Why the AI will always be uncanny

by laughlyn (johan eddebo) | Feb 27, 2025

I was part of a small symposium the other day arranged by my friend and colleague Dennis Riches that focused on technology, neutrality and soft power. Great crowd where some genuine discussion took place among people who are actually trying to find useful answers to important questions, rather than these perfunctory events of dead-end socializing, brand marketing and mindless self-promotion that academic conferences today have mostly been reduced to.

So anyway, I was in a conversation with a clever private sector operative around the character of AI output, and I remarked that it’s curious that people seem to be able to identify a certain style and tone of AI-generated content, even now when it’s comparatively advanced. His response was then along the lines of “yes, so far” — but that we should naturally expect these identifying traits to become less and less pronounced as the LLMs develop further.

Sure, maybe that’s true, I thought, but then I recalled something another friend of mine said a couple of years back that stuck with me since then, both because I found it plausible, and since it seemed to contain an important ontological observation. I realized that this particular observation, if understood correctly, meant that these identifying traits of AI output are actually an inevitable aspect of it, that should rather get more pronounced as the systems develop further.

This guy has been working in and around various types of creative output his entire life (also feel free to support his band, Morlocks) and he has been close to the impact of the genAI phenomenon and considers many of its problems a continuation of those associated with canned digital output from the 90s and onward. As we spoke last fall, he made the point that in his experience, the uncanniness seems to become more marked the better the genAI output becomes, and that this was also the case with CGI output proper.

This might seem counterintuitive, but if we take a step back and think about what genAI output really is, and if we give it a proper ontological framing, it makes perfect sense.

GenAI output is based in a particular kind of adaptive probability calculations. In modern analytical language, it’s nature is that of a certain type of function, a particular type of rule-governed output that as a pattern can be distinguished from others.

In classical philosophy, it’s thus a certain kind of form, whether we approach it along the lines of Aristotelian or Platonic or Pythagorean or even Confucian theory.

And this form is just different from that of human intelligence. It’s a different abstract entity, another kind of pattern — and the unique character of the AI’s form is by virtue of the basic laws of cause and effect going to have a particular kind of impact on its output. It’s just like how Chopin’s unique character produced a discernible style of music that’s different from that of Hemingway’s prose. Generative AI that operates in accordance with a basic and distinctive pattern of rules will also produce output that cumulatively will lead to the emergence of a certain discernible style, just because its inherent nature must necessarily be expressed in the output.

And we should rather expect this to get more pronounced as these systems develop further and become more refined, because you simply elaborate upon the same basic network of causes. It’s like if you have an algorithm that incrementally and at random adds pixels in accordance with the pattern of the Mona Lisa. At first, the output looks like nothing at all, but soon, a vague likeness will manifest itself, impefectly expressing the basic causes — and the closer you get to the finished picture, the more clearly you can discern the character and nature of the basic algorithm.

It’s the same thing with processes of emergence in systems theory — the emergent phenomenon will only ever bring out more of the underlying causal factors, it can only more strongly and clearly express the characteristics and potential of the basic facts that engender the emergent phenomenon.

And by this same principle, the more advanced the AI gets, we should expect the uncanniness to become more marked.

However, this doesn’t mean that people will necessarily retain the ability to actually detect this uncanniness.

While it’s true that Mozart’s style is a discernible pattern present from his teenage compositions all the way up to his late work like Don Giovanni, this isn’t easy to discern for the average modern listener. It takes education, training and critical reflection for one to actually be able to make this out.

The same goes for the distinctive styles of genAI. While contemporary AI slop quite obviously reads like someone set fire to a pile of shit and pushed it down three flights of stairs, the more refined stuff is a bit less conspicuous.

And what’s worse — if people lose familiarity with the character of actual human output, they will lack a framework of comparison when interacting with more advanced genAI slop over the coming years. So even if it will always have a characteristic style, even if it will always have that uncanny character, if people become less and less accustomed to genuine human creative output as a baseline, the uncanniness will actually lose its meaning.

Give it a generation or two, and we’re not going to know why it would possibly signify a problem, and we will be desensitized to the associated cognitive discomfort.

The tune had been haunting London for weeks past. It was one of countless similar songs published for the benefit of the proles by a sub-section of the Music Department. The words of these songs were composed without any human intervention whatever on an instrument known as a versificator.

She knew the whole drivelling song by heart, it seemed. Her voice floated upward with the sweet summer air, very tuneful, charged with a sort of happy melancholy. One had the feeling that she would have been perfectly content, if the June evening had been endless and the supply of clothes inexhaustible, to remain there for a thousand years, pegging out diapers and singing rubbish.

Nineteen Eighty-Four