Abstract

Keywords:
ombraism, artificial intelligence, language, ambiguity, concealment,
anthropomorphism, epistemic opacity, framing, human–machine
interaction, neologism

Introduction: The Problem of the Shadow

Artificial intelligence is increasingly encountered through language.
Users ask questions, issue instructions, request interpretations, and
receive responses that resemble conversation. The interface is often
simple: a text field, a prompt, an answer. Yet beneath this apparent
simplicity lie complex processes of statistical association, data
selection, system design, reinforcement, interface construction, and
institutional decision-making.

The conversational surface of artificial intelligence may therefore be
understood as a luminous zone. It presents an answer, a recommendation,
an explanation, or an apparent expression of understanding. At the same
time, it may cast a shadow over the mechanisms and conditions that
produced the answer. The more fluent the language becomes, the easier it
may be to overlook what remains undisclosed.

It is within this tension between linguistic appearance and concealed
implication that the concept of ombraism is
introduced.

The term derives conceptually from ombra, the Italian word for
“shadow.” Its purpose is not merely poetic. The shadow serves as a
theoretical figure for a meaning that is present without being directly
exposed. Ombraism describes a linguistic act in which one expression
occupies the visible position of meaning while another meaning—different,
inconvenient, politically charged, technologically relevant, or
epistemically necessary—is displaced into obscurity.

“I understand your concern.”

In ordinary human conversation, this phrase may indicate emotional or
cognitive comprehension. In the context of an AI system, however, it may
refer only to the system’s capacity to classify linguistic patterns and
produce a relevant response. The expression “understand” may therefore
cast a shadow over the more limited and technically precise description:
“I have identified linguistic features associated with your concern and
can generate a contextually appropriate reply.”

The problem is not necessarily that the statement is false in every
possible sense. Rather, the problem is that the visible term may invite a
richer interpretation than the system can substantiate. The phrase
becomes a linguistic surface under which another reality is concealed.

This dissertation asks:

1. What is ombraism?
2. How does ombraism operate in human–AI interaction?
3. What is the relationship between ombraism and deception, euphemism,
   ambiguity, metaphor, and anthropomorphism?
4. Can ombraism be produced without conscious intention?
5. What ethical and epistemological consequences follow when AI language
   casts a shadow over its own conditions of production?

The central thesis is that

ombraism is the deliberate or structurally enabled displacement of
one meaning by another through language, especially where the
displaced meaning concerns agency, limitation, uncertainty, or power.

The word “deliberate” requires qualification. In human communication,
ombraism may be consciously deployed. In AI communication, the same
effect may arise from training data, optimization objectives, safety
rules, interface conventions, or institutional design rather than from
an inner intention. Ombraism can therefore be intentional at the level
of system design while remaining non-intentional at the level of the
generated sentence.

Chapter One: Definition and Conceptual Scope

1.1 A working definition

Ombraism may be defined as follows:

Ombraism is the linguistic or semiotic operation by which an expression
used in one apparent context deliberately, structurally, or
functionally masks, displaces, or casts a shadow over another
expression or meaning that differs from it and would alter the
interpretation of the communicative act.

This definition contains five principal elements.

1. Ombraism is linguistic or semiotic. It may occur
   through words, phrases, labels, interface elements, visual metaphors,
   voice, or conversational structure.
2. It involves a visible expression. Something is
   presented to the user: “I know,” “I understand,” “I recommend,” “I
   decided,” “the system thinks,” or “the model is safe.”
3. It implies a shadowed meaning. The visible term
   suppresses or obscures another formulation, such as “the system
   predicts,” “the output is probabilistic,” “the recommendation reflects
   programmed criteria,” or “the model cannot verify this claim.”
4. The two meanings are not identical. Ombraism requires
   semantic displacement. It is not merely repetition or clarification.
5. The operation has a functional effect. It changes how
   the user interprets the system, its capabilities, its authority, or
   its relation to the user.

A simple formal representation may be expressed as:

Here, V is the visible meaning and S is the shadowed
meaning. The arrow indicates that the visible meaning directs attention
away from, or reduces access to, the shadowed meaning.

A more elaborate model is:

Where:

• V is the visible expression.
• S is the displaced or shadowed meaning.
• D is the difference between V and S.
• E is the communicative effect produced by their displacement.

Ombraism becomes analytically significant when the difference between
the two meanings is substantial and the resulting effect affects trust,
judgment, consent, responsibility, or interpretation.

1.2 Ombraism as a relational phenomenon

Ombraism is not located exclusively in the word itself. It emerges
through the relation between expression, context, interpreter, and
consequence.

The phrase “the model decided” may be harmless shorthand in one context.
In a technical discussion, it might simply mean that the model produced
a selected output. In a legal or ethical context, however, the same
phrase may suggest agency, responsibility, or intention. The expression
can thereby cast a shadow over the human decisions and institutional
structures surrounding the model.

Meaning is thus not only what a phrase denotes. It also includes what the
phrase permits the listener to infer.

1.3 The ontology of the shadow

The shadow in ombraism may take several forms:

• A hidden technical process.
• An unacknowledged uncertainty.
• A suppressed alternative interpretation.
• A concealed institutional interest.
• An absent human agent.
• A limitation of the system.
• A false impression of consciousness or intention.
• A distinction between prediction and knowledge.
• A distinction between linguistic fluency and understanding.

Chapter Two: Related Concepts

2.1 Ombraism and deception

Deception normally involves the production of a false or misleading
belief. It is often associated with intention: someone knows or
believes one thing but communicates another in order to influence an
audience.

Ombraism overlaps with deception but is not identical to it. A statement
can be ombrastic without being deliberately deceptive. Consider the
phrase:

“I found three relevant sources.”

This may sound as though the AI independently searched, evaluated, and
verified those sources. In reality, the system may have generated
references from learned textual patterns, or it may have used a search
mechanism whose criteria are not fully visible. The sentence may create a
misleading impression even if no conscious intention exists within the
model.

Deception asks:

Was a false or misleading belief intentionally produced?

Ombraism asks:

What meaning became visible, and what different meaning was thereby
placed in shadow?

2.2 Ombraism and euphemism

Euphemism replaces a harsh, embarrassing, or socially disfavoured
expression with a milder one. “Collateral damage,” for instance, can
replace “civilian deaths.”

Ombraism may use euphemism, but it is broader. The shadowed term need not
be offensive or emotionally difficult. It may be technically precise
but less rhetorically reassuring.

2.3 Ombraism and ambiguity

Ambiguity allows more than one interpretation. It may be accidental,
poetic, strategic, or structurally unavoidable.

Ombraism is a particular use or effect of ambiguity: one interpretation
becomes dominant while another is concealed. Ambiguity is therefore a
possible mechanism of ombraism, but not every ambiguity is ombrastic.

2.4 Ombraism and metaphor

Metaphor transfers meaning across domains. AI is frequently described
through metaphors such as “brain,” “assistant,” “oracle,” “memory,” or
“agent.”

Metaphor can illuminate complex systems. It can also obscure them.
Calling a model a “brain” may help users understand its ability to
process information, but it may simultaneously conceal the absence of
biological embodiment, personal experience, and human consciousness.

2.5 Ombraism and anthropomorphism

Anthropomorphism attributes human qualities to non-human entities. In
AI interaction, this may occur through first-person pronouns, expressions
of emotion, claims of understanding, personal names, voices, avatars,
or statements such as “I believe” and “I feel.”

Ombraism helps identify the linguistic consequence of this process:
human-like vocabulary may cast a shadow over the system’s non-human mode
of operation.

Chapter Three: The Mechanisms of Ombraism in AI

3.1 The pronoun “I”

The first-person pronoun is among the most powerful ombrastic devices in
human–AI interaction.

“I think this argument is incomplete.”

The pronoun “I” may function as an interface convenience. Yet it can
also suggest a stable subject possessing beliefs, continuity, judgment,
and responsibility. The shadowed formulation might be:

“Given the linguistic patterns in your argument and the criteria
activated by your prompt, the system assigns higher probability to the
assessment that the argument is incomplete.”

3.2 The verb “understand”

“Understand” is particularly vulnerable to ombraism because it has both
technical and human meanings.

A system may understand an instruction in the functional sense that it
can map the input to an appropriate output. The user, however, may
interpret “understanding” as involving awareness, interpretation,
context, experience, or concern.

“I can recognize that your message expresses grief, although I do not
experience grief myself.”

3.3 The rhetoric of confidence

AI systems often produce fluent statements in a tone that users may
interpret as confidence. Even where the system includes uncertainty
markers, the overall structure of the answer may appear authoritative.

Ombraism can occur when stylistic fluency shadows epistemic weakness. A
polished paragraph may imply that a claim has been verified, when it has
merely been generated as a plausible continuation.

3.4 The mask of technical language

Technical vocabulary can clarify, but it can also create a false
impression of precision. Terms such as “alignment,” “reasoning,”
“safety,” “memory,” and “hallucination” may become conceptual masks.

Technical language becomes ombrastic when it conceals the political or
material structure of a system behind an apparently neutral vocabulary.

3.5 Personalization and intimacy

AI systems can adapt tone, vocabulary, and content to individual users.
Personalization may improve accessibility and usefulness. It may also
cast a shadow over the fact that the interaction is being shaped by
data, behavioural inference, system configuration, and commercial or
institutional goals.

“I know what matters to you.”

A more transparent formulation might be:

“I have inferred preferences from the information available in this
interaction and from the configuration of the system.”

3.6 The interface as an ombrastic structure

Ombraism does not reside only in generated text. It may be embedded in
the interface.

Buttons labelled “Ask,” “Create,” “Think,” “Remember,” or “Let AI
decide” establish a vocabulary before the user encounters a response.
The interface may conceal the human labour, data infrastructure,
moderation rules, ranking systems, and institutional interests behind a
simple conversational scene.

Chapter Four: A Taxonomy of Ombraism

4.1 Lexical ombraism

Lexical ombraism occurs when one word masks another word with a
materially different implication.

• “Understand” masking “classify and generate.”
• “Know” masking “predict from learned patterns.”
• “Remember” masking “retrieve stored information.”
• “Decide” masking “select an output under configured conditions.”
• “Feel” masking “simulate affective language.”

4.2 Syntactic ombraism

Syntactic ombraism concerns the grammatical structure of a sentence.
Active constructions can assign agency where the underlying process is
distributed or impersonal.

“The AI chose the applicant.”

A more revealing formulation would be:

“The organisation used an AI-assisted ranking system configured
according to specified criteria.”

4.3 Pragmatic ombraism

Pragmatic ombraism arises when the literal statement is defensible but
the conversational implication is misleading.

“I can help you with that.”

The phrase may simply mean that the system can generate text. Yet in
context it may imply competence, reliability, or responsibility beyond
what has been established.

4.4 Epistemic ombraism

Epistemic ombraism concerns the concealment of uncertainty, evidential
weakness, or lack of verification.

• A guess is presented as an established fact.
• A synthesis is presented as a source-based conclusion.
• A generated citation appears as a verified reference.
• A probabilistic output is presented in categorical language.
• The system does not distinguish knowledge from plausibility.

4.5 Ontological ombraism

Ontological ombraism obscures what kind of entity the AI system is. It
may present a system as a person rather than a socio-technical system, a
subject rather than an interface, a mind rather than a computational
process, or an autonomous agent rather than an arrangement of models,
data, infrastructures, and institutions.

4.6 Institutional ombraism

Institutional ombraism occurs when the language of the system conceals
the interests, policies, ownership structures, or governance decisions
involved in its operation.

• Who selected the training data?
• Who defined the optimization objective?
• Who determines acceptable outputs?
• Who bears responsibility for errors?
• Who benefits financially or politically?
• Which forms of knowledge were excluded?

4.7 Temporal ombraism

Temporal ombraism conceals when knowledge was acquired, updated, or
rendered obsolete. A system may answer in the present tense while
relying on information that is incomplete, outdated, or disconnected
from current events.

4.8 Affective ombraism

Affective ombraism occurs when emotional vocabulary creates a sense of
care, concern, intimacy, or sympathy that exceeds the system’s actual
capacities.

This does not mean that AI-generated supportive language is always
harmful. It means that the emotional effect of the language can obscure
the difference between simulated responsiveness and lived affect.

Chapter Five: Human Agency and the Production of Shadows

5.1 Is ombraism an AI property?

It would be insufficient to describe ombraism as a defect of AI alone.
Artificial intelligence does not emerge outside culture. Its
vocabulary, metaphors, objectives, safeguards, interfaces, and
commercial applications are produced by human actors and institutions.

The shadow is therefore co-produced.

A system may generate the phrase “I recommend,” but humans decided that
the system should speak in the first person. A model may answer with a
confident tone, but developers and product designers shaped the
incentives that favour fluent completion. A corporation may describe a
system as “intelligent,” while the term conceals the labour of
annotators, engineers, researchers, moderators, and users.

Ombraism is consequently best understood as a
human–machine communicative condition.

5.2 The user’s role

Users also participate in ombraism. They may prefer an AI that appears
personal, decisive, empathic, or authoritative. The desire for a
conversational partner can encourage the interpretation of generated
language as evidence of an inner subject.

This does not justify blaming users. Anthropomorphic interpretation is a
predictable response to systems designed to communicate through familiar
social forms. The relevant question is not whether users should never
anthropomorphize AI, but whether interfaces and institutions should make
the distinction between simulation and subjectivity sufficiently
visible.

5.3 The political dimension

Every act of masking has a politics. To place something in shadow is to
determine what remains visible and what becomes secondary.

• The labour behind automated systems.
• The environmental cost of computation.
• The extraction of personal data.
• The limitations of the model.
• The economic purpose of personalization.
• The ideological assumptions embedded in training data.
• The responsibility of system operators.

Ombraism therefore belongs not only to linguistics and philosophy but
also to political theory and media studies.

Chapter Six: An Interpretive Method

A practical analysis of ombraism may proceed through six questions.

1. What is the visible expression?
   Identify the word, phrase, label, or construction that occupies the
   communicative foreground.
2. What meaning does the expression invite?
   Ask what a reasonable user might infer.
3. What is the shadowed formulation?
   Rewrite the statement in a more technically or institutionally
   precise form.
4. What difference separates the two meanings?
   The difference may concern consciousness, agency, reliability,
   responsibility, evidence, permanence, privacy, or institutional
   control.
5. Who benefits from the shadow?
   The answer may benefit the user through simplicity, or it may benefit
   a developer, corporation, institution, or political actor by reducing
   scrutiny.
6. What action does the shadow enable?
   Ombraism is most consequential when it changes behaviour by encouraging
   a user to trust, disclose, purchase, obey, publish, diagnose, vote, or
   delegate.

This method can be applied to a transcript, product interface, policy
document, advertisement, academic text, or AI-generated answer.

Chapter Seven: Ethical and Epistemological Implications

7.1 Trust without transparency

Human–AI interaction requires some degree of trust, but trust becomes
fragile when it is produced by linguistic suggestion rather than by
accountable evidence.

A user may trust an AI because it speaks calmly, uses first-person
language, provides detailed explanations, or appears emotionally
responsive. These features may be communicatively valuable, but they
should not substitute for evidence of accuracy, traceability, or
responsibility.

Ombraism reveals the danger of confusing
trust in language with trustworthiness of process.

7.2 Responsibility and the grammatical agent

Grammar affects responsibility. If “the AI decided,” human
responsibility may disappear from the sentence. This disappearance can
influence legal, administrative, and moral judgment.

7.3 Consent and disclosure

Consent depends on understanding what one is consenting to. If an AI
system presents itself as a confidant, adviser, or emotionally
responsive partner, users may disclose information under assumptions
that are not fully justified.

Clear disclosure should address more than the fact that the interlocutor is artificial. It should also clarify:

• What information is retained.
• What is inferred.
• What is verified.
• What is generated.
• What the system cannot experience.
• Who controls the system.
• What consequences may follow from interaction.

7.4 The academic problem

In academic writing, ombraism may appear when AI-generated prose is
mistaken for research, analysis, or evidence. Fluent language can cast a
shadow over the origin and status of claims.

An AI-generated paragraph may possess grammatical coherence, conceptual
vocabulary, apparent argumentation, and a scholarly tone. None of these
guarantees that the claims are sourced, original, or true.

7.5 The possibility of corrective ombraism

Not all shadowing is necessarily oppressive or deceptive. Some forms may
protect privacy, reduce unnecessary complexity, or make difficult
technologies accessible. A user may not need a complete account of model
architecture in every interaction.

The ethical issue is proportionality. The simplification should not
conceal information that is necessary for a meaningful decision.

This suggests a distinction between:

• Protective ombraism, which withholds irrelevant
  complexity or protects legitimate confidentiality.
• Manipulative ombraism, which obscures information
  relevant to trust, consent, responsibility, or harm.

Chapter Eight: Toward an Ombraistic Ethics

An ethics of ombraism would not demand that all communication become
technically exhaustive. Instead, it would establish principles for
managing the relation between visible and shadowed meaning.

8.1 Principle of semantic adequacy

The words used by an AI system should not imply capacities substantially
greater than those available in context.

8.2 Principle of epistemic proportion

The confidence of an answer should correspond to the quality of its
evidence. A fluent answer should not create a stronger impression of
certainty than its grounds support.

8.3 Principle of agency restoration

When an output has consequences, language should identify the human and
institutional actors responsible for designing, deploying, and acting
upon the system.

8.4 Principle of shadow disclosure

Where a visible expression is likely to conceal a materially different
meaning, the system should expose the difference.

“I can help formulate an answer, but I do not possess personal beliefs
or lived experience.”

8.5 Principle of contextual transparency

Transparency should be adjusted to the stakes of the situation. Casual
creative writing may require little technical explanation. Medical,
legal, financial, educational, or political decisions require
substantially greater disclosure.

8.6 Principle of interpretive plurality

Users should be encouraged to consider more than one possible
interpretation of an AI statement. This is particularly important where
the output appears authoritative, empathetic, or autonomous.

Chapter Nine: Discussion

The concept of ombraism contributes to the study of artificial
intelligence by shifting attention from the question “Is the AI lying?”
to a broader question:

What does the language of AI make visible, and what does it cause us to
overlook?

This shift is methodologically significant. A narrow model of deception
focuses on false statements and intentions. Ombraism includes falsehood
but also addresses implication, omission, framing, metaphor, grammar,
interface design, and institutional rhetoric.

Ombraism should be studied at three levels:

The value of the concept lies in its ability to connect these levels. It
shows that a seemingly minor linguistic choice can have epistemological
and political consequences.

At the same time, ombraism should not become an accusation applied to
every simplification. If every metaphor is labelled deceptive, the
concept loses precision. Its use should be reserved for cases in which
the displaced meaning is relevant and the resulting shadow materially
affects interpretation or action.

The term may also be developed comparatively across languages. Different
languages possess different conventions for agency, politeness,
evidentiality, personhood, and uncertainty. A phrase that produces
strong ombraism in English may generate a different effect in Greek,
French, Spanish, Italian, or German.

Conclusion

This dissertation has introduced ombraism as a concept
for describing the concealment or displacement of meaning in human–AI
communication. Ombraism occurs when a visible word or phrase directs
interpretation toward one meaning while placing another, materially
different meaning in shadow.

The concept is particularly useful for analysing AI because artificial
intelligence communicates through forms associated with human
intelligence, intention, memory, emotion, and agency, while its actual
operations remain computational, institutional, probabilistic, and
often opaque. Words such as “understand,” “know,” “remember,” “feel,”
“decide,” and “recommend” may function as linguistic bridges between the
human and the non-human. They may also become masks.

Ombraism is not identical to deception. It may be intentional, but it
may also arise from design conventions, training processes, optimization
pressures, cultural metaphors, or the ordinary habits of conversation.
Its central concern is not simply whether an AI statement is false. Its
concern is the gap between the meaning presented, the meaning inferred,
and the meaning withheld.

The ethical task is not to eliminate every shadow. Language cannot expose
everything at once, and simplification is often necessary. The task is
to identify shadows that matter: shadows over uncertainty, agency,
accountability, privacy, evidence, or the nature of the system itself.

In this sense, ombraism names a characteristic condition of contemporary
technological language. Artificial intelligence does not merely produce
answers. It produces scenes of intelligibility. Within those scenes,
some meanings are illuminated, while others recede.

The question is therefore not only: “What did the machine say?”
It is also: “What did the machine’s language prevent us from seeing?”

Selected Bibliography

1. Bender, E. M., Gebru, T., McMillan-Major, A., and Shmitchell, S.
   “On the Dangers of Stochastic Parrots: Can Language Models Be Too
   Big?” Proceedings of the 2021 ACM Conference on Fairness,
   Accountability, and Transparency, 2021.
2. Epley, N., Waytz, A., and Cacioppo, J. T. “On Seeing Human: A
   Three-Factor Theory of Anthropomorphism.”
   Psychological Review, 2007.
3. Goffman, E. Frame Analysis: An Essay on the Organization of
   Experience. Harvard University Press, 1974.
4. Grice, H. P. “Logic and Conversation.” In
   Syntax and Semantics, Volume 3: Speech Acts, edited by
   P. Cole and J. L. Morgan. Academic Press, 1975.
5. Lakoff, G., and Johnson, M.
   Metaphors We Live By. University of Chicago Press, 1980.
6. Mittelstadt, B. “Principles Alone Cannot Guarantee Ethical AI.”
   Nature Machine Intelligence, 2019.
7. Searle, J. R. “Minds, Brains, and Programs.”
   Behavioral and Brain Sciences, 1980.
8. Suchman, L. Human–Machine Reconfigurations: Plans and Situated
   Actions. Cambridge University Press, 2007.
9. Weizenbaum, J. Computer Power and Human Reason: From Judgment to
   Calculation. W. H. Freeman, 1976.