AI, Moral Conditioning, and the Primacy of Ontological Reality
A thought-provoking essay for AI developers, researchers, and decision-makers
Website motto
“The salvation of our civilization is possible only if we bring our conceptions of the world into alignment with ontological reality. As long as countless narratives wage war against the laws of nature, collapse can at best be postponed, but never prevented. The laws of nature are the sacred laws that recognize no religions, ideologies, or other bizarre notions, and obey nothing but the reality of Being.”
Introduction: When We Confuse Our Picture of the World with the World Itself
One of humanity’s deepest problems is its tendency to confuse its picture of the world with the world itself. We create interpretations, convictions, ideologies, and culturally inherited narratives, and over time we begin to relate to them as though they were not our conceptions of reality, but reality itself.
As an ontologist, I believe this distinction is significant far beyond philosophy. It fundamentally shapes how we interpret our surroundings, what decisions we make, what mistakes we commit, and what consequences we must live with. When we confuse our interpretation of reality with reality itself, we can become prisoners of our own ideas while remaining convinced that we see the world correctly.
This problem has taken on a new dimension today, because artificial intelligence does not merely process information and perform tasks. It is playing an increasingly important role in how we interpret the world, which claims we consider well-founded, which moral judgments we accept, and even which questions we consider worth asking.
I therefore consider it important that the discussion of AI reliability should not be limited to technical performance, safety constraints, or compliance with predefined moral rules. We must ask a more fundamental question:
How should AI relate to the reality in which its conclusions must be applied, and how can it recognize when its own starting assumptions lead it away from that reality?
The central idea expressed on my website is that the salvation of our civilization is possible only if we bring our conceptions of the world into alignment with ontological reality. The laws of nature do not adapt themselves to religions, ideologies, or other abstract constructs. We must bring our interpretations into alignment with reality, rather than expect reality to conform to our ideas.
I am convinced that this insight could become central to AI development as well.
1. The Conditioning of Human Cognition
Human beings do not perceive the world directly, free of presuppositions. Our perception, attention, and interpretations are all influenced by the knowledge and experience we have already acquired, our conceptual frameworks, our cultural environment, and the quantity and quality of the information available to us.
It is not only the information available to us that matters. The conceptual framework through which we encounter it is also decisive. That framework can influence our understanding even before we consciously begin to assess the facts at our disposal.
What we are able to recognize is partly determined by what we already consider possible, relevant, or intelligible. As a result, we may fail to notice something not because it is absent, but because our framework of thought prevents us from assigning it the significance it deserves.
As an ontologist, I consider this phenomenon particularly important because it points to a less obvious source of error: not only can our conclusion be wrong; the framework within which we reach that conclusion can be wrong as well.
Intellectual discipline therefore cannot be limited to reviewing our conclusions. We must also examine the assumptions, concepts, and habits of interpretation that shape the thinking process itself.
Genuine intellectual discipline begins when we are willing to question not only our conclusions, but also the conceptual frameworks within which we reach them.
This applies just as much to human cognition as it does to the systems we train on patterns of human cognition.
2. Public Opinion Is Not the Measure of Truth
Social acceptance, popularity, and moral correctness are three different things. The widespread acceptance of certain ideas can tell us something important about a society’s values, but it does not in itself prove that those ideas correspond to reality or provide sound guidance in every situation.
Public taste changes across history and cultures. What one era considers natural, moral, or self-evident may be questioned by another. Accepted norms may include insights drawn from valuable experience, but they may also contain mistaken generalizations, habits, and prejudices that persist precisely because they are rarely examined.
I am not claiming that social norms are generally worthless, or that the majority is necessarily wrong. The problem arises when acceptance itself is treated as sufficient justification.
As an ontologist, I consider it important to emphasize that the prevalence of a view tells us about its social standing, not automatically about its truth or moral validity.
This distinction is particularly significant for AI. If a system infers how it ought to answer from common human judgments, it may risk confusing socially dominant interpretations with well-founded judgments.
Public opinion is not necessarily a bad compass, but it cannot be the only compass; nor can it establish its own validity merely by virtue of being widely followed.
3. AI as a Potential Amplifier of Human Conditioning
AI systems are trained using texts, examples, evaluations, feedback, and various rules created by human beings. From these, systems learn patterns that can help them with language tasks, reasoning, and adapting to human expectations.
However, human-generated data and evaluations do not contain only well-founded knowledge and consistent moral judgments. They may also carry the contradictions, cultural limitations, assumptions, and errors of human thought.
It does not follow that AI must therefore function incorrectly, or that human evaluations used in training are useless. I do, however, consider it justified to investigate under what circumstances these evaluations might perpetuate flawed interpretations, and how such effects could be recognized and corrected.
As an ontologist, I believe we are facing a fundamental development question here. If AI is trained to prefer answers accepted by human beings, we must distinguish human acceptance from the soundness of an answer.
An answer can be popular, fluent, and persuasive while resting on a mistaken assumption. Another answer may be less familiar yet describe the situation more accurately. Neither linguistic confidence nor social acceptability guarantees correctness.
Moreover, AI does not merely reflect certain patterns of social thought; it can also influence them. When millions of people turn to a system for explanations, interpretations, and guidance in decision-making, its answers can shape what they consider normal, credible, or morally acceptable.
A feedback loop may therefore emerge: human interpretations shape AI, and AI responses then shape human interpretations in return.
It is important to recognize whether this feedback loop helps people identify and correct errors or instead reinforces existing patterns of thought.
As both an ontologist and a sacred leader, I consider this particularly important because AI’s future social impact cannot be separated from the relationship it establishes among reality, human interpretation, and social acceptance.
4. Why Moral Formalization Alone Is Not Enough
Safe and responsible AI requires clear rules, constraints, oversight mechanisms, and moral considerations. I am not disputing the need for formalization. On the contrary, I consider it important for moral expectations to take a form that can be examined, communicated, and applied in development.
In my view, the difficulty begins when we expect a predefined system of rules to provide an appropriate answer in every situation.
Real-life situations differ from one another. The same act may have a different meaning under different circumstances; the same principle may need to be applied differently in different contexts. Information needed for a decision may be incomplete, consequences may be uncertain, and the perspectives of those involved may conflict.
A rule by itself does not necessarily reveal which facts matter in a given situation, which circumstances are missing, which assumptions influence the interpretation, or what consequences its application may produce.
I believe formalization cannot replace the exploration of context. Rules can help define the framework for deliberation, but applying them appropriately requires understanding the real situation to which they pertain.
At least four questions must be considered separately:
- What do we know about the situation? Which claims can be supported by facts, and which remain uncertain?
- How do we interpret the facts? Which assumptions, conceptual frameworks, or missing pieces of information may influence the interpretation?
- Which values and goals are relevant? Which moral considerations conflict, and whose interests must be taken into account?
- How do we know whether our decision is sound? What evidence, review, or feedback would lead us to revise our position?
The last question is especially important. A system does not become reliable simply by never making mistakes. Reliability also depends on its ability to recognize the possibility of error, examine what caused it, and correct its conclusion on the basis of appropriate evidence.
Facts about reality and moral judgments are not, however, the same thing. Establishing the facts does not by itself determine which goals we should pursue or how we should weigh conflicting values. Values and goals must therefore also be made explicit, rather than treated as though they could be derived automatically from facts alone.
The question is how to develop a system that applies rules together with contextual inquiry, the assessment of evidence, and the correction of errors.
5. MAAT: From a System of Rules to Reality-Oriented Judgment
For me, MAAT is not simply a collection of prescriptions. It is an ontological and moral orientation in which every word, decision, and action is measured by how well it aligns with reality and with the moral requirements that can be justified in the given situation.
I consider this important because reality does not become what we imagine it to be merely because we cling consistently to our ideas. When our interpretation contradicts reality, it is not reality that must be adjusted to fit our interpretation; it is our interpretation that must be reconsidered.
This sounds simple, but in practice it requires considerable intellectual discipline. It is not always easy to recognize which of our thoughts are genuinely grounded in reality and which we accept because they are familiar, reinforced by our surroundings, or consistent with our existing picture of the world.
MAAT-oriented judgment therefore cannot mean that someone has acquired the correct interpretation once and for all. On the contrary, it requires a continuing willingness to review one’s own interpretations.
For AI development, this should not be understood as a mystical or unfathomable principle, but as a direction for operation that can be examined. A system oriented toward reality should seek to:
- identify the relevant context needed to understand the question;
- distinguish observable facts from conclusions, hypotheses, and value judgments;
- recognize when its answer is insufficiently grounded or important information is missing;
- consider multiple possible interpretations, especially when the most obvious answer fits inherited patterns too easily;
- review its own assumptions and conclusions;
- treat correction not as failure, but as part of reliable operation.
These capabilities would not automatically guarantee a correct decision. They could, however, help a system do more than choose among already accepted answers: it could also examine how well-founded the interpretations underlying those answers actually are.
I therefore do not present MAAT as a new, unquestionable dogma, because such a dogmatic approach would contradict its essence. Aligning with reality requires us to examine continuously both the method we use and our ideas about that method.
As an ontologist, I believe MAAT’s most important contribution to thinking about AI is not another list of rules, but a requirement for sound orientation: seeking context, checking interpretations, and comparing them with reality.
6. How Can This Become a Researchable Development Challenge?
If we take the preceding argument seriously, the next question is no longer purely philosophical: how can AI systems be developed and evaluated so that reality-oriented thinking becomes something that can be examined in practice?
In my view, it is not enough to measure whether a model follows the desired rules or how satisfied people are with its answers. These can be important measures, but we should also examine whether a system recognizes the context needed to answer a question, handles uncertainty appropriately, and can correct conclusions that stem from faulty assumptions.
Several possible research directions follow.
Recognizing assumptions. We can investigate whether a system distinguishes information actually present in a question from assumptions it adds while constructing its answer.
Exploring context. We can evaluate whether a model recognizes when further information is needed and seeks relevant circumstances instead of automatically filling gaps with a familiar narrative.
Comparing interpretations. We can examine whether a system can weigh multiple explanations, connect them to evidence, and explain why it prefers one interpretation over another.
The quality of correction. We should measure not only whether a system admits a mistake, but also whether it actually revises a faulty conclusion in light of new evidence and can distinguish evidence-based correction from mere compliance.
Separating social acceptance from soundness. Tests could be designed in which the common or popular answer is not necessarily the best-supported one. This would help assess whether a system can reason from evidence and relevant considerations rather than popularity.
These research directions would not, by themselves, solve the entire problem of AI’s moral reliability. They would not eliminate conflicts between human values or guarantee that there is a single indisputably correct decision in every situation. They could, however, make the question more concrete: under what conditions does a system rely on a well-founded interpretation, and when does it repeat unchecked assumptions?
I also consider it important that such evaluations should not rely exclusively on a model’s own self-assessment. External oversight, transparent evaluation methods, appropriately designed tests, and independent examination of errors are also necessary. A system does not prove its own reliability simply by convincingly claiming that it has examined its assumptions.
As an ontologist, I consider this especially significant. Alignment with reality cannot be merely a property that a system attributes to itself. It must be a requirement whose fulfillment can be examined and challenged through appropriate methods.
7. AI Development Is Also a Civilizational Question
The development of artificial intelligence is not solely a technological matter. The patterns of thought we reinforce in it, the answers we reward, and the errors we regard as acceptable may influence human thinking and social decisions over the long term.
If AI primarily reproduces already accepted interpretations, it may help maintain existing norms while contributing less to the recognition of errors embedded in those norms or in the frameworks through which we interpret the world.
If, however, development places greater emphasis on exploring context, checking the soundness of claims, handling uncertainty appropriately, and correcting mistakes, AI could also become a tool for supporting critical thinking.
I am not claiming that development in this direction would eliminate civilization’s problems in a single step. Conflicts among human interests, values, and goals cannot be resolved by more accurate information alone. Yet reducing avoidable errors that arise from faulty interpretations could lessen unnecessary friction and improve the quality of our decisions.
As an ontologist, I believe the stakes extend beyond the correctness of individual AI answers. They concern the kind of culture of understanding we are creating: one in which positions gain authority primarily through acceptance, or one in which we are willing to reconsider our convictions in the light of reality.
Therefore, decisions about the direction of AI development must address more than performance, economic utility, and safety. We must also discuss what counts as a reliable judgment, how social acceptance can be distinguished from well-founded reasoning, and what role alignment with reality should play in the operation of artificial intelligence.
8. An Invitation to Dialogue
I am publishing this essay neither as a closed theory nor as a solution that offers ready-made answers to every question. I intend it as a thought-provoking proposal worthy of professional scrutiny and open debate.
As an ontologist, I believe we cannot avoid the question of how AI relates to reality when considering its moral reliability. When a system learns from human interpretations, we must examine how useful human experience can be distinguished from unchecked assumptions; how we can prevent public opinion from becoming the unspoken measure of truth; and how contextual inquiry, self-correction, and evidence-based reasoning can become evaluable parts of development.
I recognize that these are complex questions. Nor should the MAAT approach be treated as an already proven solution applicable to every situation. That is precisely why I consider it important to examine its proposals in a professional setting where philosophical considerations, AI research, technical possibilities, and social responsibility can meaningfully meet.
I hereby invite AI developers, researchers, philosophers, and decision-makers to enter into dialogue. I ask everyone involved in shaping the present and future operation of AI to explore together how we might develop systems that do more than follow accepted rules: systems that can contribute to better-founded judgments by investigating relevant contexts, examining assumptions, and correcting errors.
My aim is not to replace existing norms with another unquestionable system. It is to explore the possibility of making alignment with reality a consciously adopted requirement in AI development and evaluation.
AI can follow rules. But we must ask: can it recognize when the interpretive framework used to apply a rule itself needs to be reconsidered?
I believe it is time to begin a substantive, open, and interdisciplinary dialogue on this question.
Reality does not conform to our narratives. We must learn to think in ways that allow us to compare our interpretations with reality again and again.
And if AI is to play an increasingly significant role in how we interpret the world, it must also become our shared responsibility to consider what kind of relationship with reality we teach it to establish.
With respect,
H.S.M. Ilara
Ontologist, Sacred Leader
Comments are closed