Forum: Story Discussion and Feedback

Automation is a marvelous concept for the commercial purpose of reproduction of products.
It cannot however create originals.

That seems to me to be an Intelligent Design versus Evolution argument.

AJ

The interesting thing here is that it becomes an argument for the AI actually being creative. It's not the only one, either. There was a case recently where a someone used an AI to generate a list of recommended books for summer reading. Some of them turned out not to exist. The consensus, in general, was that the recommended (but non-existent) books sounded very interesting and someone ought to write them.

Creative? Or not? In my opinion, the answer quickly winds up in the land of philosophy, because you need a decent definition of what 'creativity' means in order to answer it.

Again, we're just dancing around the problem of defining what 'creative' means.

In your specific example, I would argue that it is not, based on my understanding of 'creative.' But that's because it's an exaggerated example, not because it's a meaningful point.

Consider the oft-stated comment 'If one gives an infinite number of monkeys an infinite number of typewriters, one of them will write Shakespeare.' Is that monkey 'creative?' Again, I would argue it is not, because the creativity is inherent in the probability of infinity.

On the other hand, suppose we have a black box which reliably writes new, never-before-written Shakespeare-grade material on request. Is that black box 'creative?' Does it matter whether the black box (of arbitrary size) contains a human being, an animal, a disembodied brain, or a computer? If so, why?

You say 'random mashups of things humans have created.' I can argue (likely incorrectly, but still) that every work of fiction which does not coin new words (and, perhaps, even ones that do) is a 'random mashup of things humans have created.' Humans created the words; authors mash them up. Joseph Campbell makes a moderately persuasive case that there are only a few stories, retold in all sorts of interesting permutations but still, fundamentally, the same story.

Part of the problem is 'random.' Modern LLMs are about as 'random' as many people are when stringing sentences together, and potentially less so. They're not monkeys banging on typewriters.

One step back: is 'West Side Story' creative, or a mashup of Shakespeare? What about 'R+J', which uses Shakespeare's dialogue but is visually extremely different? Creative, or not?

Back to my black box. If one postulates a future AI that reliably emits Shakespeare-grade stories, how exactly does one argue that it is not 'creative?' Oh, you can ascribe the creativity to the programmers of the AI, but none of them can write Shakespeare-grade stories, so where did the 'creativity' come from? Or you can ascribe it to the source material being combined, but that's like saying Degas wasn't creative because he learned everything from studying the paintings of other painters. And, of course, it's also like saying that humans aren't creative. The creativity comes from either God or evolution, whichever 'programmed' humans.

Next, a big step back. There's an interesting intersection of physics and philosophy that holds that the universe is deterministic and predestined (from the instant of 'creation' - presumably the Big Bang), and that free will is an illusion. If so, Shakespeare himself is no more than a biological machine who wrote exactly what he was 'designed' by nature to write, no more and no less. His creativity exists as a byproduct of the exact conditions of the Big Bang.

All of this just chases the fundamental question of what is actually meant by 'creativity,' which was my original point. People use the word, but I suspect it does not mean what they think it means. Without a solid definition, saying something is, or is not, 'creative' is about as useful as saying it's 'pretty' or 'cute' or 'attractive.' It's an eye-of-the-beholder view, not something really subject to factual analysis.

If I, say, pull up the Merriam-Webster definition, I get 'having the quality of something created rather than imitated : imaginative.' If an AI produces something that does not, on a word-for-word level, imitate something, it is 'creative' by that definition, like it or not. It might, in fact, be based on its inputs, but so is 'West Side Story'.

I don't find it the most useful definition, but it certainly doesn't preclude software from being creative. And it will take a lot of philosophy to make either the argument that 'creativity' is either inherently biological or is not, especially when philosophy already considers there to be an open question as to whether any human, anywhere, at any time, has ever been 'creative' at all.

But, if one assumes, for the purpose of argument, that at least some humans are 'creative,' does that say anything about whether machines can, or cannot, be creative? That isn't at all clear to me. We have a reasonable idea of how and why LLMs generate the output they produce (though it is increasingly clear that it is a 'reasonable' idea, even for the top experts in the field, not an authoritative understanding), but we don't really understand how humans generate the output they produce, so there's no prima facia case that we don't produce things in the same way, but using biological computation rather than technological computation.

so there's no prima facia case that we don't produce things in the same way

Rather obviously, there's the case you made towards the start of your post.

AJ

If you mean infinite monkeys, that's the difference between purely random means and 'programmed means.' If human beings are, in the end, entirely mechanical, and the universe is without free will (as noted, a position currently embraced by a number of people in both the physics and philosophy communites), then there is no practical difference between human creativity and AI creativity (whatever 'creativity' means).

Note that the 'free will' argument does not necessarily remove randomness. There is a subset of the community holding that at quantum randomness is a real and non-predetermined thing, but that quantum randomness does not introduce free will, since it is entirely and uncontrollably random.

Mind you, I am not a subscriber to the viewpoint that we / the universe is mechanical and predetermined - but, presumably, I may be predetermined to reject that viewpoint, which would make my rejection of it something of no consequence :)

No, I was thinking of the creation of new words. Okay, it's not right at the top, for which I apologise.

Allegedly DeepSeek's LLM's finest level of granularity is the word, so it can never create new words. When the level of granularity is the syllable, it's not so clear. Since the preceding and following syllables are determined statistically, I suspect those LLM's won't be able to create new words either. But for humans, it's easy. Take a root and build on it according to unwritten/intuitive? rules regarding tenses, prefixes, suffixes etc until you get the exact meaning you want and occasionally you'll come up with a word not in a dictionary.

AJ

There are now AIs that can (and do) create new words based on existing ones. It's even a subject of academic study - see e.g. https://arxiv.org/abs/2502.14900

There are also well-documented cases where two AIs interacting with each other created their own languages. This is perhaps not the definitive article, but it's interesting: Unraveling The Curious Mystery Of Two Different AI Models Suddenly Forming A New Language Of Their Very Own. Note that, in that article, it also mentions AI-created words being formed as part of 'hallucinations.'

With the benefit of further consideration, some AIs seem to be creative in some circumstances.

Those impinging upon the legal profession have been shown to invent non-existent cases to support claims of precedent. And those who ask AI to critique their stories always seem to be told their writing is peer to a Great American Novel even though their prose seems to me clunky and prosaic.

Both those instances are cases where the AI is presumably following an instruction to please the user. But can AIs produce such creativity when the reins are off?

AJ

This is, again, related to the question of what 'creativity' means. The paint shaker is a (semi-)random mechanical process. It 'knows' nothing about art, unlike Jackson Pollock.

On the other hand, a generative art-making AI 'knows' a great deal about art. It has studied an enormous number of artistic works, after all.

There are examples of AI-generated artworks winning art contests, and the prize being taken away once it was known that the work was created by an AI. To my way of thinking, that means either that the judges failed in judging 'creativity' (if the AI is incapable of creativity), or correctly judged 'creativity' but then took a knee-jerk human-centric approach to the term after finding out the creator was not human.

In a way, we're looping around to 'The Imitation Game' aka the Turing Test. While nominally obsolete in 2025, the question posed is not obsolete (again, to my way of thinking): if one posits a black box that might contain a human being or a program, and one is incapable of determining which is within the box based solely on interacting with the entity inside the box, is it meaningful to say that, if the entity is indeed a program, it is not 'human' in some essential way?

Or is there some fundamental way in which only biological organisms can be 'human?' If so, why?

Mind you: I do not believe any extant AI in 2025 has achieved sentience or should be considered human. But that does not necessarily mean they are incapable of being creative. Chimpanzees, dogs, and elephants (among other specials) may or may not be sentient (depending on the definition), and they are not 'human'; they are, however, arguably capable of creativity (again, depending on the definition).

An LLM has no intention behind what it produces - if you give it the same input, any deviation between two outputs is the result of a configurable amount of randomness.

A number of LLMs will actually display their decision-making process, which (at minimum) is a very strong proxy for 'intention.' That it is not human intention is obvious, but that doesn't mean it's not an intention, by a reasonable definition of the word.

While one can turn up or down the randomness of the output, in general an LLM picks every token it outputs with deliberate care, thus matching your claim in the first paragraph.

Again, I'm not arguing that any extant AI is sentient or has human-level 'intention.' But the discussion seems to presume that a lack of human-level intention/creativity/etc equals a lack of any intention/creativity/etc.

Dogs, for instance, would beg to differ. They are certainly not 'human-level', but they have intention and creativity. It is not unreasonable to claim that an LLM - which can explain its decision-making process and the goals it is seeking to achieve at every step along the way - has a much stronger intentionality than a dog. At least in my experience, dogs often fail at maintaining a consistent plan to achieve their goals, yet it would be very hard to argue that they do not have goals or make plans.

I will also say that many authors say their output varies wildly based on environmental factors (time of day, hunger, sleep, etc). Are those 'intentional,' or are they 'random'? Even if every word is carefully chosen in the context of the author's intent, the entire work may be different if the author, say, starts writing at 8am every day versus starting writing at 8pm every day.

Or, try the following experiment: write a non-trivial reply to this comment, or any comment on the forums. Delete it. Write it again. Delete it again. Write it a third time. Each time, maintain the intention to write the same reply.

Do you get the same text each time? I certainly don't, even if I intend to recreate it. Random factors! Are those random factors a feature, or a bug? Or both?

Really, at base, we're endlessly looping around to a fundamental question no one can answer: are human beings enormously complex biochemical machines that intake knowledge, store it, and then emit outputs based on the knowledge they take in, or are they possessed of 'free will' which exists outside of the knowledge they take in and the biochemical means of processing it. If the first, we are merely enormously complicated, rather buggy LLMs, with no more 'creativity' or 'intention' than they have. If the second, then we are 'more,' but the way we are 'more' is highly ill-defined and very poorly understood.

And, as noted, there is a very active community within physics and philosophy that staunchly argue the first point - that 'free will' is entirely an illusion and that the entire universe was predestined at the time of 'creation' (e.g. the big bang). All 'creativity' and 'intentionality' along the way is merely an illusion.

I find that depressing, but presumably I am predestined to feel that way, and my finding it depressing has no impact whatsoever as to whether it is true.

In some ways, this is an unexpected side benefit of LLMs, even at their current rather limited state of development. They do a pretty interesting job of holding up a mirror to humanity and provoking a better discussion of what things like 'creativity' and 'intention' actually mean.

Humanity, thus far, is doing a lousy job of answering those questions. But, maybe we were predestined to do a lousy job if, in the end, they were nonsense words all along.

it's meaningless to say "humans are enormously complicated, rather buggy LLMs"

Meaningless to whom, exactly? It's entirely meaningful, if you take the position that the universe is one big, entirely predetermined machine with no free will or true randomness.

There's a lot of things that humans do which LLMs don't, and a lot of things that humans are good at that computers aren't

Again, not if you believe in the argument that there is no such thing as free will. All humans are 'good at,' in that view, is following the programming created for them at the time of the big bang.

LLMs, thus far, have not demonstrated free will.

The argument has nothing to do with whether LLMs have free will. I am not claiming that they do.

The argument is whether humans have free will. There is a large and active community in philosophy and physics making the claim that humans do not have free will. If they don't, one cannot argue 'free will' as an essential difference between humans and AIs. Nor between either and monkeys banging on typewriters, for that matter.

Hence, I stand by the starting point you copied. Are we just biochemical machines? If so, certainly one can argue that we are far better and more complex machines than extant LLMs, but that has all of the impact of saying that El Capitan (the current fastest supercomputer in the world) is far better and more complex than an IBM 701 (first programmable IBM computer). It is, but they are fundamentally both Turing machines, and the difference in computing is one of scale, not of theoretical power.

Well, it's meaningless if humans have free will because it's wrong.

Yes, by presupposing the thing that is in contention, certain conclusions follow. That completely sidesteps the entire point.

And if the universe is a big entirely predetermined machine with no free will or randomness, then it's meaningless because there's no difference between humans or LLMs or anything else

Yes, that's exactly the point being made here. If that is the case, humans are superior to LLMs in programming and computational complexity, but not in kind.

'Buggy' is a reference to our tendency to do 'strange' things. LLMs are predictable, by comparison, so less 'buggy.' I agree, though, that the distinction becomes somewhat meaningless.

And, again, the point isn't to make the claim that the universe is a single giant deterministic mechanism, nor to make the claim that LLMs and humans are computationally equivalent. The point is to note that such a view is held by a fair number of people with the credentials to back it up, and that one must account for it before making statements like 'humans have creativity; LLMs do not' or 'humans can do things LLMs cannot.'

Or, of course, one need not account for it, since, in that case, one is only making the statements one has always been predestined to make :)

If 'it' is 'AIs are not creative', then yes, we are in agreement. It's a meaningless statement without an enormous amount of context.

Something we agree upon! It's not just a religious concept.

See, for instance, Does Quantum Mechanics Rule Out Free Will? or Quantum Mechanics and the Return of Free Will, amongst many others.

Or books like "Determined: A Science of Life without Free Will" by Robert Sapolsky or "The Science of Free Will" by Samir Varma.

Those are all science-centric discussions of 'Free Will' versus 'Determinism' in one sense or another.

There's a lot of interest in the topic right now, and people approaching it from a variety of perspectives. Some of those people are avowed atheists and not interested in the religious aspects at all.

I don't agree that it's meaningless if humans don't have free will. Did you miss the text you quoted saying 'It's entirely meaningful, if you take the position that the universe is one big, entirely predetermined machine with no free will or true randomness'? How can you quote that and then proceed to assert that I said it was 'meaningless if humans don't have free will'?

You would have a case in saying that literally everything is meaningless if humans don't have free will. That's an overly broad statement, in my opinion. One must presuppose some things are meaningful or there is no point in a discussion. And, if some things are meaningful, that statement would be one of them.

I'm paying plenty of attention. You're trying to avoid the entire point of the discussion by presupposing something that is the point of contention ('human beings have free will') or trying to use meta arguments to avoid obvious conclusions ('if human beings don't have free will, everything is meaningless').

But, certainly, if you are in agreement, you also would have to agree that 'AIs are not creative' is a meaningless statement without an enormous amount of context not generally in evidence when people make that statement. After all, if my statement ('humans are enormously complicated, rather buggy LLMs [in the case that there is no free will]') is 'meaningless', as you contend, then the statement 'AIs are not creative' is at least as meaningless, and for the same reasons.

For reference, 'the plot' here is the argument that terms like 'creativity' require far more definition than is commonly given, and any such definition needs to take into account arguments from philosophy and physics arguing that human beings may also effectively lack 'creativity' in any meaningful sense of the term.

I'm certainly still connected to the plot. Are you?

I do believe that the question of whether or not free will exists is a meaningless one, for what it's worth!

It's meaningful in the context of 'the plot', as you called it.

If free will exists, there is, very likely, a difference of kind between creativity in humans and extant LLMs.

If free will does not exist, there is, almost certainly, no difference of kind between creativity in humans and extant LLMs.

That such a difference exists affects the meaning of the usually ill-defined term 'creativity' when discussing LLMs vs humans. Since that is 'the plot' here, it is absolutely meaningful to this discussion.

Note: we will not get an answer to whether free will exists. Obviously. That is not 'the plot.' The point is that as long as the existence of free will is a matter of controversy, it is a convenient way of pointing out how poorly defined 'creativity' is, since a topic of current controversy totally changes the discussion as to 'creativity.'

Your wit wins the day, sir. I am obliged to doff my hat in appreciation.

Brilliant analogy!

I've given several speeches in my Paying-Job life on the topic of Enshittification and its effects on the workplace. Spoiler Alert: poorer tools correlates with poorer outcomes/products. The concept is aimed at corporations first and foremost, i.e. Microsoft, Google, E.A. When it comes to writing, the concept is more complicated. Is Dan Brown (The Da Vinci Code) writing crappier novels or was the first novel just as crappy, but it was new crap to us?

I was told as a teenager that I would not read 90% of the books in a bookstore. She never told me why. Derivative plots and writing has always plagued the writing world, and funnily enough, was the source of the boom that was the pulp novels of an earlier age or the Romance novels of the past fifty years. Is this enshittification?