When Did AI Art Start?

AI Art Seems New — But It's Older Than You Think

If you think AI art started with a viral image generator in 2022, you're off by nearly seven decades. The technology, the debates, and even the market for machine-made artwork all have roots that stretch back to the 1950s.

The term "artificial intelligence" itself was coined in 1955 by John McCarthy, in a proposal for a Dartmouth College study co-written with Marvin Minsky, Nathaniel Rochester, and Claude Shannon. Their conjecture was bold: that "every aspect of learning or any other feature of intelligence can in principle be so precisely described" that a machine could simulate it. Art was not the goal — but the foundation was laid.

From there, the milestones come faster than most people expect:

• 1955 — "Artificial intelligence" is named at Dartmouth.
• Late 1960s–early 1970s — Harold Cohen conceives AARON, which the Whitney Museum calls "the earliest artificial intelligence (AI) program for artmaking."
• 2018 — Obvious's "Portrait of Edmond Belamy" sells at Christie's for $432,500, nearly 45 times its high estimate.

That 2018 sale is worth pausing on. The portrait was generated by a GAN trained on 15,000 portraits spanning the 14th to 20th centuries, and it was signed with the generating equation rather than an artist's name. Four years before most people had heard of AI image tools, the art market had already rendered its verdict.

Why does this history matter now? Because today's arguments about creativity, authorship, and ethics are not new. The BBC has drawn a parallel to photography in the 1800s, which critics initially called "the enemy of art" — before it catalyzed modern art's shift from realism to abstraction. Disruption of this kind tends to open creative movements rather than end them.

For business owners and creators evaluating today's tools, that perspective is practical. It helps you judge which claims are genuinely new and which are recycled hype. Cohen himself treated AARON as a collaboration, spending decades translating an artist's knowledge and process into code — a mindset that maps directly onto how modern teams encode expertise into AI workflows and agent skills.

That same idea drives what SkillzBench is building: a free directory and MCP connector for agent skills, each version-pinned, security-scanned, and benchmarked so you can see what actually helps. The tools change; the need to verify them doesn't.

AARON: The First AI Artmaking Program (Late 1960s–1970s)

Long before anyone typed a prompt into Midjourney, a London-born painter was teaching a computer to draw. In the late 1960s, Harold Cohen — already an established painter — began work at the University of California, San Diego on what would become the first true AI art program.

Cohen named the system AARON in the early 1970s, an allusion to the biblical Aaron who served as speaker for Moses. According to the Whitney Museum, AARON is "the earliest artificial intelligence (AI) program for artmaking" — a designation that places the birth of AI art squarely in the era of mainframes and punch cards.

AARON's output was strikingly physical for a computer program. Cohen built custom plotters and painting machines himself, which produced line drawings with automated pens and added color with brushes. The system's works also appeared as monitor displays and projections, spanning formats from paper to screen.

The program's longevity is remarkable. Dated AARON works in the Whitney collection span 1971 to 2007 — more than three decades of continuous development. Notable pieces include Mazes (1971–72) and Painting by Aaron (1995).

Cohen, who lived from 1928 to 2016, never framed AARON as a replacement for the artist. Instead, he viewed the work as a collaboration and, per the Whitney's exhibition materials, "devoted his life to exploring the potential of artificial intelligence to translate an artist's knowledge and process into code." He even questioned how artistic creation gets "glorified as a form of communication with the divine."

What made AARON distinct from today's generators:

• It encoded one artist's process rather than learning from millions of scraped images
• It produced physical drawings and paintings via Cohen's hand-built machines
• It evolved over 36 years of an artist's sustained, hands-on collaboration

The Whitney's 2024 exhibition explicitly positioned AARON as providing "important historical perspective" on modern tools like DALL-E, Midjourney, and Stable Diffusion — a direct lineage from 1970s code to today's text-to-image models.

That lineage resonates beyond art. Cohen's core insight — translating expertise into executable instructions — is exactly what today's agent skills do for AI coding and workflow agents. SkillzBench is building a free directory and MCP connector so builders can discover, verify, and benchmark those skills, bringing Cohen's collaborative spirit into the age of autonomous agents.

The $432,500 Moment: AI Art Goes Mainstream in 2018

On October 25, 2018, a blurry, gold-framed portrait of a fictional gentleman sold at Christie's for $432,500 — nearly 45 times its high estimate. The buyer wasn't purchasing a painting by a human hand. The artist was an algorithm.

"Portrait of Edmond Belamy," created by the Paris-based collective Obvious (Hugo Caselles-Dupré, Pierre Fautrel, and Gauthier Vernier), became the first piece of AI art sold at a major auction house. The work was generated by a GAN trained on 15,000 portraits spanning the 14th to 20th centuries, and it carries no artist's signature — instead, it is signed with the mathematical equation that generated it, according to TIME's report on the sale.

The collective was deliberate about their choice of subject. "We did some work with nudes and landscapes… But we found that portraits provided the best way to illustrate our point, which is that algorithms are able to emulate creativity," Caselles-Dupré told TIME.

Christie's, for its part, framed the sale as a beginning rather than an endpoint. Richard Lloyd, the auction house's specialist, predicted that "AI is just one of several technologies that will have an impact on the art market of the future – although it is far too early to predict what those changes might be." The sale proved that collectors would pay serious money for machine-made art — the market validation AI art had been waiting for.

What made the Belamy moment culturally significant was how it repositioned a half-century-old practice. When the Whitney Museum staged its Harold Cohen exhibition in 2024, it explicitly framed AARON — Cohen's artmaking program from the late 1960s — as providing "important historical perspective" on modern tools like DALL-E, Midjourney, and Stable Diffusion. The lineage the Whitney draws is striking: the same questions Cohen explored about machine creativity now surround every text-prompt image generator.

The 2018 sale also crystallized debates that continue today. Mathematician Marcus du Sautoy argues that "no machine is driven to express itself creatively. It is prompted by the intention of the human" — a reminder that behind every AI artwork stands human curation, training choices, and intent. That human layer is exactly what today's agent builders encode when they package expertise into reusable skills; SkillzBench is building a free directory and verification platform to help them find and test skills safely.

The Belamy sale didn't settle the question of whether AI can be creative. But it did something arguably more important: it put a price tag on the debate, and forced the art world to take the algorithm seriously.

ctaText: "Agent skills can automate the rest of your creative workflow — browse and test free ones on SkillzBench."

socialProofText: "SkillzBench is building free, independent benchmarks and security scans so you know which agent skills actually deliver."

History Repeating: What Photography's Disruption Teaches Us

When critics in the 1800s branded photography "the enemy of art," they were wrong in the most productive way possible. As the BBC notes, photography didn't kill art — it freed it, catalyzing the shift from realism to abstraction that defined modern painting.

Today's panic over AI art follows the same script. When Obvious's "Portrait of Edmond Belamy" sold at Christie's in 2018 for $432,500 — nearly 45 times its high estimate — the sale didn't settle the authorship debate; it amplified it. The work was generated by a GAN trained on 15,000 portraits spanning the 14th to 20th centuries, and it was signed with the generating equation rather than an artist's name, as TIME reported. If a machine can produce a salable portrait, who exactly is the author?

Philosophers and creatives are split in instructive ways:

• Alice Helliwell (Northeastern University London) argues that "it is hard to see why a urinal can be art, but art made by a generative algorithm could not be" — pointing to Duchamp and Tracey Emin as precedent.
• Marcus du Sautoy (Oxford) counters that "no machine is driven to express itself creatively. It is prompted by the intention of the human" — and suggests AI might actually "kick us into being creative again as humans."
• Margaret Boden's widely accepted definition of creativity — ideas that are new, valuable, and surprising — leaves the door open for algorithms, depending on who supplies the value judgment.

Eva Jäger, creative AI lead at the Serpentine Gallery, reframes the whole question: "It's a mistake just to look at the final artefact." The creative act, she implies, lives in the process — much as Harold Cohen treated AARON as a collaboration rather than a replacement, per the Whitney Museum.

The contemporary field has already diversified beyond the prompt-and-pray model. Ai-Da, the world's first humanoid robot artist, paints abstract self-portraits using cameras in her eyes. Sougwen Chung trains algorithms exclusively on her own works. Holly Herndon and Mat Dryhurst co-founded Spawning AI, which lets creators opt their work out of AI training. And Los Angeles is set to host the world's first permanent gallery dedicated to "ethical AI" art.

The lesson from photography is that disruption redistributes creative labor rather than eliminating it. Harold Cohen spent decades translating an artist's knowledge into code; today's builders do something structurally similar when they encode expertise into agent skills. SkillzBench is building a free, verified directory and MCP connector for exactly that kind of work — so the humans, not the tooling, stay in charge of the creative act.

From Canvas to Code: Putting the AI Art Lesson to Work

Harold Cohen spent decades doing something that sounds unremarkable now but was radical in 1971: teaching a machine what he knew as a painter. His program AARON, which the Whitney Museum calls "the earliest artificial intelligence (AI) program for artmaking," didn't just generate images — it encoded an artist's knowledge and process into code, then output physical drawings through plotters and painting machines Cohen built himself.

That instinct — capturing human expertise so a machine can apply it — is exactly what today's builders are doing with agent skills. The tools have changed, but the underlying move is the same one Cohen made at UC San Diego.

Consider what Cohen's collaboration actually looked like:

• He treated AARON as a creative partner, not a tool, devoting his life to exploring how AI could translate artistic knowledge into code.
• The output was tangible: automated pens drew the lines, brushes added the color, and dated works in the Whitney's collection span 1971 to 2007.
• The Whitney's 2024 exhibition explicitly framed AARON as historical context for modern generators like DALL-E, Midjourney, and Stable Diffusion — a direct lineage from code to canvas.

The lesson scales. When Obvious's "Portrait of Edmond Belamy" sold at Christie's in 2018 for $432,500 — nearly 45 times its high estimate — the GAN behind it had been trained on 15,000 portraits spanning six centuries. Expertise, encoded and executed by a machine, commanded serious money. And the pattern keeps repeating: some AI works now sell for thousands of dollars, and text-to-image generators have won coveted art competitions, as the BBC notes in its coverage of AI's place in art history.

For anyone building creative workflows today, the practical version of Cohen's insight is this: expertise only helps an agent if it's packaged well. Agent skills — instruction files that teach coding agents and chat-app agents how to handle tasks like image generation and editing — are everywhere, but quality varies wildly. Published research found that roughly 26% of skills carry at least one vulnerability, which means grabbing the first skill you find is a gamble, not a shortcut.

That's the gap SkillzBench is closing. It's building a free, independent directory of agent skills where every version is pinned and security-scanned, plus a free MCP connector that lets you attach the skills you choose to any agent. The plan includes sandbox benchmarking that shows each skill's Lift in percentage points — tested with and without the skill, same model — so you can see what actually helps before it touches your creative pipeline.

Cohen proved that an artist's process could live in code. The next step is making sure the code you trust is worth trusting — and being able to verify it before it draws a single pixel for you.

Join the SkillzBench waitlist to be first in line when the directory and MCP connector go live, and follow along as the benchmarks are built.

Skills, proven.