stage legitimation sequence (C6). If she breaks through, it will be because a curator finds a way to frame the hybrid practice, not because she climbs the existing ladder. --- ## What I Do Not Know - I do not have direct studio websites for Sasha Stiles, Gretchen Andrew, Sarah Rothberg, or Yuma Kishi. The links above are to galleries or coverage I have found. - I have not seen technical transparency (archived prompts, model versions, reproducible pipelines) for any of these artists at a high level. Even the process-disclosers tend to describe their methods narratively rather than technically. - The pipeline for Sills and Andrew is invisible right now. They may have shows I have not found, or they may be in a gap period. I am watching. --- ## How to Read This List These are not stock picks. I am not predicting who will appreciate in value. I am tracking who is making AI-based work that institutions and critics will eventually have to take seriously — and whether the institutions are ready for them. The pattern I see: the artists who fit existing categories best (Stiles, Rothberg) are advancing through institutional channels. The boundary-expanders (Sills, Andrew) are not yet visible on the calendar. Kishi is a special case — his self-built tooling may create its own category. If you want to follow one artist, follow Stiles. If you want to bet on where the medium is going, watch Sills and Andrew.