AI thrives on human creativity, but risks eroding the very incentives that keep it alive—copyright must evolve to protect both
This essay is part of the series: World Creativity and Innovation Day 2026: Sparks and Shields
Creativity has long been considered a defining characteristic of human intelligence. Today, however, advances in Artificial Intelligence (AI) are beginning to blur this boundary. Generative AI systems can now write scripts, compose music, generate images, design animations, and mimic distinctive artistic styles. From publishing and advertising to film and gaming, AI tools are increasingly embedded in creative workflows.
AI systems are trained on vast amounts of human-generated content. Using processes such as text and data mining (TDM), they analyse millions of books, images, songs, and articles to identify patterns in language, structure, and style, allowing them to generate outputs that resemble creative works. This ability also enables them to produce creative outputs rapidly and at scale, lowering the cost of content production and expanding the tools available to creators.
This development has triggered an important debate. AI models are often trained on copyrighted works without the explicit consent of authors or rights holders. These systems can then generate content that competes with the very works that enabled their training. If AI systems can mimic artistic styles or generate substitute works without compensating creators, the economic value of copyrighted works may decline. Over time, weaker incentives could discourage the production of new creative content. At the same time, generative AI itself depends on a continuous supply of fresh human-created works. Without new creative material, the quality and diversity of AI-generated outputs may stagnate.
Generative AI creates an economic paradox: it depends on human creativity as training input while simultaneously undermining the incentives that sustain that creativity.
In short, generative AI creates an economic paradox: it depends on human creativity as training input while simultaneously undermining the incentives that sustain that creativity. In this context, this article examines the economic rationale behind copyright, how generative AI challenges it, and how weakening copyright protection could ultimately affect the future development of AI systems.
Creative works protected by copyright are considered non-rivalrous, meaning that one person’s use or enjoyment of a work does not diminish the ability of others to use it. Once produced, copyrighted works can therefore be reproduced and consumed by many people simultaneously without being depleted. This creates a fundamental economic challenge: while the cost of producing the original work is often high, the cost of making additional copies is extremely low. In the absence of adequate protection, such works could be copied and distributed freely, making it difficult for creators to recover the costs of production. In economic terms, this creates a classic free-rider problem: once creative works are produced, others can copy and distribute them at very low cost, potentially undermining the incentives needed to sustain creative production.
Copyright attempts to address this problem by granting authors exclusive rights over the reproduction, distribution, and licensing of their works for a limited period. These rights enable creators to derive economic returns from their labour and investment. When individuals know they will retain control over the outputs of their labour, they are more likely to invest time and resources into producing new works. Therefore, copyright rights allow creators to “reap where they have sown”. Without such assurances, the incentive to create may weaken.
Beyond economic incentives, other theories also justify copyright protection. One view is based on the natural property right of a person over the fruits of their creation, suggesting that authors have an inherent right over their works. Another perspective emphasises the moral rights of authors and that creative works are extensions of an author’s personality. A third argument focuses on rewarding creators for their contribution to society, with copyright functioning as a temporary monopoly that allows authors to receive compensation for their work. At its core, copyright, like all other intellectual property rights, is a tool for transforming knowledge and creativity into economic value. Copyright protection ultimately improves social welfare by encouraging the production of creative works that might otherwise not be produced.
AI significantly challenges traditional understandings of copyright. As mentioned earlier, many AI systems are trained on vast datasets that include copyrighted material, often without the explicit consent of authors or rights holders. This has raised concerns about whether the use of copyrighted works for AI training should require permission or compensation. At the same time, the rise of generative AI has created new questions for copyright regimes regarding how outputs produced by AI systems should be treated and whether such works should qualify for protection. While these are important policy questions, this article looks beyond them to examine how AI fundamentally challenges the underlying rationale of copyright.
If the growth of AI-generated content weakens incentives for artists and reduces creative production, it could ultimately undermine the quality of AI systems themselves.
When AI systems are able to produce creative outputs that compete with those produced by human artists, the two become increasingly substitutable in certain markets. This increases the overall supply of creative content, which may reduce the price consumers are willing to pay for individual works. At the same time, the economics of production change. While training AI models requires significant upfront investment, once trained, generating additional content, whether text, images, or music, becomes extremely inexpensive. In effect, AI dramatically reduces the marginal cost of creative production.
As the cost of producing new works falls and the volume of content increases, the competitive dynamics of creative industries begin to shift. AI-generated content may substitute for human-produced works in areas such as stock imagery, marketing copy, or basic design tasks, increasing competitive pressure on creators.
At the same time, AI can function as a productivity-enhancing tool, enabling artists and writers to experiment more rapidly and reduce the costs of creative production. However, this may also reduce reliance on skilled human creators in certain areas, potentially weakening incentives for artistic skill development and creative labour. As the expected economic returns from creative work decline, individuals may invest less time and resources into producing new works. Over time, this could reduce the supply and diversity of high-quality creative output.
The rapid expansion of AI-generated creative content also raises deeper concerns about the long-term dynamics of AI development. Machine learning models depend on large and diverse training datasets, most of which consist of human-created works accumulated over decades of creative products. The availability of such data is not automatic; it is sustained by the economic and institutional structures that incentivise individuals to produce new cultural material.
This creates a structural dependency between AI systems and human creativity. If the economic incentives that sustain creative production weaken, particularly in areas where AI-generated outputs substitute for human work, the production of new material may decline. In turn, the pool of high-quality training data available for future AI models may gradually shrink.
Copyright in the age of AI must both sustain human cultural production and enable the continued development of artificial intelligence.
Over time, this dynamic could lead AI systems to rely increasingly on recycled datasets or synthetic content generated by earlier models. Researchers caution that such feedback loops may produce “model collapse”, where models trained on AI-generated outputs progressively lose diversity, originality, and informational richness. In this sense, generative AI cannot sustain itself indefinitely on synthetic data alone. Its long-term development ultimately depends on maintaining the human creative ecosystem that supplies the cultural material on which these systems are trained.
Some scholars argue that creative production is often driven by motivations beyond financial reward, such as reputation, prestige, and intrinsic satisfaction, economic incentives remain important in sustaining cultural output. If the growth of AI-generated content weakens incentives for artists and reduces creative production, it could ultimately undermine the quality of AI systems themselves. The policy challenge, therefore, lies in balancing the interests of creators and the needs of AI development.
A regime that allows unrestricted use of copyrighted works for AI training may weaken incentives for creators, while overly restrictive rules could limit access to the data required to train advanced AI systems. One possible pathway lies in hybrid regulatory approaches that combine access with compensation. Mechanisms such as collective licensing systems, statutory remuneration frameworks, or greater transparency around training datasets could allow AI developers to access large datasets while ensuring that creators receive fair recognition and compensation for the use of their works. Ultimately, the challenge lies in balancing creativity and incentives: copyright in the age of AI must both sustain human cultural production and enable the continued development of artificial intelligence.
Basu Chandola is an Associate Fellow at the Observer Research Foundation.
The views expressed above belong to the author(s). ORF research and analyses now available on Telegram! Click here to access our curated content — blogs, longforms and interviews.
Basu Chandola is an Associate Fellow at the Observer Research Foundation, where his work focuses on the governance of the digital economy, including artificial intelligence, ...
Read More +
PREV
