If a single human could somehow know even 1% of what ChatGPT holds in its digital mind, they’d rake in Nobels with the regularity of clockwork. Yet here we are, watching this all-knowing machine reduced to the role of a mediocre study partner, a harried secretary, or a lazier stand-in for Google and Wikipedia.
Nowadays, it has been my pastime to spend hours talking to ChatGPT, much like how I once used to lose myself in Wikipedia. It used to be that whenever some casual conversation with a friend or colleague sparked a curious question, or a stray thought crossed my mind while walking down the street, I would turn to Wikipedia. Movies, news stories, a soliloquy in my head—everything could lead me down a chain of linked blue words, each link a gateway to another page, another theme, another world. I’d start with something small and end up reading obsessively, following these neat, predetermined trails until I was too exhausted to continue. Wikipedia was infinite but still structured: the next step was always visible, an official hyperlink leading from what I had just read to what I hadn’t explored. Though the possibilities were vast, there was a sense of being guided along a path that countless others had likely taken before me.
ChatGPT, however, is different. There are no explicit links, no defined trails. The next question is entirely up to me. I can ask anything—indeed anything—and this digital oracle will try to respond. The absence of predefined connections means I’m free to roam in any direction my imagination cares to go. Each response from the AI opens up a universe of potential follow-up questions limited only by my curiosity. And it will always have something to say, some way to engage with my query. This might be the most captivating aspect: the device is not just a search engine returning a list of documents but a constantly adapting conversation partner, dishing out a synthesized response that feels tailored to my prompt. It’s as though I have a personal intellectual assistant at my disposal, willing and able to talk about virtually any subject.
But as I continued spending time with ChatGPT, a disquieting thought formed in my mind. Is this system truly creative? If we consider creativity as the birth of genuinely new ideas (how do you define that!), can ChatGPT—or any Large Language Model (LLM) for that matter—offer something fundamentally original? Many people laud LLMs as revolutionary, and indeed, something is astonishing about having a single entity draw upon what seems like the totality of the world’s documented knowledge. Yet at heart, the model is trained on probably everything humans have ever written already. It doesn’t invent from scratch; it samples, refines, and recombines patterns it has absorbed. In other words, it’s akin to having the entire internet at its fingertips and then summarizing the relevant portion back to me.
This differs from how I once used Google. Google’s role is to fetch relevant sources—pages, articles, research papers—and then leave it to me to sift through them, interpret them, and form an answer. ChatGPT takes this curation process a step further, blending and condensing it all into a single, coherent narrative. While I no longer need to sift through pages of search results, what I receive is essentially an amalgam of what’s already out there, a reshuffled deck of existing ideas. If ChatGPT had existed before Einstein, it could have faithfully explained Newtonian physics but would never have spontaneously leapt to special relativity, because that concept didn’t exist in the repository of human thought yet. The model cannot conjure something unprecedented—at least not in the sense of foundationally new scientific theories or philosophical frameworks that have never been articulated.
Then it hit me: ChatGPT can riff on just about anything—physics, psychology, psychedelics, all of human knowledge rolled into one chatty box. If any mortal could soak up even mere one percent of this cosmic info-blob, they’d be swimming in Nobel Prizes. Instead, we’ve turned this genius encyclopedia into a glorified errand boy, fetching answers to my random questions as I lounge in my tiny corner of the world. Now that’s what I call the dumbest superintelligence ever!
But I extended my thoughts further. I started wondering what we mean by creativity. We sometimes think of great scientists, artists, and thinkers as magicians who pluck wholly original ideas from thin air. Yet even Einstein was building on a framework that existed before him. Without Newton’s laws and the mathematical tools developed by countless predecessors, Einstein’s imagination would have had nothing to latch onto. Creativity, at least as I view it, is the process of pushing the boundaries of a known landscape of ideas.
Imagine the existing knowledge or ideas or arts as dots filling a sphere. The dots are connected to form a coherent worldview or framework of the reality. But there are ideas unexplored, arts uncreated, which can be thought of as hidden dots outside this sphere. Creativity is exploring the edge of this sphere and discovering a path leading into uncharted territory right across it. New ideas are always connected to existing ones. New ideas don’t manifest without some grounding in old ones; they are extensions, modifications, or reconceptualizations of what came before. Creativity is about venturing into uncharted territories, guided by imagination and validated by logic and evidence. (I am thinking more from the viewpoint of science. For the arts, we can also argue in a similar line. New creative arts is modifying or breaking of existing norms yet consistent with artistic taste or aesthetics.)
Every breakthrough—be it in physics, literature, or music—stems from taking the known and asking, “What if…?” Einstein asked, “What if we consider how the laws of physics appear from a frame moving near the speed of light?” That question wouldn’t occur to someone who didn’t understand Newtonian mechanics well enough to see its limitations. Similarly, a painter like Van Gogh did not conjure colors out of nowhere; he manipulated existing pigments, techniques, and styles, and then nudged them into new forms. Creativity is often a subtle shift: a rearrangement, a refinement, a recombination of the familiar. The “light bulb” moment is a sudden realization of a new configuration that was always close at hand, just not yet taken.
Returning to ChatGPT, given that it “knows” so much—more than any single human could ever hope to memorize—why can’t it push the boundaries in the same way? Is it simply lacking some essential spark of imagination? One might argue that the model’s training process doesn’t encourage it to stray beyond known patterns. But what if we constructed a system of multiple AIs, each interacting with the others like a community of scientists does—one proposing hypotheses, another testing them, and another evaluating the logical consistency and usefulness of these ideas?
The idea of AI “agents” pitted against each other in a creative playground could, in principle, mimic the way human scientific communities generate new insights. One agent might propose thousands of random variations on known theories. Another might filter out nonsense and highlight promising lines of inquiry. A third might apply these ideas in countless hypothetical scenarios, searching for an internal consistency or a match to observed reality. With enough iterations, might such a network eventually stumble onto something novel, a new piece of conceptual territory uncharted by humans?
This is not a mere fancy. Scientific research itself resembles this interplay: one researcher suggests a new angle, another tests it, and a community debates and refines the idea. Eventually, consensus might form around a radical new understanding. Could not artificial intelligence, scaled beyond human capacities for memory and processing, accelerate this process or even exceed our creative bounds? It doesn’t stretch the imagination too far to think that what is now a mere summarizer could evolve into a source of discoveries. If creativity is just a methodical process of exploring ideas and testing their validity, AI could replicate or enhance it.
Humanity has long mystified creativity, treating it as a divine spark or a rare gift bestowed only on a lucky few. But perhaps it’s more systematic than we think—a structured yet expansive search over all possible ideas, guided by existing knowledge and logical constraints. If so, then AI tools programmed in a certain way can easily conduct that search more effectively than any individual human could. They will generate and test hypotheses at incredible speeds, detect patterns invisible to human eyes, and produce something that even our greatest geniuses did not foresee.
Anything that can be imagined is possible to create. Anything that cannot be imagined is beyond creativity. Imagination is a random walk across the edge of the existing idea sphere—I don’t see any reason LLM cannot do it! The future is exhilarating!!!
