The Difference Between Random and New

Spend enough hours generating things with AI and you start to feel a texture to the output, a smoothness that is hard to name at first. The copy is competent, the image is plausible, the melody resolves exactly where you expected it to resolve, and nothing is wrong, which is the unsettling part, because after a while the work starts to feel like it was drawn from the middle of something, the centre of a vast average, and you begin to wonder whether we have found the ceiling of what a machine can make feel new.

I do not think the ceiling is a matter of model size. I think it is built into the mechanism, and the mechanism is worth looking at plainly, because once you see it you also see where the exit might be, and the exit runs through physics most people file under science fiction.

The mechanism is a closed one

A language model produces text by sampling from a probability distribution it learned from human writing. At every step it estimates how likely each next token is, given everything so far, and it picks from that distribution. Image and music models do a version of the same thing in their own spaces. This is an extraordinary capability, and it is also a closed one, because everything the model can produce is a recombination of patterns already present in what it was trained on. Its novelty is interpolation, the filling of gaps between existing points. Turn the temperature up and you sample further from the centre, which feels wilder, until you notice that wild and incoherent arrive together, because the edges of a learned distribution are mostly noise. The ceiling is not a lack of imagination. It is that the possibility space itself is bounded by the data, and the average of everything we have made sits right in the middle of it.

Most randomness is not random

To see why this matters, it helps to be precise about a word we use loosely. Most of the randomness in computing is not random at all. A pseudo-random number generator runs a deterministic algorithm from a starting seed, and the sequence looks unpredictable only because the algorithm is complicated and the seed is hidden. Anyone who knows the seed and the algorithm can reproduce every number exactly. The unpredictability is a property of your ignorance, not of the world. A classical computer cannot do better than this, because a deterministic machine following fixed rules has no source of true surprise inside it. It can disguise its determinism beautifully, and it can never escape it.

Quantum is where the floor stops being deterministic

When you measure a quantum system that sits in a superposition of states, the outcome is not merely hard to predict, it is unpredictable as a matter of physics, with no hidden variable underneath that would let a better-informed observer call it in advance. The randomness belongs to the event itself. This is the part that is consequential and easy to miss. In June of 2025, NIST and the University of Colorado Boulder switched on a randomness beacon that draws its numbers from measurements of entangled photons and certifies, through the structure of the measurement itself, that the bits could not have been known beforehand by anyone. They built a tap connected directly to the indeterminacy of the universe. The numbers that come out are not sampled from any dataset, not interpolated from anything that already existed, not recombined from the past. They are new in a sense no algorithm can match, because their source is an event that had no determined answer until it happened.

Randomness on its own is not creativity

It would be too easy to leap from there to the claim that quantum makes machines creative, and the leap is wrong, because noise is the most random thing there is, and noise is not art. Creativity is constrained selection, a movement through a space of possibility guided by taste, intention, and the pressure of a problem. The randomness only matters as the raw material of variation, the thing that proposes something the selector did not already contain. A generative model already has a very good selector. What it lacks is a source of proposals that come from outside the distribution of the already-made, because its variation is reshuffled memory, and reshuffled memory regresses to the mean of what was remembered.

Creativity may depend on a genuine outside

This is where the relationship between randomness and creativity gets interesting, because human creativity may quietly depend on a genuine outside. The history of invention is full of moments where the new thing did not interpolate from the old, the mutation that natural selection works on is genuinely undirected, and there is a serious if unsettled argument that the brain itself exploits noise at the smallest scales. I am not claiming the mind is a quantum computer, which is a stronger and shakier claim than I want to make. I am pointing at something simpler, that creation seems to need a supply of the genuinely undetermined, a way to leave the space of what is already implied, and that a deterministic machine sampling its own training data has been cut off from exactly that supply.

Originality has become an engineering question

So the honest question is whether feeding true quantum entropy into a generative system would do anything, or whether it would only produce better-disguised interpolation. Seed the sampling with certified quantum randomness and the immediate effect is modest, because you are still selecting from the same learned distribution, just arriving at different points within it. The more interesting possibility is architectural, using genuine indeterminacy as a generative source the model is asked to make sense of, the way a person makes sense of an accident in the studio. Whether that crosses the line from novel-to-us into truly original is not something I can settle from a keyboard, and that is the point worth sitting with. We can now build a source of the genuinely undetermined and wire it into the machines that make our images and our sentences, and we are about to find out whether the thing that has always felt most human about creation was the taste, or the access to real surprise that the taste was selecting from.