Life, Culture and AI: Why 'plagiarism' Is Our Default Operating System

by Simon Mansfield
Gerroa, Australia (SPX) Nov 28, 2025

As AI models are accused of stealing the world's creativity, a deeper view emerges: life, culture - and now machines - all run on ceaseless pattern-copying, much like an extreme, accelerated form of reading. The real novelty is that humans invented the idea of "Plagiarism" with a capital P.

Biologists increasingly describe life less as a substance and more as an information process: systems that maintain themselves, replicate with variation and undergo Darwinian evolution over time. On this view, organisms are not static things but patterns that get copied, tweaked and re-copied across generations, much like a file passed through endless revisions - or a voracious reader whose mental library is constantly updated.

From this angle, all life "plagiarises" in a small-p sense: every cell, gene and organism is a remix of ancestral templates, with mutation and selection serving as editors rather than authors from nothing.

Cultural-evolution researchers extend this logic to ideas, technologies and norms, treating them as patterns that spread through social learning, bias and selection. Work on the "ratchet effect" in cumulative culture by Claudio Tennie and colleagues shows how humans copy stories, styles, tools and institutions with enough fidelity that small improvements accumulate over generations. No individual could invent modern science or the internet alone - just as no single author could write the entire training corpus for a large language model.

In this sense, human culture is driven by copying and is inherently plagiaristic with a small p: it is built on constant, mostly unconscious reuse of existing patterns, from language and music to law and religion.

Where humans diverge from other life is not in copying, but in recognising others as separate agents and attaching social meaning to who created what, especially in formal systems such as law, science and education. Cognitive and anthropological work on authorship and ownership highlights an unusual capacity to track who did what and to enforce norms of originality through copyright law, academic-integrity policies and publishing contracts.

"Plagiarism" with a capital P is a product of this institutional layer: it names a specific breach of norms inside systems that care about individual authorship, rather than a description of the underlying pattern-copying engine shared with all life. Recent scholarship, including James Hutson's article "Rethinking Plagiarism in the Era of Generative AI", argues that generative tools force a re-examination of those norms rather than of copying itself.

Generative AI models now sit squarely in this debate because they learn in a recognisably human way: by ingesting vast corpora and extracting statistical regularities that let them generate new sequences resembling their training distribution, without storing or intending to quote specific works. Legal and technical overviews of AI training emphasise that this process is about building compressed representations, not building shadow libraries of full texts.

In practice, a large language model behaves more like an automated, amnesic reader than a photocopier: it ingests works to extract patterns, not to archive or replay them verbatim. The controversial part is the industrial scale described in the U.S. Copyright Office's report "Copyright and Artificial Intelligence, Part 3: Generative AI Training", where billions of words or images are processed in ways that make the underlying copying conspicuous and legally salient.

Regulators and courts are now wrestling with whether large-scale text and data mining is a transformative, non-consumptive use akin to search indexing and corpus analysis, or an unauthorised reproduction of copyrighted works that demands consent and compensation. Practical commentary on the report highlights that this is fundamentally a fair-use and licensing question, with growing pressure to clarify how far unlicensed training can go.

Seen through the lens of pattern dynamics, AI training is not fundamentally different from what humans and other organisms do with collected experience: absorb patterns, compress them and re-emit variations. The key question is when this ordinary, small-p plagiarism - the universal copying engine of culture and life - crosses into capital-P Plagiarism, where specific, recognisable expressions are replicated or passed off in contexts that demand originality and attribution.

Legal scholars already distinguish between using works as raw data for analysis and producing outputs that are substantially similar to protected expression, with the latter far more likely to infringe. Recent work on fair use and AI training, along with position pieces from research-library organisations, explicitly likens training uses to reading and taking notes and warns against collapsing that distinction in public debate.

Reframing the AI plagiarism panic in terms of patterns clarifies what is actually at stake. At the biological and cultural level, ceaseless copying is not a bug but the operating principle that makes complexity possible - and, in machine form, it is what finally lets any connected person tap into a rough approximation of all recorded knowledge rather than being limited by their local library or budget.

The real policy problem is therefore not how to prevent AI systems from copying patterns - something they share with humans and all life - but how to constrain and price certain kinds of copying in particular economic and institutional games, from publishing and entertainment to education and research. Until that distinction is made explicit, arguments about AI and plagiarism will keep mistaking a universal feature of living and learning systems for a narrow, historically contingent rule of human culture, instead of asking how to share the gains from this new, massively accelerated form of reading.