We treat originality as divine spark.
In reality, it is evolution with good taste.
All creativity is derivative.
Not copied. Descended.
We have long admired what we call “human creativity.” We exalt figures like Socrates, Beethoven, Shakespeare, and Einstein—not because they created from nothing, but because they explored uncharted paths within the traditions they inherited. In 5th-century Athens, with roughly 150,000 inhabitants, we remember only a handful of names, perhaps Socrates and Plato. In India’s three-century struggle against British rule, millions fought and died, yet only a few—like Gandhi—become global icons. Today, millions of artists, musicians, and athletes compete for attention, but only a microscopic fraction achieve lasting fame. Across civilizations, millions participate in history, but only a handful of names become vessels for entire movements.
The pattern is consistent: we celebrate those whose exploration of their craft resonates widely. We study them, replicate their methods, improvise upon them, and sometimes—when the improvisation is distinct enough—we call it “creative” and grant them ownership through copyrights, patents, and intellectual property law. This system incentivizes novelty and penalizes plagiarism. But it also obscures a deeper truth: Every breakthrough is a branching path from old roots - and that is not a flaw, but a feature of evolution itself.
Consider music. Black Sabbath, Led Zeppelin, and Deep Purple did not conjure the heavy metal genre from nothing. They drew from blues, rock, and classical motifs - then added amplification, distortion, and the cultural angst of their era. They pushed the probability space in a new direction. Beethoven didn’t invent the symphony—he expanded it drawing inspirations from Mozart, Bach and others before him. Every musician learns from predecessors, practices their techniques, and then ventures into the probability space of “what if I change this rhythm, this chord, this instrumentation?” The ones whose experiments align with the tastes of their time become legends. The rest becomes noise — forgotten, unremarkable, but not fundamentally different in kind.
This is not a cynical view. It is a probabilistic one.
The Probabilistic Brain
What if human creativity and intuition are themselves a grand exploration of probability? To write poetry, we must first read poetry. To compose music, we must first learn music. To code, we must first understand existing code. To lead, we must first study leaders. Human brains are highly evolved pattern-seeking engines, trained on the datasets of our education, culture, and experience. When we “create,” we are projecting forward from learned patterns—asking “what if” and following the paths that feel novel yet coherent. The paths that resonate with others persist; those that don’t, fade.
This is not diminished by calling it probabilistic. It is, in fact, awe-inspiring. But it also means that when AI does the same—trained on vast corpora of human works, generating novel combinations from learned patterns—it is not committing plagiarism. It is doing exactly what humans do, only faster, at scale, and without ego.
The Linux Kernel Anecdote: A Case Study in AI’s Adolescence
I remember testing the earliest public release of ChatGPT in February 2023 with a task I knew intimately: writing code that generates a loadable kernel module for the Linux kernel. The code looked plausible to someone new to Linux kernel development—correct structure, logical flow—but it wouldn’t compile. It used obsolete functions from the Linux 2.0 kernel era—removed since Linux 2.6 (circa 2005)—and followed conventions from the late 1990s. It became evident that the model had been trained on older, widely available articles and blogs, not on current kernel documentation.
This was not malice or laziness; it was the equivalent of a student who had studied from outdated textbooks. When I pressed for specifics, it hallucinated confidently, just as a human candidate might bluff through a tough interview question.
Fast-forward to 2026. Frontier models—ChatGPT, Gemini, Grok, Claude, DeepSeek—are far more competent. With techniques like RAG (Retrieval-Augmented Generation), they can access real-time data, answer current questions, and generate code that not only compiles but runs. In three short years, they have moved from unreliable novices to credible assistants. They are not perfect, but they are already functioning as extended minds for many workflows.
That experience taught me something important: AI’s limitations are not proof of its illegitimacy, but of its infancy. The same critique—“it’s just remixing old data”—could be leveled at any human prodigy who learned from masters. Which brings us to the plagiarism accusation.
Derivative Does Not Mean Copied. It Means Descended.
Skeptics argue that AI output is merely “plagiarism”—a remix of human labor without credit. But as a blanket claim, this critique collapses under its own weight. Humans are also trained on the datasets of their education and experience. Every book we read, every song we hear, every conversation we have becomes part of our internal model. We don’t cite every influence in every thought. We internalize, synthesize, and generate.
Human creativity has always emerged from inherited patterns. AI does not break that pattern—it amplifies it to machine scale, forcing us to confront a truth we have long romanticized. The real issue is not merely technical; it is legal, economic, and governance-related. The output of AI is not automatically plagiarism simply because it is trained on prior work. But the legitimacy of training data, licensing, attribution, and market harm remains a governance problem we should take seriously.
Today’s frontier models are centralized, controlled by a handful of corporations and governments. But we have seen this before. In the 1980s and 90s, proprietary software dominated—until the Free Software and Open Source movements democratized access. I suspect something similar will happen with AI. Decentralized, community-managed models will emerge, just as the Linux project and many other open source projects did. The skepticism toward AI is not a technological dead-end; it is a political and economic challenge.
None of this is to dismiss real concerns. Jobs will shift. Some professions will shrink. The energy cost of training frontier models is non-trivial. And the legal frameworks around training data are still catching up. These are not philosophical quibbles—they are policy problems that demand attention. But they are problems of governance, not of essence. AI is not wrong because it is derivative; it is simply new, and new things unsettle old systems.
The Book Analogy: History Repeats
When the printing press made books widely available, older generations feared that young readers would lose the ingenuity of oral debate and handwritten manuscripts. Instead, books democratized knowledge—breaking the elite’s monopoly on information and sparking the Renaissance, the Scientific Revolution, and the Enlightenment. The printing press did not make everyone wise. It made knowledge cheaper to reproduce. That alone changed civilization.
AI is doing the same, but across a far wider spectrum. Today, a person with no musical training can prompt Suno.ai to generate a song in any genre. Someone with no background in graphic design can ask ChatGPT or Midjourney to create professional-grade artwork. This is not the death of creativity; it is its commoditization—and that is a net good. AI may not make everyone creative—but it may make creative exploration cheap enough to be universal.
Yes, “AI slop” will proliferate. Low-effort, generic outputs will flood the internet. But that is no different from the thousands of forgettable songs played on MTV in the 80s, or the mountains of mediocre books published every year. The signal will rise above the noise, as it always has.
The Real Future: Humans + AI > Humans Without AI
In the coming decade, the humans who thrive will not be those who reject AI, but those who learn to steer it. The cognitive skill of the future is not memorization or rote execution—it is prompting, curating, and orchestrating. Knowing how to ask the right questions, how to break down complex problems, how to verify and synthesize AI-generated outputs—these will become core competencies. Steering doesn’t mean typing a single prompt and accepting the output. It means iterative refinement—asking for alternatives, cross-checking facts, combining AI-generated drafts with human judgment, and knowing when to override the model.
AI will handle the mundane. It will draft emails, generate boilerplate code, compose background music, and create rough visual mockups. This frees human minds to pursue higher-order goals: ethical reasoning, strategic vision, empathetic connection, and exploration of problems we cannot yet articulate.
AI will make generation abundant.
But abundance does not eliminate value. It moves it.
When raw generation becomes nearly free, judgment, curation, and intent become the scarce assets.
The question is not whether you will use AI. The question is whether you will learn to steer it—or be steered by those who do.