The reading room · Digna Legi
Arguments in Favor of AI Fair Use
/100
80–100: high value. 70–79: worth the time. Below 70: below the usual publication threshold.
Evidence-reviewed score based on available publisher text. The text intentionally excludes counterarguments, so it is not sufficient on its own for high-quality decision-making.
Scores reflect one reader’s profile, not an objective quality rating. Best is a separate personal selection.
How scoring works →This brief · about 3 min with detail
Why read this
Kelly defends AI training as fair use by treating latent space as transformative, non-reversible structure rather than stored copying.
AI brief · Checked against source text
The main idea
Kelly argues that large language model training should be treated as fair use because it changes human expression into latent space: an abstract, non-reversible structure that does not resemble stored copies. He adds that the resulting system is used for different purposes than the source works, gains value from distributed interactions among many sources, and could support public intelligence commons rather than only commercial products.
Go a little deeper
Transformation has two parts
The argument is not only that training changes form. Kelly says the model converts expression into latent space, and then users apply that transformed material to tasks unlike ordinary consumption: repairing devices, planning marketing, or proving math. The fair-use claim therefore rests on both technical transformation and a changed public use, not on one alone.
Non-copying depends on irreversibility
A central mechanism is asymmetry: source content can shape latent space, but latent space cannot simply be reversed into the original works. Kelly uses this to separate training from storage. The model may contain information derived from books or images, but he argues that its weights and latent structure do not amount to copies of those works.
Individual contribution becomes hard to price
Kelly’s valuation argument has two layers. First, most individual sources are highly redundant once added to a vast conceptual space. Second, when trillions of parameters interrelate, the marginal influence of one source becomes practically unmeasurable. The point is not that sources contribute nothing, but that ordinary attribution and compensation logic may break down at this scale.
The political endpoint is a commons
The final move shifts from commercial AI to public infrastructure. Kelly argues that if training remains fair use, society can build publicly funded and accountable language models trained broadly across texts and languages. In this framing, latent spaces are not just business assets; they become a new public domain that expands access to knowledge work.
A case from the article
Star Wars without the movies
Kelly uses Star Wars to show why source influence is not limited to primary works. A model might learn much of what it knows from reviews, commentary, and other secondary material rather than the films themselves. This supports his claim that indirect sources supply context and may do much of the work that makes model output useful.
How the case is made
The case is made as a think-aloud legal-policy argument using technical analogies, usage examples, redundancy claims, and public-domain reasoning.
Where the idea has limits
Kelly explicitly limits the piece: it is in the context of U.S. copyright law, omits many arguments for and against, and is not arranged as a tight case.
A question to take away · from Digna Legi
If influence cannot be practically traced, what should replace source-by-source compensation as the unit of fairness?
What the original adds
The full source supplies a broader catalog of pro-fair-use arguments, including transitional digital copies, redundant source value, indirect cultural influence, and public-interest models for non-commercial AI.
About this brief
AI-written, then separately checked for source support, useful detail and clarity. The author’s claims and our editorial question are kept separate. The original remains the author’s work. How we select and summarise →
How was this brief?
Rate this summary, separately from the author’s article.
Optional. Saved in this browser; shared only if you allow analytics.
How was the original article?
Rate the author’s original after reading it.
Optional. Saved in this browser; shared only if you allow analytics.
Digna legi. Worth reading.