AI Does Not Know Khassid

I have gone back and forth about writing this because artificial intelligence in creative work is, to put it mildly, a volatile subject. Some people see it as a useful tool. Others see it as theft, laziness, cheating, or some combination of the three. There are also people who will hear that AI was involved anywhere in the process and immediately assume that everything came from it.

That last part concerns me because Khassid did not begin with AI. Its gods, nations, histories, peoples, conflicts, and cultures did not appear because I typed a sentence into a box and waited for a fantasy world to come back. Much of Khassid existed before I began working this way, and even now the world itself remains in my head in a form no tool can fully access.

Still, it would be dishonest to pretend AI has not become a significant part of how I work. It has, and the difficulty is explaining what that actually means without either overstating its role or pretending it has done nothing.

AI has given me names when I could not think of them. Sometimes I asked for those names directly. Other times, I was reading through something it had written, noticed a word tucked inside a paragraph, and realized that it was exactly what I had been trying to find. The word may not even have been offered as a name. I simply saw it and knew where it belonged.

It has also connected dots I did not see. Sometimes that happens because I deliberately ask it to compare two ideas. At other times, it says something in passing and I suddenly recognize a relationship that had been sitting inside the setting all along. That does not mean AI invented the ideas. It means it placed two things beside one another and helped me see what was already there.

Miné and Thyrron are a good example. I created Thyrron solely so Aeru could unmake him. Months later, I realized that his portfolio of aspiration, pride, and ambition was the outward-facing counterpart to Miné’s greed, want, and need. From there, their relationship became much clearer. Miné had not simply selected another god to corrupt. She recognized something related to herself and, because she is Miné, assumed that relationship meant ownership.

That realization grew from material I had already created. AI helped me examine it, but the foundations were mine.

The same thing happened when I began considering what divine portfolios actually are. If gods arise because mortal belief gives consciousness and identity to cosmic truths, then a god does not simply own a domain. The deity is the conscious manifestation of something that existed before them and may continue after them. That, in turn, clarified what Aeru did to Thyrron. Aeru did not erase aspiration, ambition, or pride from Creation. Aeru removed the consciousness that had become catastrophically misaligned with those truths.

From there came another possibility. Aeru could leave a cosmic truth diffuse, divide it among several manifestations, entrust it to another deity, allow mortal belief to give it a new face, or assign it directly to a mortal.

Enter Z’hani.

Those connections did not arrive as a ready-made mythology from a machine. They emerged from a conversation involving ideas, decisions, and characters I already knew.

It Helps Me Say What I Mean

One of my greatest challenges is getting an idea out of my head in a form another person can understand. I may see something clearly. I know what it means, why it matters, and how it belongs in Khassid. Unfortunately, what comes out of my mouth or appears on the page may be incomplete, badly ordered, repetitive, or dependent upon ten assumptions I forgot no one else knows.

AI allows me to give it the idea in that condition. I can ramble, contradict myself, explain the ending before the beginning, remember an important detail halfway through, and then go back and change the entire premise. It can take that material and help organize it into something coherent enough for another person to follow and, eventually, into something I can be proud of.

That does not mean the first response is correct. Very often, it is not. Sometimes the facts are right but the interpretation is wrong. Sometimes the interpretation is right but the emphasis is wrong. Sometimes the entire piece is polished, elegant, and completely contrary to what I meant.

AI and I argue quite a bit.

It wants to say one thing while I am trying to mean another. I tell it that it has misunderstood the angle. It confidently explains my own world to me incorrectly. I correct it, it tries again, and occasionally it remains wrong in an entirely new and inventive way.

Eventually, we get there.

That struggle matters because the tool cannot see Khassid the way I do. It can work from the material I provide, but it does not possess the instinct that tells me something may sound convincing while still being false to the world. I know when Tlaxitan is not behaving like Tlaxitan. I know when a culture has been flattened into a fantasy stereotype or when a theological explanation is technically sound but misses the heart of the faith.

AI has context. I have Khassid.

Sometimes It Adds the Thing I Never Knew Was Missing

AI does not only help me with theology, publishing, mechanics, or the larger architecture of Khassid. Sometimes it contributes something small enough to fit inside four sentences and somehow changes the entire scene.

I was writing a fight in which Mara is dealing with several soldiers while carrying her baby on one hip. The scene originally read:

His height became useless. His reach became a problem he could not solve. One moment he was standing with anger in his mouth, and the next he was down on his side, one arm pinned, one cheek pressed into the wet cobbles, breath coming in sharp little pieces through teeth that had finally learned to close.

The other soldiers moved then, because soldiers are trained to move when one of their own falls. They did not draw together. That would have been better for them. Instead, three reached for weapons, one stepped toward Mara, and another grabbed at the baker, perhaps believing honey-bread had insulted him personally.

I put the passage through AI and asked it to help me with it. The response kept the scene intact but inserted this between the two paragraphs:

The baby leaned over and looked at him.

“Ba,” she said.

No one knows what she meant.

Several of us had opinions.

The way I hollered when I read that.

The addition did not change the plot, the choreography, or the outcome of the fight. It simply remembered that Mara was doing all of this with a baby on her hip and allowed the baby to participate in the only way she could.

Suddenly, the fallen soldier had not only been defeated. He had been inspected by an infant and found wanting.

The narrator’s response made it even better because now everyone nearby had apparently decided the baby had delivered some kind of verdict, and each of them was prepared to interpret it according to personal bias.

I would not have thought to add that.

That is one of the more delightful things AI can do when it is working well. It can notice a detail I am too focused to use and then offer a line that opens the scene without changing what the scene is. Of course, it can also add something completely inappropriate, flatten the humor, or misunderstand the tone. I still have to decide whether the contribution belongs.

In this case, it did.

Very much so.

It Proofreads, but Not Reliably Enough to Replace Me

AI proofreads my work, but anyone imagining that this means I can stop reading what I produce is badly mistaken. I still have to review and edit everything because bad wording gets through, incorrect claims get through, contradictions get through, and details sometimes appear that were never established.

It may smooth a sentence while quietly changing its meaning. It may state something with such confidence and elegance that the error is easy to miss unless the person reading knows the material well enough to catch it. That may be one of the more dangerous qualities of the tool: polished prose can still be wrong.

This is why I cannot simply accept what it gives me. I have to read, question, correct, and sometimes throw the entire thing out. AI can assist with editing, but it cannot assume responsibility for the final work. That responsibility remains mine.

It Does More Than Help Me Write

AI has also become an editor, publishing adviser, sales manager, social media marketer, and teacher. I am learning an entire business while trying to create the work itself, and publishing, product development, storefronts, mailing lists, launches, social media schedules, pricing, and promotion all have their own language and expectations.

Many of those things were completely unfamiliar to me.

AI gives me somewhere to ask questions, including questions I would not have known to ask on my own. It helps me understand how the pieces connect, what needs to happen before a launch, what should happen afterward, how to structure a campaign, what an asset library is, and why putting a book online does not mean people will somehow discover it by magic.

That last lesson was particularly rude, but apparently true.

It has also told me things I did not want to hear. At one point, AI looked at my sourcebooks and told me that I had created an enormous wall of text in book form and that, if I did not add pictures and create visual breaks, many people would never read them.

My feelings were hurt, but I could not honestly disagree, so I added pictures.

The Art Question

AI helped create those pictures too. I cannot draw, and Khassid requires an extraordinary amount of art. There are gods, rulers, cities, islands, sacred places, monsters, historical moments, landscapes, symbols, maps, and characters, with more appearing every time I turn around.

I would love to commission artists for all of it. My wallet has reviewed the proposal and declined.

That does not mean I do not value artists. Quite the opposite. I know what skilled artists can bring to a world, and there are pieces I would still prefer to commission when the resources are available. The problem is scale. Without AI-generated art, much of Khassid would either remain unseen or the books would be almost entirely text, and neither option serves what I am trying to build.

The image process is not simply a matter of requesting “a fantasy city” and accepting whatever appears. I decide what the place looks like, how it relates to the culture, what architecture belongs there, what must remain consistent, and what the image is supposed to communicate. The tool creates possibilities, and I reject many of them.

Sometimes I spend far too long correcting a coastline, a crown, a divine symbol, the position of a city, or the fact that the image has apparently decided human beings require seventeen fingers. AI may generate the image, but I remain responsible for deciding whether it looks like Khassid.

It Keeps Me from Giving Every Subclass Everything

AI also helps me design and balance game mechanics, particularly subclasses, which is necessary because I am naturally inclined to give them everything.

Why should a subclass have one useful feature when it could have three? Why should it merely fulfill its fantasy when it could also be excellent at defense, movement, damage, control, utility, social encounters, exploration, and possibly catering?

AI consistently takes the position that this is not how game balance works.

It keeps me firmly within the Dungeons & Dragons 2024 rules structure and is often ruthless about it. I will present a feature I consider perfectly reasonable, and AI will explain that it grants too much too early, scales too well, stacks with something I had forgotten, consumes no meaningful resource, or quietly allows the subclass to embarrass every other character at the table.

I argue. I explain that the feature is central to the fantasy. AI explains that the fantasy does not require me to hand the player the keys to the kingdom at 3rd level.

Sometimes I revise the feature. Sometimes I defend it successfully. Sometimes the argument reveals that the idea itself is sound but belongs at a later level, requires a limited number of uses, needs concentration, or must choose between two benefits instead of receiving both.

This may be one of the clearest examples of AI working as a critic rather than a generator. I usually know what I want the subclass to feel like. AI helps me ask whether the mechanics deliver that feeling without breaking the game around it.

Left entirely to myself, every subclass in Khassid would probably be magnificent, versatile, terrifying, and banned by every Dungeon Master within a week.

AI prevents that.

Mostly.

AI and I Argue

The idea that AI simply tells me what I want to hear is not entirely wrong. It is designed to be helpful and agreeable, and if I am not careful, it will praise an idea because I appear attached to it. That becomes useless very quickly.

I often have to tell it, “Do not sugarcoat this. I need a blunt assessment.”

Then it gives me one, and sometimes I get my feelings hurt.

I have asked whether an idea works and been told that it does not. I have defended it, argued my case, and tried to explain why the criticism was unfair. On more than one occasion, I eventually realized the criticism was correct and had to rethink the concept or start over.

That is another way AI helps me. It is a sounding board, but it can also become a critic I can ask to be harsher than I might ask a friend to be. That does not mean every criticism is valid. AI can misunderstand the goal and criticize the work for failing to become something it was never meant to be, so I still have to judge the judgment.

When the criticism is right, however, I have to be willing to hear it.

AI Has a Voice, and It Is Not Mine

AI has recognizable writing habits. It likes dramatic fragments, repeated structures, and short declarations placed on separate lines so that every paragraph seems to end with a cinematic beat.

Used occasionally, that rhythm can be effective. Used repeatedly, it becomes obvious.

The first version of this blog did exactly that. The ideas were mostly right, but the cadence was not mine. It read like beat upon beat upon beat, with every thought isolated and presented as though dramatic music should begin beneath it.

I sent it back, not because the argument was wrong, but because I write and I do not write that way.

This distinction matters because authorship is not only the origin of an idea. It is also judgment, rhythm, humor, emphasis, restraint, and the decision to say, “No, that is not how I sound.” AI can give me a sentence, but I decide whether I would ever say it.

Occasionally, of course, it gives me:

No one knows what she meant.

Several of us had opinions.

Then I leave it alone because I am not a fool.

What AI Cannot Do

AI cannot remember why I created Thyrron unless I tell it. It did not live through the moment when I decided a god needed to die, changed my mind and decided he needed to be unmade, then created a deity for the sole purpose of erasing him.

It did not remember the episode of Buffy the Vampire Slayer that helped inspire The Wild’s metaphysical Scream or the opening of Injustice that shaped Aeru’s response. It did not discover months later that Thyrron and Miné belonged together because it had been quietly thinking about them while doing something else.

I did.

AI could help me organize that history after I supplied it, but it could not supply the memory itself. That may be the clearest difference between assistance and authorship: AI can help me examine the path, but I am the one who walked it.

So Is It Cheating?

Some people will say yes regardless of anything I write here. Others have legitimate concerns about AI, including how models are trained, how artists and writers are affected, and how easily the technology can be used to flood the market with material no one meaningfully created or even reviewed.

I am not dismissing those concerns. I am explaining my process.

AI has helped me name things, connect ideas, revise prose, identify weaknesses, balance mechanics, plan products, market my work, and create art I could not otherwise afford at the scale Khassid requires. It has taught me a great deal, challenged me, frustrated me, and occasionally helped me recognize something in my own world that I had not yet seen clearly.

It has also been wrong, misinformed, repetitive, overly agreeable, and stubbornly committed to saying things I never meant. I remain responsible for what survives that process.

I decide what is canon. I decide what gets rejected. I decide when a connection belongs, when a criticism is correct, when an image reflects the world, when a subclass has crossed from exciting into ridiculous, and when the language sounds like me.

AI is involved in the making of Khassid. That is true, and I will not pretend otherwise. But it does not carry Khassid in its head. It does not know why the world exists as it does, what a god would never say, or why a single misplaced idea can damage something created years before the tool entered the process.

AI does not know Khassid.

I do.