The Spectrums of Originality: A Problematic Proposal for AI Copyright

View on LinkedIn

The Spectrums of Originality: A Problematic Proposal for AI Copyright

When the regulation of artificial intelligence is debated, a recurring proposal is to create a spectrum that classifies works according to the degree of human versus AI participation in their creation. At first glance, this approach seems logical and orderly, but a deeper analysis reveals potential problems that could generate more confusion than legal certainty.

Originality Cannot Be Measured

There is a fundamental problem that must be addressed before all others: originality is not objectively quantifiable. U.S. copyright law, since Feist Publications, Inc. v. Rural Telephone Service Co. (1991), established that a mere "modicum of creativity" is sufficient to obtain protection. As the Supreme Court stated in that decision:

"The vast majority of works make the grade quite easily, as they possess some creative spark, no matter how crude, humble or obvious it might be."

This 1991 case reveals a reality: if we cannot objectively measure how much human creativity exists in a purely human work (redundant as that may sound), how do we expect to precisely calibrate the proportion of generative AI output versus human creation? The thresholds assume a capacity for measurement that simply does not exist.

This impossibility of gradation exposes the fundamental contradiction of the entire proposal: building precise categories around something that is inherently immeasurable.

An Apparently Sensible Proposal

The typical proposal establishes five levels of creation:

1. Fully human (0% AI): The person creates everything without generative AI assistance. (100% registrable)

2. Human with AI assistance (AI as a tool): "Fix the spelling of this text I already drafted, or harmoniously redraft this paragraph." The AI functions like Photoshop or any other editing software or even AI (ChatGPT).

3. Co-creation (balanced participation): "Give me 3 ideas for structuring an essay on AI and law; I'll choose and edit afterward." Both the human and the AI contribute significant elements.

4. AI with minimal human guidance: "Write an 800-word article on AI regulation in Mexico." The human provides only the initial spark.

5. Fully AI [AKA Made by the AI] (no human intervention): The machine generates everything autonomously. Not registrable for copyright.

This scheme has undeniable appeal: it appears to offer clarity and classifications that could potentially translate into different levels of legal protection. However, the reality is considerably more complex.

Why It Does Not Work in Practice

1. The Determination Dilemma, or "Where Do We Draw the Line"

Who decides which category a work falls into? This apparently simple question triggers a cascade of practical complications. Will it be a technical evaluation based on logs and metadata? A sworn declaration by the author? A judicial expert opinion on a case-by-case basis?

Each option presents problems. Technical records can be manipulated or may simply fail to capture the real complexity of the creative process. Author declarations are subjective and potentially strategic. Expert opinions in disputed cases would be costly and slow, creating prolonged uncertainty. What do we do with Pixar, or with software developers — would we have to create exceptions or variants by industry?

2. The Error of Artificial Divisions

Creation with AI does not occur in sealed compartments: it is a continuous back-and-forth between the human and the LLM. Consider a typical process: a writer uses AI to generate an initial draft (level 4), then manually rewrites 70% of the content (level 2?), then asks the AI to improve certain paragraphs (level 3?), and finally makes final edits (level 2 again?).

At what point on the spectrum do we place this work? The classification becomes arbitrary, depending on which moment of the process we consider "definitive" or what percentage of changes we deem "substantial."

We already had this debate with photography. No one considers that automatic recordings from security cameras constitute authorship: they are merely instrumental. Likewise, the key distinction lies not in which tool is used, but in the creative intention and the human control exercised over the result.

4. Disproportionate Commercial Consequences

If rights vary according to classification, the commercial implications would be enormous. A work classified at level 4 could have lesser protection than one at level 2, regardless of its quality, utility, or actual market value.

This could create perverse incentives: artificially inefficient creative processes designed to achieve a higher-protection classification, or strategic disclosure (or concealment) of AI use depending on what is most convenient.

Technological Neutrality and a More Practical Approach

Rather than creating complex taxonomies that measure degrees of participation, I propose a binary approach based on the principle of technological neutrality:

Is there human creativity? Yes = protectable work. No = no rights.

Technological neutrality establishes that the law must be agnostic with respect to the tools used. It does not matter whether you use a pencil, a computer, Photoshop, or AI: legal protection should not depend on the sophistication of your instruments, but on the human creativity present in the result.

This binary criterion maintains the standard established in Feist: if that "modicum of creativity" is present, the work deserves protection, regardless of what tools were used to develop it. If there is no human creative intervention whatsoever, there are no rights (as in the case of Naruto [the monkey selfie]).

This approach eliminates the problems of graduated classification and focuses on what truly matters: the presence or absence of human creativity, not its exact proportion or the technological purity of the process.