top of page
Search

Beyond AI Watermarking: Should We Measure the “DNA” of the Prompt?

Aug 17
2 min read

By: Faisal Awartani (Ph.D.) , email: faisal@insights.ps


Anthropic's decision to watermark AI-generated content is an important step toward transparency. But I wonder whether simply labeling something “AI-generated” tells us enough about its intellectual origin.


Perhaps we should think about generative AI through a biological analogy:

Prompt = Genotype (DNA)AI model + context = Developmental environmentGenerated output = Phenotype


A sophisticated prompt is not merely an instruction. It can contain original ideas, hypotheses, data, evidence, domain expertise, methodological choices, constraints, and years of accumulated professional knowledge.


Consider a psychiatrist who provides a detailed prompt containing clinical knowledge, research evidence, an original hypothesis, and a proposed interpretation, then asks AI to synthesize it into an article. Compare that with:


"Write an article about depression treatment."


Both outputs may come from exactly the same AI model and therefore carry the same watermark—but their intellectual DNA is fundamentally different.


Perhaps, therefore, we need something beyond an AI watermark: an Intellectual Contribution Signature, Prompt DNA Score, or Human–AI Contribution Index.


Such a measure could potentially examine the originality and domain specificity of the prompt, evidence and data supplied by the human, hypotheses and analytical structure introduced by the user, and how much substantive generation was subsequently contributed by the AI.


Instead of simply:

“AI Generated”


we might eventually see:

Prompt DNA: High human intellectual specificationDomain expertise: HighOriginal conceptual input: HighAI generative expansion: ModerateProvenance: Claude-assisted

The biological analogy is imperfect—but interestingly, that may strengthen it. DNA does not independently determine phenotype. Expression depends on environment, development, interactions, and stochastic processes. Likewise, the same prompt can produce different outputs across different AI models.


Conceptually:

Output = f(Prompt, Model, Context, Randomness)

This raises what I believe is an important research question for the AI community:

How much of the semantic and intellectual structure of an AI-generated output can be traced back to the human prompt, and how much was contributed by the model?


Perhaps future AI transparency should therefore tell us not only which machine generated the words, but also something about where the intellectual DNA originated.


The prompt provides the genotype. The AI provides the developmental environment. The output is the phenotype.


Maybe the future of responsible AI needs both a watermark and an intellectual DNA signature.


 
 
 

Comments


Insights for Research, Polling and Training
 

Based in the Ramallah - Delivering research across West Bank, Gaza, East Jerusalem, Jordan & Lebanon

  • Youtube
  • LinkedIn

© 2008–2026 All Rights Reserved.

Send us a message
 and we’ll get back to you shortly.

bottom of page