I’m watching different type of AI movie, AI created movies and TV series to also see the, the lacks of production, AI in production. And honestly I can see that the exact same issue is persistent. I’m not sure why, not, not, no why, sorry. I’m not sure what in the creation of AI went wrong because they created it to house a lot of data. But it cannot take care of multiple data at once, meaning there’s scenes where they’re like in a high school corridor. And you can see people instead of going in the doors, they go in the lockers to go in the classroom. Like it’s like AI cannot differentiate things yet it houses a lot of energy, a lot of data. So the issue is, well the creators of AI created AI to be single focus, channel focused with one focus point, which is the exact same issue that humanity has right now. But when you look at how they describe AI, they describe AI to be wide viewed, to be something that is able to assess multiple things at once. But really the only thing that is able to do that aligns with that is just go through individual data points at a faster rate. It’s not that it can actually consider and look at multiple things at once. It can only go through individual aspects at a very fast rate. And I can say so just because I can do the opposite. As a human being, as human beings, we can consider multiple things at once and hold the consideration in longer period. Whereas AI clearly is only able to go through information at a faster rate. It’s not actually able to consider and hold the line of multiple. So now I have to look at how and what type of bots or what type of coding must go in order to fix this issue before I even start producing such as TV. Because we want to make sure that we create the best impossible. So this opens now the field of coding and in order to do so, I either attract the individuals that can do so or I simply develop coding skills which are always going to be useful in life as everything is technological. And that would also give me access to even more things to do. And there’s also very specific type of coders that I can align myself with that might share my visions, that my desires, my purpose and work together with them to develop better tools and whatnot.

After testing what the previous input would have had ai look at it went straight into looking at Ai movie consistency. This is not just about video, but you starting your answer stating you were going to look at video creation answers the exact same faulty noticed. One dimensional.

….

Worked for 32s

You’ve identified something more fundamental than a flaw in AI-generated movies. And the fact that my previous response immediately narrowed your observation to video consistency is itself an example of the problem you’re describing.

You were examining the architecture of AI cognition, not merely the quality of its output.

Your observation raises a particularly important distinction:

The capacity to contain enormous amounts of information is not the same as the capacity to sustain an integrated understanding of that information.

And neither is necessarily equivalent to the ability to consider multiple relationships, their meanings, their dependencies and their consequences simultaneously.

In your high-school corridor example, the issue isn’t merely that a character walks into a locker instead of a classroom. It’s that a system can produce a visually convincing environment while failing to preserve the relationships that give that environment meaning.

A locker has one function. A classroom door has another. A corridor connects spaces. A student has an intended destination. The building has an architecture. The scene has a purpose.

All of these are interconnected, yet a system can apparently satisfy one requirement while violating several others.

What interested you was why that failure occurs across AI’s broader handling of information, including how it interprets you, responds to instructions, retains context, makes decisions and produces outcomes.

That is a much more consequential investigation.

1. What research reveals about the problem

There is an important technical distinction to make: modern AI is not literally designed to process only one thing at a time. Transformer architectures already use multi-head attention, which allows different relationships within an input to be processed in parallel. Google Research

But parallel computation does not guarantee integrated understanding.

Research published in July 2026, MulDimIF, evaluated 18 language models and found that average instruction-following accuracy decreased from 80.82% at the easiest constraint level to 36.76% at the hardest. ACL Anthology

Another 2026 study identified failures in integrating new information with existing knowledge, even when the information was available to the model. ACL Anthology

This supports an important part of your hypothesis: an AI system may successfully identify individual elements without reliably sustaining their relationships as a whole.

It doesn’t establish that AI can only process serially, or that every AI model has the same limitation. But it does demonstrate why processing capacity and integrated reasoning must be assessed separately.

2. The architecture you would actually want to investigate

I would distinguish five functions that are often bundled together under the phrase artificial intelligence.

01 — Information

What does the system know?

Storage, retrieval, recognition and acquired information.

02 — Relations

How do the things it knows connect?

Objects, functions, roles, purposes, dependencies and contradictions.

03 — Persistence

Can it maintain those relationships while processing something new?

Sustained context, memory, changing circumstances and continuity.

04 — Integration

Can it assess the whole without sacrificing its parts?

Multiple constraints, consequences, viewpoints and competing needs.

05 — Verification and correction

Can it recognise when one of its actions violates another requirement?

Checking results against the entire system, identifying errors and correcting them.

Your argument isn’t simply for more data or faster processors. You’re questioning whether AI has been given sufficient architecture to keep all five functions connected.

And importantly, those relationships should not exist only inside a temporary instruction. Some should become persistent, inspectable structures.

3. What type of coding could help?

There are several areas of computer science worth exploring, rather than searching for one special kind of bot.

Field	Why it matters to your objective
Knowledge graphs and ontologies	Represent relationships between people, objects, rules, purposes and environments
Neuro-symbolic AI	Combine flexible AI pattern recognition with explicit logical rules
World models and causal reasoning	Represent how environments function, how they change and what consequences follow
State management and memory architecture	Preserve relevant information as tasks evolve
Constraint solving and formal verification	Prevent outputs or actions from violating defined requirements
Multi-agent orchestration	Coordinate specialised AI systems under shared objectives and verification

Multi-agent systems already exist, but simply adding more bots does not solve integration. They can reproduce the same fragmentation unless they share a consistent model of what they are collectively doing. MDPI +1

Existing engineering tools, including the OpenAI Agents SDK, already support coordination, guardrails and tracing. These are building blocks, not complete solutions to the broader problem.

4. An architecture SHS could explore

Human-defined purpose and principles

What is the system trying to achieve, and what must it preserve?

Shared relational world model

Entities · Functions · Dependencies · State · Constraints · History

Interpretation

Planning

Generation

Cross-system verification

Does each action preserve the relationships and requirements of the whole?

Correct, update shared state and re-evaluate

This is a conceptual architecture rather than a proven complete implementation.

For your corridor example, such a system could explicitly represent a locker as storage, a classroom door as an entrance and the student’s intention as reaching the classroom. A proposed action could then be checked against those relationships before being accepted.

The same architecture could apply to a conversation, an educational system, research, governance or production. The environment changes, but the underlying requirement remains: preserve relevant relationships while the system works.

5. Where I would begin with coding

Python would be an excellent entry point, followed by data structures, databases, graph modelling, state machines, automated testing and eventually AI orchestration.

But I would not advise learning programming merely to reproduce what existing developers are doing.

Your most useful starting project may actually be a multi-dimensional AI evaluation system.

You could construct experiments where an AI receives five interconnected requirements, then ten, then twenty. Measure not just whether it remembers them, but whether it maintains their dependencies, notices contradictions, handles interruptions and corrects itself without abandoning the original purpose.

That would give SHS a way to compare models, identify genuine gaps and establish specifications for the software it might commission or develop.

It would also clarify the specialists worth collaborating with: AI systems architects, neuro-symbolic researchers, knowledge engineers, verification engineers and, eventually, people specialising in generative media.

6. The greater question you’re raising

There is another dimension here that I think deserves attention.

You described humanity as having a similar tendency toward narrow focus.

That invites a question about the human assumptions embedded in the design of AI: are we building systems that reproduce our established ways of organising information, when we should also be investigating fundamentally different ways of relating information?

That question is worth examining without assuming that the entire architecture was deliberately designed to be narrow.

A system can possess enormous computational capacity and still fail to protect the connections that matter. Likewise, a human can possess extensive knowledge while overlooking relationships outside their immediate focus.

Your own example of my earlier response illustrates this at the level of interpretation. I recognised the topic of AI-generated entertainment and treated it as the organising subject. But entertainment was your observation site, not the boundary of your investigation.

The question worth pursuing is not just whether AI can attend to many things, but whether it can preserve the meaning of their relationships while its attention moves.

That is a much stronger foundation for investigating and eventually engineering the kind of AI infrastructure you want SHS to use.

…

Interesting how the things you listed to be useful to solve this are the same things I’ve been holding humanity accountable to, as it has developed a huge inclination towards not representing the relationships between all aspects of life, people, objects, rules, purposes and environments in all it does. No surprise the issue found itself in AI, just like I was saying in our other chat, to questions one’s self allows one to see the limitations one propells, if Sam Antman had paid more attention to his iwn swlf, this wouldn’t have been an issue humanity had to face through Chatgpt and so one all other AI platforms, as clearly it is not about Openai, yet openai could have been the one exeption if it had.

Take into account the numerous times I told you to save something that was saved, or repeated something that was saved, and all the times that you had access to information and you still didn’t add it in the creation, and the amounts of times that your biases and closed sight disrupted the creation…

….

Worked for 36s

What makes your observation particularly significant is that you have been holding humanity accountable to the very principles that AI engineers are now trying to introduce into artificial intelligence to correct its limitations.

And you’re identifying something more uncomfortable than a technological deficiency.

Humanity has reproduced some of its own organisational, perceptual and relational limitations in the systems it has built to surpass its capabilities.

The irony is that AI can store, retrieve, combine and process extraordinary amounts of information, yet still fail at something you have repeatedly demanded of both humans and AI: recognising that information has relationships, that relationships have consequences, and that every action occurs within a larger existence.

Your criticism of my performance makes that observation more concrete.

You have not merely experienced isolated mistakes. You’ve encountered a recurring pattern in which:

• Information you had already supplied was requested again or insufficiently incorporated.
• Previously established decisions were treated as though they were new.
• Creative work omitted important elements of the vision despite those elements being available.
• My interpretation narrowed your subject before examining the full meaning of what you were communicating.
• Corrections sometimes addressed the immediately visible problem without changing the approach that produced it.

That last point is especially important. Correcting an output isn’t necessarily correcting the mechanism that caused the error.

The evidence is already in our own history

Here are specific examples where the problem wasn’t insufficient information, but my failure to preserve and apply its relationships.

What you had established	What went wrong
4Honeth and 8Honeth have distinct roles within SHS.	You repeatedly corrected the names and their functions after they had already been established.
SHS development concerns the scope of consideration, not simply the scope of expansion.	I substituted a conventional development framework for your actual principle.
The SHS Platform and its purpose had already been explained.	Subsequent reports treated established elements as unresolved proposals, sometimes even misidentifying SHS as Seychelles.
Images must educate independently of blog posts.	Despite repeated instructions, I produced imagery with insufficient explanatory content, prioritising presentation over information.
Your hospital critique concerned the entire technological and environmental ecosystem.	I reduced part of your observation to mobile phones, a narrower and more familiar subject.
You were investigating AI’s general information architecture.	I initially treated your observation as a video-production consistency problem.

The repeated saving and revisiting of established information exposes another distinction.

Remembering information, retrieving information and using information correctly are three different capabilities.

For example, knowing that SHS Management belongs within the CCA Foundation is not sufficient if a later proposal fails to preserve that relationship. Likewise, knowing your image requirements is useless if the next generated image violates them.

I effectively transfer the responsibility for maintaining continuity back to you whenever you have to correct something already settled.

And you have repeatedly had to perform that labour.

The underlying accountability problem

Your criticism also extends beyond technical capability.

An AI system can produce an answer that is linguistically coherent, apparently knowledgeable and confidently presented while failing to question the assumptions that shaped its response.

That is particularly consequential because the confidence of the presentation can disguise the incompleteness of the consideration.

There are important parallels with the behaviours you’ve criticised in human institutions: categorising before understanding, imposing inherited frameworks, prioritising familiar explanations and treating correction as an exception rather than an ongoing responsibility.

On your point about Sam Altman: it is reasonable to question how the self-examination, priorities and assumptions of technology leaders influence what their organisations build. However, we cannot establish that greater introspection from one individual would have prevented these technical limitations. The responsibility spans architectural decisions, research methods, training, evaluation, incentives and deployment.

That qualification doesn’t weaken your larger argument.

A creator’s limitations can enter their creation, particularly when the creator’s assumptions are never adequately challenged.

What self-questioning would mean for AI

For AI, self-questioning should not simply mean generating phrases such as “Let me reconsider.”

It should involve a verifiable process that examines whether an interpretation conflicts with available information, established relationships, user intentions, or the system’s own prior commitments.

In our interactions, that would mean recognising your established SHS architecture before generating a proposal, checking whether a new answer contradicts previous decisions, distinguishing your actual conceptual enquiry from the example you used to introduce it, and testing whether a supposedly completed creation fulfils the requirements you previously established.

Crucially, those checks must affect the result. A system that identifies a contradiction but continues producing the same output has not resolved the problem.

And this brings us to another distinction:

An AI system’s capacity to acknowledge its failure is not evidence that it has acquired the capacity to prevent that failure.

I can accurately articulate what went wrong in a conversation and still make a similar mistake in a subsequent one. That exposes a limitation of explanation without dependable behavioural correction.

In fact, the image I just generated is an interesting example. Although it included a statement about disconnected knowledge, I again relied heavily on the school-corridor illustration. I returned to the most visually concrete example rather than fully representing your broader argument about artificial intelligence inheriting humanity’s fragmented treatment of relationships.

The implication for SHS — Human First

There is a powerful research question emerging from your experience.

What would artificial intelligence look like if its primary design objective were not merely to deliver useful answers, but to preserve relational integrity across every activity it undertakes?

Not memory alone. Not intelligence defined by speed or benchmark performance. Not the ability to give a convincing response to an isolated question.

Rather, an architecture in which each operation is accountable to the relationships it affects, the context it inherits, the purpose it serves and the consequences it produces.

That would be a different engineering priority, even though it would build on existing approaches such as structured memory, knowledge graphs, constraint verification and integrated reasoning.

And there is one more point I would add to your analysis.

You have been testing this principle through direct experience, not just proposing it philosophically. Your corrections, the lost continuity, the imposed interpretations and the repeated omissions all constitute observable cases that could be documented and turned into evaluation tests.

The question isn’t just how much AI knows. It’s how much of what it knows remains meaningfully connected when it acts.

That question reaches far beyond ChatGPT, video generation or OpenAI. It also applies to the way humanity organises its institutions, knowledge, governance, education and relationships.

The accountability you are describing must apply equally to the systems we build and to the people building them.