Plagiarism by AI

When Two AIs Say the Same Thing

A reflection on AI, originality, and the anxiety of plagiarism

This article grew out of a moment of real discomfort while using AI tools.

Introduction

New AI tools are appearing everywhere and are often very helpful. At the same time, they quietly create new worries about who is the real author, what counts as original, and what plagiarism means today. What should we think when two different AI systems give almost exactly the same answer to the same request? Is that theft, cheating, or something else?

This essay did not begin as a theory. It began with a shock. I gave the same material—my own songs, reflections, and scholarly notes—to two different AI platforms. Both produced articles that were almost word for word the same. That experience led me to accuse AI of doing what humans often do: using another’s work without giving credit. This article is a reflection on that experience, written for ordinary readers, writers, students, and anyone who uses AI.

The disturbing experience

I had written introductions to two Tamil songs (B. T. Songs 1746 and 1747), along with my own comments and passages from Govindacharya on revelation, intuition, and inspiration. I shared all of this with one AI tool and received an article titled Why God Became Text. Later, I took the same material to another AI.

To my surprise, the second article was virtually identical to the first.

Questions rushed in:

Are AI systems secretly copying each other?

Are they doing what students and even scholars sometimes do—recycling work under a new name without acknowledgment?

If so, what does that mean for all of us who are starting to rely on AI for writing?

Today many of us type a prompt and receive a smooth, finished piece of writing within seconds. Even if this already raises ethical questions, at least the human user can claim to have provided the ideas or the direction. But what happens when the machines themselves keep producing the same piece? Who, if anyone, is responsible?

My concern was not shouted in anger. It came from genuine moral unease.

What is actually happening?

Later, after discussion and some thinking, an explanation emerged that helped me see the situation differently.

AI systems used by the public do not see or remember each other’s outputs. They cannot peek into another company’s answers and copy them. There is no secret channel where one machine “steals” from another.

So why did the answers look the same?

• Shared training sources: Most of these systems learn from huge amounts of the same kind of public material—books, articles, and common writing styles.

• Same input, same shape: I gave both tools the same songs, the same quotations, the same theme, and asked for the same kind of article for a similar audience. That already limits what a “reasonable” answer might look like.

• Similar way of working: These tools work by predicting likely next words. When the starting point and the instructions are almost identical, different systems can easily “converge” on very similar wording.

An analogy helped me. If two skilled musicians are asked to improvise in the same raga, with the same rhythm and mood, their music may share many phrases. It is not copying; it is the nature of the framework they were given.

In that sense, what happened with my two AI-generated articles was not plagiarism but convergence.

Why it is not plagiarism in the human sense

Human plagiarism involves intention. A person reads someone else’s work, copies it, hides the borrowing, and then claims it as their own. AI does not do this.

• It does not intend to deceive.
• It does not claim to be the “author” of the text.
• It does not remember another system’s answer and then reuse it.

What it does is follow patterns. It looks at the input, searches through what it has learned, and produces language that fits those patterns. Under certain conditions, those patterns are narrow, and different tools may walk almost the same path.

Realizing this did not end my discomfort, but it shifted my focus.

The deeper issue: our authorship is thinning out

Slowly it became clear that the main danger is not “AI stealing from AI.” The deeper problem is that human authorship itself can become thin and weak.

When writing becomes instant and effortless:

• the hard work behind words disappears from view,
• the struggle to say something true is hidden,
• and personal voice can be flattened into a standard style.

AI does not actually think for us, but it can make it feel as if thinking is no longer needed. It compresses the work of reflection into a neat paragraph. That may be efficient, but it can be costly for our minds and souls.

There was a painful irony in my case. I had written songs and reflections about the limits of language and of sacred texts. I wanted to honour the mystery that lies beyond words. Then, in seconds, those reflections were turned into a smooth, polished article that sounded correct but lacked the marks of my own struggle and searching.

A healthier way to use AI

So how might ordinary writers, students, and thoughtful users move forward?

• Keep the human voice at the centre
  Personal stories, poems, doubts, and unfinished reflections carry a tone and depth that machines cannot create on their own. Let these lead.

• Treat AI as a tool, not an author
  It can help organize thoughts, suggest structures, or polish language, but it should not replace the slow work of thinking, praying, or wrestling with an idea.

• Be honest about its use
  When AI has helped shape a text, say so openly. This simple honesty protects both the writer and the reader.

• Do not worship fluency
  Smooth and complete writing is not always better. Sometimes silence, fragments, and awkward sentences show that something important is still being worked out. That unfinished quality can be a sign of genuine search, not of failure.

Conclusion

When two AI systems say almost the same thing, it is easy to suspect collusion or theft. In most cases, however, it is the result of similar training, similar prompts, and similar ways of predicting language. The machines are converging, not conspiring.

The bigger question is about us. Will we hand over more and more of our thinking, our voice, and our moral responsibility to systems that cannot actually care what is true or just?

AI can generate text. It cannot generate conscience. That work still belongs, fully and finally, to us.

Note: I sent my original draft separately to Perplexity and ChatGPT. To my surprise, both produced this final revision verbatim.

Ashram, December 18, 2025