Is AI art theft?

One of the most contentious claims in the AI art debate is that generative AI systems are, at their core, engaged in theft. Critics argue that by training on billions of artworks without explicit permission, these systems are stealing the creative labour of human artists. This accusation demands scrutiny – and as we'll see, it fundamentally misunderstands how these systems work and the nature of artistic influence itself.

How diffusion models create images

To address the theft accusation, we must first consider how diffusion models like Stable Diffusion or Midjourney function.

These systems do not store a library of images they've seen. Instead, they learn abstract statistical patterns from their training data. The training teaches the model to use the statistical patterns over hundreds of small steps to create images in response to input prompts.

When generating a new image, the model:

1. Starts with pure random noise – not a blank canvas or an existing image

2. Progressively denoises this randomness according to learned patterns

3. Uses text prompts to guide which patterns to apply

4. Follows probabilistic paths rather than deterministic recreation

I will dive deeply into this process in a later blog post. 

This mathematical process makes direct copying of training images virtually impossible. The model stores weights and statistical relationships, not images. This is fundamentally different from sampling or collage – the AI isn't cutting and pasting elements from existing works; it's creating something new based on its understanding of visual concepts.

Dismantling the "Theft" Arguments

Let's examine the most common arguments claiming AI art constitutes theft:

Argument 1: "AI copies artists' styles without permission"

This argument suggests that an AI's ability to create images in the style of specific artists constitutes theft of intellectual property.

Why this fails: Artistic styles have never been protected by copyright law, and for good reason. Human artists have always learned by studying, emulating, and transforming other artists' styles. Van Gogh studied Japanese prints, African masks influenced Picasso, and countless contemporary artists built on established styles. When a human painter studies Monet and creates impressionist landscapes, we don't call it theft – we recognise it as part of artistic tradition.

Imagine the chaos if styles were legally protected: Artists would need to seek legal clearance before painting in well-known styles. Artists would always be concerned that an artist they had never heard of might challenge their next work in court. They would be forced to take indemnity insurance to protect against legal claims. New artists would be discouraged from joining the profession. We could imagine the growth of an industry making fake copyright infringement claims against artists in the hope that artists would pay rather than be dragged through the courts. It would not benefit artists or art lovers – only lawyers and insurance companies would gain.

Imagine if the principle that art styles can be copyrighted were extended to other areas. Architects could be sued for designing buildings with similar aesthetic sensibilities to existing structures. The entire ecosystem of creative arts could collapse.

The history of art is a conversation between artists across time. AI systems have joined this conversation.

Argument 2: "Artists didn't consent to having their work used for AI training"

This argument contends that using publicly available artwork for machine learning without explicit permission constitutes theft.

Why this fails: When artists place their work in public galleries, websites, or publications, they implicitly consent to having their work viewed and learned from. We don't require human artists to obtain permission before visiting museums or browsing art websites for inspiration. Learning from observation is fundamentally how all human or artificial intelligence develops.

No artist has ever been required to document which works they've studied, nor do we expect them to seek permission from other artists before viewing their publicly displayed work. Applying a different standard to AI systems creates an irrational double standard.

Argument 3: "AI companies profit from artists' work without compensation"

This argument suggests that commercial AI companies are monetising training data created by artists who receive no compensation.

Why this fails: This argument conflates economic impact with theft. Artists have always understood that their publicly displayed work might inspire others who could subsequently profit from that inspiration. When artists display work in a gallery, they know other artists might view it, learn from it, and create commercially successful work influenced by it.

In reality, for AI art generation, the diffusion of influence across billions of training images means the impact of any single artwork is minimal. Thus, any artwork will have less influence on an AI than a human apprentice who studies intensively under a single master or an artist who admires the work of a fellow artist.

The practice of artists benefitting from the work of earlier artists without paying compensation is long established and uncontentious. It is perverse and inequitable to demand that AI art compensate artists when it makes minuscule use of any artwork. 

It might at least be fair if human artists were also obliged to compensate the creators they were inspired by. However, imagining how such a system would work is difficult. Any system to collect money and pay artists worldwide would inevitably be complex, contentious, and unpopular. Since we have managed without such a system until now, the argument for introducing a system is extremely weak.

Argument 4: "AI devalues human creativity"

Some argue that AI art tools, by making image creation more accessible, devalue the uniqueness of human artistic expression.

Why this fails: This is an argument about market impact, not theft. New technologies have repeatedly transformed creative fields, and predictions of creative apocalypse have proven wrong each time. Instead, these tools have consistently expanded creative possibilities and allowed more people to participate in artistic creation.

The value of art has never been solely about technical difficulty. We wouldn't value Rothko's color fields or Pollock's drip paintings if it were. Art's value lies in its ability to move us, challenge us, or make us see the world differently – qualities that remain relevant regardless of the tools used to create it.

Learning vs. copying

The distinction between learning and copying is often overlooked in this debate. When humans observe the world – including art – we don't make perfect internal copies. We abstract patterns, recognise relationships, and integrate what we've seen into our mental models. These models then inform our own creative outputs.

AI systems, in their computational way, follow a similar pattern. They extract statistical relationships from observed data, build abstract representations of visual concepts, and generate new outputs influenced, but not determined, by what they've "seen."

This process mirrors how human creativity has always functioned. No artist creates in a vacuum – we all build upon the accumulated cultural knowledge of those who came before us. Every artistic movement, from Renaissance painting to jazz music, emerges from this process of learning, transformation, and creation.

The path forward

The most productive path forward isn't found in misguided accusations of theft or attempts to restrict this technology. Instead, it lies in embracing AI as a new creative tool with unique capabilities and limitations.

Just as photography created new forms of visual expression without eliminating painting, AI art offers new creative possibilities without replacing human artistry. Artists who view AI as a collaborative tool rather than a threat are already creating works that wouldn't have been possible with either human or machine working alone.

Our real challenge isn't protecting existing creative practices from theft (a premise that doesn't hold up to scrutiny). Instead, it's about navigating the broader societal transformation as AI technology reshapes creative industries and virtually every domain of human activity.

How will we distribute the unprecedented productivity of this technological revolution? How will we redefine meaning and purpose when many traditional forms of work – not just in art – become automated? These questions extend beyond art and demand sophisticated social solutions, not technological retreat.

Conclusion

AI art isn't theft – it's the latest chapter in humanity's ongoing exploration of creative expression. Like photography before it, it will ultimately be judged not by the controversies of its early days but by the new forms of beauty and meaning it enables us to create.

As an AI artist, I create with confidence that my work stands on solid ethical ground. I use these tools to extend human creativity, not to replace or devalue it. The result is a new form of collaborative creation that honours a rich tradition.