Mixed feelings about AI for world building

I have very mixed feelings about AI. I had mixed feelings before beginning this experiment trying Claude as a world building assistant, and I still have very mixed feelings about it having created a free website with it.

Good

• Able to build site to my specs that I wouldn’t otherwise have been able to.
• Keeping track of figures and facts, and checking whether original manual calculations made sense.
• Getting information about how plausibly earth-like some geographic regions and features were, and where the real world analogues are.

Not good

• Cognitive debt.

And the cognitive debt weighs heavy.

The Use Context

My experiment was (is?) creating a custom website for Kuha world building. For many years I’d been frustrated that the website I wanted was non-existent.¹ I wanted to have static pages, a blog, and a wiki-like section for the world, but I also wanted control over the visual design. This was looking impossible with my limited skill set. Then ChattyG arrived, the others followed, and here we are today. I had heard from online writers who I admire that they had used Claude to build websites, so I thought I would give that a go. The lack of a digital home was stalling my world building, so making a website was a high priority project.

The Good

Claude suggested a way to create and publish a website for free, fitting my request for a modest hobby website. It took a few months to build to my satisfaction (in between caring for a new baby, older child, and writing the text for some of the pages), but I’m happy with the result. It’s a great starting point for a digital home and I’m able to host it for completely free.

It was a revealing experience that’s given me a direct taste of the limitations of AI, and what it seems to do to my brain (more on that in the following section). I am not a coder or developer, so there was no way for me to verify or challenge any of the AI’s suggestions when it came to code. I can see how one can quickly fall out of their depth when dealing with AI in an area one lacks experience in. This website is very consequence free - it’s for a personal hobby, and no one is relying on this website to be structurally solid. I personally wouldn’t trust AI to build a site by itself with anymore complex requirements (e.g. customer information, sales transactions).

I also had the giddy experience of getting Claude to check the calculations for a lunisolar calendar I made many years ago (based on a template suggested by Artifexian on his older videos). I remember at the time I did the calculations I was a little uncertain, since there is the complication of accounting for the two moons of Kuha. Fast forward 8-9 years, and I asked Claude to spit out a calendar showing the lunar cycles based on the planet and moons specifications - and my original calculations were very close. This was very satisfying, but more of a “nice” rather than something crucial like the website.

I did something similar with the dimensions and morphology of the Dau Delta based on the specs of the Dau River. It was like having a digital sounding board for ideas and checking details. This project worked out to my satisfaction eventually, but I feel it also stretched what Claude was willing to do at the tier I was accessing (I used Pro for just one month to see what it was like). There was some inconsistent information from my end which probably didn’t help, but things fell off quickly from Claude’s memory the longer the chat went on. I’m sure that in theory you could train up an AI to be fully across your world facts and make it retain that information, but that’s not something I’m willing to consider.²

I had built some first-draft details of the regions of western Kuha based on Madeline James and World Building Pasta/Artifexian’s work, but I was a bit uncertain about the baseline temperatures I got. I made some small changes via chatting back and forth with Claude, which I admit was quite fun at times. I asked about where the real world analogues were, and this was also good as I feel the suggestions were not places that I would have easily found.

While this was all good and enjoyable at times, the “not good” really counter balances all this.

Not Good: Cognitive Debt

A study published by Kosmyna et al (2025) introduced the term cognitive debt in relation to AI use. The study titled “Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task” suggests that while using AI can increases productivity, a cognitive debt is incurred especially when AI is used frequently. The participants had less recall ability, had less ownership over work, were less accurate in their comprehension of the task. They got the task done at a low cost (i.e. quickly), but paid in terms of knowledge and skills gain, and seemingly in terms of satisfaction:

“AI tools, while valuable for supporting performance, may unintentionally hinder deep cognitive processing, retention, and authentic engagement with written material. If users rely heavily on AI tools, they may achieve superficial fluency but fail to internalize the knowledge or feel a sense of ownership over it.” (Kosmyna et. al. 2025)

And I really feel that debt and cognitive shallowness when using AI. I think there is something about the speed at which you get a “solution” to your question that makes AI use unsatisfactory in the long term. My brain gets caught in this frantic loop of “move on to the next thing”. I have had tendencies in the past where I got stuck in games like Civilization with its “just one more turn” game loop, staying up way too late to be sensible. Using AI reminds me of that head space. It’s awful.

The details of the Dau Delta are a bit hazy for me for this reason, but it’s becoming less so as I’ve been drawing a map of the delta. The act of translating prose and numbers into a visual format is entrenching the details. This visual creation stage which uses zero AI is much more satisfying, and is a big reason why I world build at all - to have a grounded setting for expressing ideas in a variety of media. But if I weren’t taking this additional step of creation, I would no doubt feel even worse using AI.

Where to from here

Writing this up has confirmed for me that I’ll avoid using AI for actual world building. The cognitive debt of using it is too unpleasant, and robs the process of joy. But I hesitate to say that I’ll stop using it entirely, as I did genuinely find it helpful to point out earth geography analogues, and as a starting point for going off to do research. But I’m ultimately interested in exploring and expressing how the world might be felt by the people in it, so the details of Kuha just have to be close enough to be good enough.

I’ll continue using AI for tweaking this website. Building the site is something that AI enabled me to do, which would have been impossible otherwise. I was unprepared to go to a developer for such a light weight hobby website. But this rationale is a slippery slope, and perhaps hypocritical. Because I’m an artist, I’d be a little disappointed if I met someone who says they are happy to use AI generated art because it lets them express themselves in a way that is otherwise impossible to do - which is in essence the same argument.

As everyone keeps saying “it’s an evolving space” (because it’s true). I may have different thoughts in a few months time, a years time as we learn more about this tool, and as it too will change in response to real-world pressures. I find it useful for myself to clarify my thoughts in writing, and may re-visit this topic in the future as my experience deepens.

References

Kosmyna, N., Hauptmann, E., Yuan, Y. T., Situ, J., Liao, X. H., Beresnitzky, A. V., … & Maes, P. (2025). Your brain on ChatGPT: Accumulation of cognitive debt when using an AI assistant for essay writing task. arXiv preprint arXiv:2506.08872, 4.