When a Few Pixels Redrawn Change Everything: The Battle Over Human Authorship in the Age of AI

saurabhsarkar
Oct 18
7 min read

Updated: Nov 3

Understanding Intellectual Property in the Modern Age

At its core, intellectual property (IP) is not merely about ownership of ideas. It revolves around incentive design.

Societies discovered long ago that creativity and invention flourish when individuals can reap benefits from their mental labor. However, pure ideas are non-rivalrous: once shared, they can be copied infinitely. Therefore, IP law carves out temporary monopolies on specific expressions or applications of ideas. The logic is utilitarian: reward creativity just enough to encourage it, but not so much that it stifles others’ creativity.

That’s the tightrope IP walks. Every IP regime—copyright, patent, trademark, trade secret—reflects this incentive structure:

Copyright encourages artistic and literary creation.
Patents reward technological innovation.
Trademarks protect signals of trust and origin.
Trade secrets safeguard investments in confidential knowledge.

In all cases, IP assumes two critical things:

Scarcity of creation: human creativity is limited and valuable.
Identifiable creator: someone made conscious choices that justify reward.

The Impact of AI on Intellectual Property

What happens to intellectual property when AI creates?

AI disrupts both assumptions. Generative models can produce thousands of outputs per minute, erasing scarcity. The creator becomes diffuse: is it the algorithm, its developer, the prompt writer, or the original authors of the training data? This isn’t just a technical issue; it’s a philosophical one. Can there be authorship without intent?

IP systems were built around the idea that creation reflects willful human expression. A poem, a design, or a melody reveals human judgment. A machine lacks such intent; it merely optimizes functions.

So, if IP aims to incentivize human creativity, granting rights to machine outputs seems illogical. You can’t incentivize an algorithm. However, humans use AI as an instrument. The key question becomes: when does AI remain a tool of human creativity, and when does it cross the line into being the creator itself?

The Continuum of Human Involvement

Instead of a binary “human vs. AI” divide, it helps to envision a spectrum:

| Human Involvement | Example | IP Logic |

|-----------------------|------------------------------------------------|-------------------------------------------------|

| Fully human | Painting from scratch | Full copyright: clear expression of human intent |

| AI-assisted | Human uses AI for drafts, then edits, curates, and arranges | Partial protection: human expressive control present |

| AI-directed | Human gives a detailed prompt, accepts raw output | Weak or no protection: limited creative decision-making |

| Fully autonomous | AI generates output without human input | No protection: no human authorship or incentive target |

The principle is simple: IP protection follows human creativity, not machine computation. That’s why most jurisdictions deny copyright or inventorship to purely AI-generated works but allow protection for AI-assisted works where human authorship is evident.

The Mirror Problem: Whose Shoulders Is AI Standing On?

If AI outputs lack ownership, another fundamental question arises: can a work with no owner still infringe on others’ rights? Yes, because creation doesn’t occur in a vacuum.

Generative AI models are trained on vast corpora of prior human works. From a first-principles standpoint, training is a kind of large-scale learning, similar to how humans learn from what they read or see. The legal challenge is that machines copy to learn; humans abstract.

If IP law’s intent is to preserve incentives for human creators, two tensions emerge:

Over-protection (restricting AI training) risks freezing progress.
Under-protection (allowing unrestricted use of copyrighted material) risks disincentivizing future human creation.

Neither extreme satisfies IP’s foundational goal: sustainable creativity. The logical answer lies in structured access: systems where data can be used for learning, but rights holders retain control over direct exploitation (e.g., licensed datasets, opt-out mechanisms, or collective remuneration).

Rethinking “Ownership” in the Age of Abundance

From first principles, ownership exists to manage scarcity. But AI collapses scarcity: digital abundance becomes the default. When content can be produced infinitely, the locus of value shifts:

From ownership to authenticity (who made it?)
From creation to curation (who selected it?)
From control to trust (can I rely on it?)

This suggests that IP protection may evolve from exclusivity rights toward reputation and provenance systems, digital signatures, provenance chains, and authenticity marks that distinguish “human-made” or “trusted-source” works.

Rather than expanding copyright, future frameworks may prioritize:

Attribution transparency
Traceable AI model inputs
Verified human authorship labels

These mechanisms preserve trust—the social contract IP law originally stood for—without trying to enforce scarcity where none exists.

Patents, Inventions, and the Meaning of “Inventor”

Patents reward problem-solving ingenuity. The first principle here is that inventions must arise from non-obvious insight.

AI challenges this by generating solutions that may be novel but not conceptually understood by humans. If a model designs a new material, who had the inventive insight: the engineer, the algorithm, or the dataset?

To remain faithful to first principles, patent law should protect human problem formulation and validation, not brute-force generation. Otherwise, patents would become property rights over computation, not invention, undermining the incentive system entirely. Thus, the emerging global consensus—that only natural persons can be inventors—is not a bureaucratic quirk but a first-principles safeguard: only minds capable of intent deserve monopoly rewards.

The Code Parallel: AI as Co-Author, Not Creator

The copyright debate over AI-generated art mirrors a quieter but equally significant one in software. Just as the Zarya of the Dawn case questioned whether minor human edits (like adjusting a character’s lips) amount to authorship, developers now face a parallel question: does fixing a few bugs or adding structure to AI-generated code make it “yours”?

Let’s unpack this from first principles.

Code as Expression vs. Function

Copyright protects expression, not function. This means two developers can independently write similar logic—one with Copilot, another from scratch—without infringement unless specific lines are substantially identical. But the human-authorship principle still applies:

Code written from scratch reflects original creative judgment: architectural choices, abstractions, naming, flow.
Copilot-generated or LLM-suggested code is machine output based on learned statistical patterns. If accepted verbatim, it’s not authored in the legal sense; the developer merely selected it.

Thus, like Zarya of the Dawn, passive selection or light editing isn’t enough. Authorship lies in intentional structure, not incidental correction.

The Spectrum of Involvement in Coding

| Scenario | Description | IP Implication |

|-----------------------------|-------------------------------------------------------|----------------------------------------------|

| Human-written | Developer designs architecture, writes all code manually. | Full copyright protection. |

| AI-assisted debugging | AI suggests minor fixes or refactors; human integrates them knowingly. | Human remains author; edits are functional aids. |

| Copilot-generated scaffolding | Developer accepts generated boilerplate or snippets without substantial modification. | Ambiguous authorship; likely unprotectable. |

| Fully automated generation | Prompt-only input; AI outputs full program. | No copyright in the code itself; falls into public domain in most jurisdictions. |

In short: creative reasoning defines ownership. Using Copilot to accelerate coding does not equal co-authorship. The copyrightable part is the architectural design, integration logic, and novel structure a human imposes.

The Hidden Risk: Derivative Code

If the LLM reproduces copyrighted code fragments from training data—an issue Copilot faces in ongoing cases (e.g., Doe v. GitHub, Microsoft, and OpenAI, N.D. Cal., 2023)—developers may unknowingly deploy infringing material. This risk resembles an artist unknowingly publishing an AI image that mirrors a copyrighted photograph. Thus, due diligence and traceability become the new form of authorship discipline.

From First Principles

Returning to first principles:

IP exists to incentivize human innovation, not algorithmic replication.
The value lies in conceptual originality, not mechanical production.
Debugging AI code with intent, just like editing AI images meaningfully, crosses the threshold from passive use to authorship. Hence, whether it’s a brushstroke on digital lips or a function refactor in Python, what matters isn’t the medium; it’s the presence of human judgment.

A Coherent Framework from First Principles

Let’s rebuild a compact logical model for IP in the AI era:

Purpose: Encourage human creativity and knowledge expansion.
Subject: Protect expressive or inventive decisions, not mechanical outputs.
Ownership: Belongs to humans who contribute meaningful, original judgment.
Training Data Use: Should balance public learning and private incentive; copying for learning may be permitted, but copying for substitution is not.
Liability: Falls on humans or entities that deploy or benefit from AI outputs, not the AI itself.
Evolution: As AI lowers creation cost, protection will shift from ownership to provenance and trust systems.

The Road Ahead

IP law has always evolved with technology, from printing presses to photography, radio, and the internet. Each time, the principle remains constant: preserve incentives for meaningful human contribution.

AI doesn’t break IP logic; it merely stretches it. The future legal frameworks will likely look like this:

AI as Tool: Human creativity + machine efficiency = protectable.
AI as Creator: Machine autonomy = public domain.
AI as Learner: Training on data = regulated, licensed, or fair-use-bounded.
Human as Steward: The real value shifts from ownership to stewardship: ensuring transparency, fairness, and attribution in an age of synthetic abundance.

Final Reflection

From first principles, the question of IP and AI is not who owns machine output but how society preserves human purpose in creation.

If we treat AI as a new kind of creative amplifier—not a replacement for authorship but an extension of it—then IP law’s mission remains intact: to sustain a culture where humans still choose, design, and dream, even as machines compose the rest.