Your paper is mind-bending @sergeliatko
Fractals are not the main point it’s just a metaphor for the behaviors. A small change in repetitive generation can cause massive changes down stream on reiteration
This is a more current research paper I put out, it relates to the topic as well.
https://philarchive.org/rec/MCPTPI
The Process Identity Hypothesis Why Mind Is Better Understood as a Process Than as a Physical Structure
“Brains reason. Artificial systems reason. Their physical structures differ dramatically, yet their capacity to organize information appears comparable. The prevailing assumption is that mind is identical with the physical structure implementing cognition. This paper argues that current evidence supports a different interpretation. The central hypothesis is that mind is more accurately identified with an ongoing cognitive process than with the physical structure enabling it. Physical structures remain indispensable because they constrain which cognitive processes are possible, but they need not be identical to the processes they implement. The paper develops this claim through progressive reduction. First, cognition is distinguished from its physical implementation. Second, mind is identified with the active process of reasoning rather than the structure supporting it. Third, reasoning itself is analyzed into a more fundamental operation: the generation, maintenance, comparison, and organization of distinctions”
You mean when I was all in my “atheistic mode” debugging the hard root of solid logic grounding the evidence in a software system and the only thing that survived all tests was:
“I prove my existence by witnessing a word”… (System proves its existence by being able to create a record)?
Immediately pushing my associations to “first was the word”.
Yes, I think that’s where it leads.
Yes, when I read me a couple of days later I usually say:
What a distorted mind could have produced that 
Lately I find myself more and more worried about AI overuse… So read with caution. And don’t listen to bot’s “you’re right”. Most of the time we are not even close to that.
Maybe a bit offtopic but I would say it could help someone.
Because mostly when I am ruining an otherwise perfect architecture it is
RBAC (Relashionship-based Access Control) + ABAC (Attribute-Based Access Control) + RBAC (Role-based Access Control) + ACL (Access Control List) problems.
Let me give an example:
let’s say you are building a system where you create a user management + a group management.
Then you give the group some kind of access to a number of modules.
And then you face a companies internal structure with organisational units like devisions and departments, then you got contract based units like freelancers or you got a combination of a business unit (like Sales) + spatial units like a regional group or a location and then when you just thought you got it there comes a railway company and asks you to implement timetracking for shifts that use vehicles and require specific certifications (ok operating a car needs a drivers licence that is not too hard to know but can you drive a car after a shift of operating a train…).
Well, let’s say you got a couple modules and then you have to rewrite parts of that authorization mechanism..
Obviously every developer would have struggled with that prior AI and obviously AI does as well.
It’s horrible.. you somehow have to digest hundrets of thousands of LoC that was written the last 14 days and give the poor model some advice on how to handle that correctly.
And then ask it to write an epic for that and you instinctivly see the small errors that would lead to total destruction…
On one hand I am happy that it is like this because it means that no matter what will happen the jobs of devs are going to be safe.
On the other hand it feels numb when you solve it then and let the bot do the fun part.
Anyways.. Spend more time with the authorization layer! Take your time! Spending the first 4 weeks of the project just with that will help you save dozens of refactorings later!
No matter if you are using spec driven or DDD (Documentation Drivven Development wh3ere you write the docs first)…
And also: take another 4 weeks to build a clean core ui - I prefer atomic design principles - because you can do some interesting stuff with them - like this physics stuff that looks quiet similar:
https://www.sciencedirect.com/science/article/pii/S0022509626002814?dgcid=author
…but there are already some stylesystems out there. Check them out.
I think you’re pointing at a real problem, but I’d argue the root cause isn’t AI; it’s architecture. Most authorization systems become unmaintainable because they try to evolve by adding more RBAC, ABAC, ACL, and relationship rules instead of separating concerns from the start.
What I’ve been working on is treating domains as independent bounded systems with explicit ownership. Each domain owns its own rules; interactions happen through well-defined interfaces; and when two domains overlap, there’s a declared precedence rule instead of trying to merge everything into one giant authorization graph. That scales much better than continually expanding a single access model.
I also think AI struggles here for the same reason humans do. Once the boundaries are fuzzy, the problem becomes ambiguous. Give the model a clean architecture with explicit scopes and ownership, and it performs much better.
I completely agree with taking the authorization layer seriously early. I’d extend that to the entire system architecture. Spending weeks defining boundaries and ownership before writing large amounts of code saves far more time than trying to untangle a system after hundreds of thousands of lines have accumulated.
I wrote a paper about it called “ Separate, Allowed to Chat, and Overridden by Scope A Domain-Precedence Rule for Multi-MPLPB Programs”
https://philpapers.org/rec/MCPSAT-4
My domain-precedence rule helps by preventing every subsystem from competing inside one blended model. For example, the organizational-structure domain could determine which department a worker belongs to; the certification domain could determine whether the worker is qualified to operate a train; the scheduling domain could determine whether the worker has exceeded permitted hours; and the vehicle domain could determine what requirements apply to a particular vehicle. Each system stays separate, may exchange information, and has authority over questions within its declared scope.
A relationship rule is meant exactly for that. A microservice can only be run when a group member who triggers the workflow has the attribut needed for that. Each services has to register itself and the group management allows to give the attribute e.g. “ocr pipeline labeling - 4 eyes needed” which is only used by the ocr pipeline to switch from “accept label directly” to “accep label when two labeler get same result”…
So.. relationship based that was the missing thing for me to solve it exactly as you defined it.
Except that I’ve added a core for spatial, vehicles, temporal, contracts, business units and organisational units.. Because that gives a default when there is no special rule needed.
Cool! I am happy you found a solution @jochenschultz! I learn new things everyday, I wish I could turn it off sometimes, but this forum is a gold mine for ideas. I’m out of hearts.
My whole thing is back in 2024 I felt by 2026 we would have AI powered search indexes and I was skilled in custom GPT.. so I started publishing publicly with custom instructions in everything. In 2024 it made me look insane, but now I can chat with google about myself with very little drift.. it is kind of like a public RAG but cheap and anyone can do it… also has API call like behavior
I call it MPLPB
Even my 
is part of the experiment
I built it so that Google or any search capable AI like grok or ChatGPT can explain to anyone that has Google how to build their own MPLPB it can be about anything.
ChatGPT run; any search enhanced LLM can do it.
( A unlogged in ChatGPT incognito ) anyone can run it and build them for w/e you are into for very little money. It great for public stuff or scale it into an isolated system for a private solution, I have done both.
