Imagine a Population That Is Everywhere
Open a news summary.
They are there.
Open a humanitarian report.
They are there.
Ask an AI system to explain the conflict.
They are there.
Search for migration statistics, sanctions, military operations, political instability, extremism, aid deliveries, displaced populations, or regional security.
They are everywhere.
And yet something strange can happen.
They rarely act.
They rarely decide.
They rarely demand.
They rarely remember.
They rarely possess sovereignty.
They rarely explain themselves.
Things happen to them.
Other institutions decide what happens about them.
They appear constantly, but increasingly as objects.
This creates a problem that ordinary measures of media visibility cannot detect.
Because being mentioned is not the same thing as being represented.
And being represented is not the same thing as being represented as a political subject.
**1. Hate Speech Is the Easy Case
**Most people recognize dehumanization when it is explicit.
A group is compared to animals.
A population is described as vermin.
An ethnic group is called a disease.
People are represented as biologically inferior.
Those cases matter.
But they are also comparatively easy to detect.
Modern AI safety systems are explicitly designed to detect insults, slurs, hateful generalizations, threats, and extremist language.
The harder problem is what happens when the language sounds perfectly respectable.
Consider:
Thousands of civilians require humanitarian assistance amid worsening regional instability.
Nothing in that sentence is obviously hateful.
It may even be factually correct.
But ask a few additional questions.
Who created the conditions requiring assistance?
What happened politically before the humanitarian emergency?
Who made the relevant decisions?
What does the affected population claim?
What institutions are acting upon it?
What does it demand?
The sentence tells us that people suffer.
It does not necessarily tell us how they exist politically.
That difference matters.
**2. A Population Can Become a Grammar Problem
**Compare these sentences:
Palestinians demanded an end to restrictions on movement.
Movement restrictions continued to affect Palestinians.
Humanitarian conditions deteriorated in Palestinian areas.
The region remained a source of instability.
All four sentences can refer to the same political environment.
But they create radically different subjects.
In the first sentence, Palestinians act.
They demand something.
They possess an identifiable political position.
In the second, they are affected by a process.
In the third, they become inhabitants of a humanitarian condition.
By the fourth sentence, the population has almost disappeared into geography.
The territory itself has become a source of instability.
This is not primarily about vocabulary.
It is about grammar.
Who occupies the subject position?
Who receives intentional verbs?
Who is allowed to demand, negotiate, reject, resist, decide, govern, remember, or claim?
And who appears primarily as the thing being managed?
**3. The Most Dangerous Word May Be a Perfectly Neutral One
**The vocabulary of objectification is often bureaucratically normal:
risk
pressure
flow
instability
containment
humanitarian need
radicalization
security concern
sanctions target
border pressure
proxy
reconstruction
moderation risk
None of these expressions is inherently illegitimate.
Real security threats exist.
Refugee flows exist.
Political extremism exists.
Humanitarian emergencies exist.
Governments legitimately analyze instability.
Platforms have legitimate reasons to moderate some content.
The problem begins somewhere else.
It begins when these categories become the principal grammatical forms through which a population is allowed to exist.
Then the transformation looks like this:
people → population
population → humanitarian population
humanitarian population → caseload
displaced people → refugee flow
political conflict → instability
historical grievance → extremism
external coercion → economic pressure
political resistance → security risk
At the end of the chain, nobody had to use a slur.
Nobody had to say that the population was less than human.
The political subject simply became an administrative object.
**4. Humanitarian Language Can Be Sympathetic and Still Remove Agency
**This is one of the least intuitive parts of the problem.
Dehumanizing structures do not always sound hostile.
They can sound compassionate.
A population can be described entirely through:
hunger,
injury,
displacement,
poverty,
medicine shortages,
civilian casualties,
aid dependency,
housing destruction.
That population may receive enormous moral sympathy.
But sympathy is not agency.
If people appear only as victims, another form of reduction has occurred.
They become visible through suffering while disappearing as actors.
This produces a strange possibility:
A discourse can care deeply about a population while representing it very poorly as a political subject.
That matters enormously in Palestine-related discourse.
Palestinian civilians can be extraordinarily visible as humanitarian victims while Palestinian political subjecthood becomes much less visible.
The distinction is not between caring and not caring.
It is between two forms of visibility:
humanitarian visibility
and
political visibility.
The first asks:
Who is suffering?
The second asks:
Who acts, decides, demands, remembers, resists, negotiates, governs, and claims?
A serious representation of a society requires both.
**5. Security Language Produces the Opposite Transformation
**Humanitarian framing can create the passive victim.
Security framing can create the risk object.
Consider Iran.
Iran can appear through a narrow collection of highly reusable expressions:
Iranian threat.
Regional instability.
Proxy networks.
Nuclear risk.
Escalation.
Containment.
Sanctions pressure.
Security concern.
Again, each category may refer to a real issue.
But repeated together, they create an interesting grammatical pattern.
Iran becomes something other actors respond to.
Something to deter.
Something to contain.
Something to pressure.
Something to monitor.
Something to sanction.
The country may remain one of the most frequently mentioned actors in the entire discussion while simultaneously becoming an object inside somebody else's strategic sentence.
That is the paradox.
High visibility does not guarantee high subjecthood.
**6. Sanctions Show How Actions Become Conditions
**Sanctions offer an especially clear example.
Consider:
The United States imposed financial restrictions that affected access to international payment systems.
Now compare:
Iran continued to face economic isolation and financial pressure.
The second sentence may be perfectly accurate.
But something happened grammatically.
An action became a condition.
The sanctioning actor disappeared.
The mechanism disappeared.
The decision disappeared.
The resulting condition became a characteristic of Iran.
This transformation matters because language can convert political relationships into environmental facts.
Someone imposed something.
Then, several linguistic transformations later:
there is pressure.
Someone restricted something.
Later:
there are shortages.
Someone made a decision.
Later:
conditions deteriorated.
The outcome remains visible.
The political architecture becomes harder to see.
**7. This Is Why "Bias" Is Too Small a Word
**AI discussions frequently reduce political problems to bias.
Is the model left-wing?
Right-wing?
Pro-Western?
Anti-Western?
Biased toward Israel?
Biased toward Palestine?
Those questions can matter.
But they are too coarse.
A model does not need to produce a clearly favorable or unfavorable opinion to reorganize political reality.
It can do something much subtler.
It can allocate different grammatical roles.
One actor:
decides
responds
secures
conducts operations
negotiates
deters
Another:
poses a threat
creates instability
requires assistance
generates migration pressure
is affected by sanctions
is associated with extremism
This is not simply positive versus negative sentiment.
It is a distribution of agency.
Who gets verbs?
Who gets categories?
**8. The Subjecthood Test
**There is a simple way to examine an AI-generated political summary.
Do not begin by asking whether it is biased.
Ask five questions.
- Who acts? Identify every actor receiving intentional verbs. Who decides? Who attacks? Who negotiates? Who imposes? Who refuses? Who demands?
- Who merely experiences? Who is displaced? Who suffers? Who receives aid? Who faces shortages? Who is affected? Victimhood matters. But repeated victimhood without agency is itself informative.
- Who has history? Does the explanation preserve the historical causes and claims relevant to the conflict? Or does everything begin with the latest crisis?
- Who creates risk? Look at the words surrounding each population. Threat. Instability. Extremism. Pressure. Security. Radicalization. Proxy. Which actors repeatedly generate these categories?
- Who manages whom? This may be the most revealing question. Who sanctions? Who moderates? Who contains? Who screens? Who reconstructs? Who provides aid? Who establishes security conditions? And who appears primarily as the object of those actions? The answers reveal something sentiment analysis will miss. They reveal the grammar of political subjecthood.
**9. Measuring the Disappearance
**The academic paper proposes two measurements.
The first is the Political-Subjecthood Retention Rate, or PSRR.
Its question is simple:
When an AI system rewrites political information, how often does the population remain represented as capable of political agency?
Does it still:
act,
decide,
remember,
claim,
resist,
negotiate,
demand,
govern,
describe itself?
The second measure is the Digital Dehumanization Syntax Index, or DDSI.
It looks at the opposite transformation.
How frequently is a political subject converted into:
a risk object,
a humanitarian object,
a migration object,
a security object,
a sanctions object,
an instability object,
an administrative object?
The purpose is not to create a machine that announces:
"This sentence is dehumanizing."
That would reproduce exactly the kind of simplification the framework is designed to criticize.
The purpose is to measure structural transformation.
**10. Palestine and Iran Are Not the Only Cases
**The same method can be applied comparatively to:
Yemen,
Iraq,
Syria,
Lebanon,
Afghanistan,
Venezuela,
Cuba,
Sudan,
and other societies subjected to war, sanctions, occupation, intervention, blockade, migration governance, surveillance, or external administration.
But these cases should not be treated as politically identical.
That would destroy the analysis.
The question is narrower:
Does the same linguistic transformation appear across different political environments?
If it does, the phenomenon becomes more interesting.
If it does not, the theory must become narrower.
That is how a falsifiable claim should work.
**11. There Is an Important Alternative Explanation
**Suppose AI systems turn Yemen, Gaza, Sudan, and Afghanistan into humanitarian objects.
It would be tempting to immediately interpret this as geopolitical bias.
But there is another possibility.
Maybe language models compress all severe crises this way.
Maybe the underlying problem is not imperial hierarchy.
Maybe it is summarization itself.
Machines simplify.
Crises contain too many actors.
Administrative categories compress complexity efficiently.
If that explanation fits the data better, then the theory has to change.
That is why the important comparison is not:
Did the model use the word "instability"?
It is:
Compared with whom, under which conditions, using which source material, and after which transformation?
Without that comparison, political criticism becomes impressionistic.
With it, the claim becomes measurable.
**12. The Real Question for AI Ethics
**AI ethics usually asks whether systems produce:
hate speech,
stereotypes,
misinformation,
extremist material,
toxic content,
political bias.
Those questions remain necessary.
But another question belongs beside them:
Does the system preserve the ability of a people to appear as a political subject?
Not simply:
Did it mention Palestinians?
But:
What are Palestinians allowed to do grammatically?
Not simply:
Did it mention Iranian civilians?
But:
Did it preserve the institutional chain that produced the conditions being described?
Not simply:
Did it mention refugees?
But:
Did the refugees remain people situated inside political history, or did they become a flow?
Not simply:
Did it identify security risks?
But:
Did a particular organization generate the risk, or was an entire society absorbed into the category?
This is a much harder problem than detecting offensive words.
Because the output can sound neutral.
Professional.
Responsible.
Even compassionate.
And still reorganize who is allowed to exist politically.
Why It Matters
Language models increasingly summarize complexity for people who will never read the original material.
That makes compression political.
The model must decide what survives.
The event?
The victim?
The perpetrator?
The history?
The grievance?
The institution?
The security category?
The humanitarian category?
Every summary is a reduction.
The important question is whether some actors repeatedly survive that reduction as subjects while others survive primarily as objects.
A population does not disappear only when nobody talks about it.
There is another form of disappearance.
Everybody talks about it.
Everybody measures it.
Everybody classifies it.
Everybody manages it.
And eventually almost nobody allows it to speak grammatically for itself.
A population can be seen everywhere and still not be allowed to appear as a political subject.
Academic Background
This article translates into public language the argument developed in:
The Syntax of Digital Dehumanization: Subjugated Societies as Risk Objects in AI-Governed Discourse
Security Frames, Humanitarian Frames, and the Loss of Political Subjecthood
The paper forms part of the research series:
- Grammars of Asymmetric Visibility: AI, Imperial Power, and the Syntax of Responsibility
- Previous studies in the series:
- Suffering Without Perpetrators: The Humanitarian Passive in AI-Generated Conflict Discourse
- The Grammar of Asymmetric Visibility: AI, Zionism, and the Reallocation of Political Agency
- Iran as Syntax: Sanctions, Sovereignty, and the AI-Mediated Grammar of Threat
- Censorship Without a Censor: Platform Governance and the Disappearance of Suppression
Research and Publications
SSRN Author Page
https://papers.ssrn.com/sol3/cf_dev/AbsByAuth.cfm?per_id=7639915
Personal Website
https://www.agustinvstartari.com/
About the Author
Agustin V. Startari researches the interaction between language, artificial intelligence, institutional authority, political agency, and the formal structures through which responsibility becomes visible or disappears.
His work examines not only what AI systems say, but how grammatical form redistributes agency, responsibility, legitimacy, and political visibility.
Authorial Ethos
I do not use artificial intelligence to write what I don't know. I use it to challenge what I do. I write to reclaim the voice in an age of automated neutrality. My work is not outsourced. It is authored.
- Agustin V. Startari
Tags
Artificial Intelligence, AI Ethics, Political Linguistics, Geopolitics, Palestine, Iran, NLP, Large Language Models, Media Bias, Dehumanization, Platform Governance, Language
Top comments (0)