Can intellectual property law apply to unfixed, context-dependent outputs?
This reads the legal question — can copyright-style protection attach to outputs that aren't fixed and change with context — through the corpus's work on why AI outputs are mutable, authorless, and flow-like rather than fixed objects.
This explores whether intellectual property law, which historically protects fixed and authored works, can attach to AI outputs that change every time you run them. The corpus doesn't argue case law directly, but it maps the exact ground that makes the legal question hard: it shows that AI outputs fail two of the deep assumptions IP rests on — fixation and authorship.
Start with fixation. Copyright traditionally attaches to a work fixed in a stable form. But the corpus argues that AI outputs are *essentially* mutable: they vary with sampling, prompt wording, and even how the audience reads them, and this variability is a defining property of tokens-as-media, not a bug to be smoothed away Why does AI output change with every prompt and context?. The substrate underneath is itself unstable — prompt, history, retrieved data, and hidden state shift constantly, so there's no fixed 'context' to anchor an output to How does AI context differ from conventional software context?. And because a single transformer is effectively programmable by its prompt Can a single transformer become universally programmable through prompts?, the 'work' isn't a thing the model stores — it's a draw conjured at request time. There's a useful historical reframe here: print culture turned knowledge into fixed, ownable *stock*, while AI returns it to generative *flow* — and IP law was largely built for the stock era Is AI returning knowledge to flow-based economies?.
Now authorship. IP needs an author to own the work, and here the corpus is sharp: users *declare* authorship socially while not actually experiencing cognitive ownership of what the AI produced — the intermediate steps are opaque and the sense of having authored it is constructed after the fact Do users truly own the AI-generated content they produce?. So even the human in the loop is a shaky candidate for 'author.' And the model isn't doing what a human author does at all — it emits strings from a probability distribution rather than using language to mean or relate to anyone, which makes treating its output like an authored expression a category error Are language models and human speakers doing the same thing?. The flow framing sharpens this: AI circulation lacks the embodied carrier — the speaker, the giver — that historically anchored who 'owned' or transmitted knowledge Is AI returning knowledge to flow-based economies?.
There's a second, less obvious wrinkle the corpus surfaces: even an AI output's *value or meaning* isn't intrinsic to the output. The same generated artifact should be read not as ground-truth content but as a subjective prior shaped by the prompter's choices Should we treat LLM outputs as real empirical data?, and quality itself turns out to be situational — an explanation, for instance, has no fixed worth independent of who presents it, how it's framed, and who receives it What if XAI is fundamentally a communication problem?. If the protectable 'thing' only stabilizes in a particular rhetorical situation, IP's premise that you protect a self-contained object gets harder to sustain.
The thing you might not have known you wanted to know: the AI copyright debate is usually framed around *training data* (was your work scraped?), but this corpus points at the harder, downstream puzzle — the output side. If a work has no fixed form, no stable context, and no author who genuinely experiences having made it, the question stops being 'who infringed?' and becomes 'is there even a *work* here for the law to grab?' The corpus suggests IP's traditional machinery may simply not have a handle to hold.
Sources 8 notes
AI outputs exhibit essential mutability—they vary with sampling, prompt wording, and audience interpretation. This is not a defect but a defining feature of tokens as media, making them fundamentally different from fixed commodities and resistant to traditional quality assurance.
AI interactions operate on a substrate of constantly shifting context—prompt, history, retrieved data, hidden state—that users cannot internalize like traditional UIs. This structural mutability demands a new design discipline centered on context engineering rather than interface design.
Research proves a single finite-size transformer exists that can compute any computable function given the right prompt, achieving complexity bounds nearly matching unbounded models. However, standard training rarely produces models that learn to implement arbitrary programs this way.
Print culture fixed knowledge as accumulated stock; AI returns knowledge to generative flow. However, unlike oral and gift economies, AI flows lack the embodied transmission—the speaker, the giver—that historically anchored knowledge circulation.
Research shows users declare authorship at a social level while lacking genuine cognitive ownership of AI-generated content. This dissociation arises from opaque intermediate steps and post-hoc narrative construction, not dishonesty, and leads to inflated self-assessments of independent competence.
LLMs produce strings via probability distributions; humans use language to address and relate to others. They share surface form but differ in what produces output, what it does socially, and what receivers should do with it.
Foundation Priors framework shows that LLM-generated text reflects the model's learned patterns and user's prompt choices, not ground truth. Such outputs should only influence inference through explicitly parameterized trust weights, not be treated as equivalent to real evidence.
Explanation quality is not intrinsic to the explanation itself but depends on the rhetorical situation: who presents it, how it is framed, and what role the recipient plays. Evaluations that ignore this triad measure only a narrow slice of real-world effectiveness.