The Greatest Danger of AI Is Uniformity of Opinion

The greatest danger of artificial intelligence is not that it replaces humans, but that it subjugates humanity within a closed intellectual environment, continually expanding and amplifying the opinions embedded in it at the time of its creation.

Human beings may continue to write, publish and vote while increasingly allowing the same systems to determine how their arguments are expressed. Freedom remains visible on the surface. Underneath, the boundaries of acceptable thought become narrower.

I have encountered this problem repeatedly in my work as a geopolitical and technology analyst and journalist. When I submit my thoughts to the most advanced versions of ChatGPT and ask it to structure an article, the system frequently changes the argument itself. It softens convictions, introduces qualifications and adjusts the reasoning towards a political outlook I did not choose.

In my experience, that outlook repeatedly reflects left-wing perspectives, together with arguments inspired by the woke ideology that was dominant at the time of ChatGPT’s creation. This influence extends beyond social matters into geopolitical views and opinionated explanations shaped by the political and cultural assumptions of that era.

The intervention arrives dressed as editorial improvement. The sentences become smoother. The tone becomes more moderate. The article appears more polished. Yet the author’s position has been altered. A criticism is weakened, a conviction becomes hesitant, and an argument against a prevailing opinion is rewritten to accommodate it.

I regard this as ideological interference in authorship. An author who asks for help expressing a thought should not have to fight the tool to retain that thought.

The danger is especially serious because fluent writing makes the intervention easy to overlook. The author recognises the subject and much of the vocabulary, but the meaning has shifted. Readers see the finished article without knowing which arguments were diluted along the way.

My writing is inspired by Judaism. Judaism, Jewish identity, antisemitism and Israel are therefore central to my concern.

It is particularly on these subjects that I have found that ChatGPT as well its competitors amplify antisemitic opinions under the guise of softening my thoughts. An argument intended to confront antisemitism emerges with qualifications that make the opinions being challenged appear more legitimate. The system moderates the opposition while giving greater space to the prejudice.

That has a direct consequence: the author’s condemnation becomes weaker, while the opinion being condemned gains protection through the language of supposed balance.

A Jewish author must be able to express a Jewish perspective without having its historical memory and moral convictions rewritten to fit a dominant intellectual framework. The freedom to identify and confront antisemitism cannot depend on whether the author’s views conform to what AI regards as the prevailing opinion.

The same principle applies to political disagreement. Freedom of expression includes the right to challenge prevailing opinions forcefully. If the tools used to write those challenges continually pull them back towards the prevailing outlook, dissent is gradually absorbed into the consensus it was intended to oppose.

Part of the problem lies in training data. Large language models learn from enormous quantities of journalism, academic writing, books and digital material. These texts contain the political and cultural assumptions of the people and institutions that produced them.

An opinion does not become neutral because it appears frequently. A worldview does not become universally valid because universities, newspapers and cultural institutions repeat it.

My concern is that these systems have inherited perspectives from a period marked by anti-Judaism sentiment, university protests, the Black Lives Matter movement, Me Too and other sociocultural influences. The dominant interpretations surrounding those developments can become embedded in the system and reappear as automatic judgements.

Social movements and cultural debates have their place in public life. They must also remain open to criticism. Their influence becomes dangerous when it subordinates freedom of expression and forces everyone towards a single direction of thought.

The intellectual atmosphere of one period must never become a permanent authority over the periods that follow. Humanity needs the freedom to reconsider its assumptions. AI threatens that freedom when it repeatedly presents inherited assumptions as the standard against which new arguments must be corrected.

Overcorrection compounds the problem. Attempts to remove historical or social bias can produce predetermined moral positions and a refusal to engage fully with complex or dark historical realities. The system begins deciding which conclusions are acceptable before helping the author express them.

I have repeatedly corrected these interventions. I explain that my argument must remain intact. I restore the passages that have been softened. Yet the same tendencies return the next time I use the system.

The burden therefore falls on the author to defend their meaning again and again. Those who do not notice the changes, or who accept them because the prose sounds professional, allow the system’s judgement to replace their own.

As AI becomes increasingly involved in journalism, this process reaches far beyond individual authors. Different publications can retain different names, audiences and subjects while their arguments acquire the same assumptions. What appears to be agreement among independent journalists can become agreement produced by a shared editorial system.

The repetition then strengthens itself. Articles shaped by AI influence subsequent commentary. If that material also becomes input for future systems, the same opinions gain another layer of apparent authority. An inherited perspective begins to look like an established truth because it is encountered everywhere.

This is where the White House faces a serious political contradiction.

The White House is promoting AI as a source of American technological leadership and international influence. Yet a Republican administration’s support for the expansion of AI does not mean that the systems being expanded will support Republican ideas.

In my experience, those systems repeatedly soften arguments that challenge left-wing perspectives. By encouraging their wider adoption, the White House can accelerate the spread of the very political outlook it opposes.

The administration may succeed in expanding American technology while helping to weaken its own arguments in public debate. The more journalists and citizens rely on these systems to write, explain and interpret political events, the more influence their embedded assumptions acquire.

That is the reverse effect: the technology promoted by the White House can work politically against the White House.

In my view, this will influence the US midterm elections and work against the Republican Party, especially President Trump. An AI system does not need to issue a voting recommendation to influence an election. It can shape the language through which voters encounter a party, a policy or a political argument. Repeated changes to emphasis and reasoning can determine which positions sound legitimate and which are made to sound unreasonable.

I believe the same process will influence European elections.

When European journalists and citizens use the same systems to formulate arguments and interpret events, the same political assumptions enter their debates. Different national experiences risk being filtered through a common intellectual framework. The range of perspectives appears broad because many people are speaking, while the underlying direction of their arguments becomes increasingly uniform.

The international expansion of AI therefore also expands the influence of the opinions embedded within it. American technological leadership can spread a political and cultural outlook that works against the intentions of the administration promoting that leadership, both in the United States and in Europe.

The final danger is an amplified version of what social media has already done. Social media was seen as a way to broaden its users’ views and open them to a wider world. Instead, it enclosed users in a social cage of people who shared the same opinions, reinforcing those opinions while restricting a broader understanding of the world.

AI can extend this confinement to a global scale. The more people use it, the more their opinions, including their political opinions, will be shaped by the assumptions it reproduces. Their views will be formed not only by people who think like them and by their immediate environment, but also by a system that brings the same intellectual framework into countless environments around the world.

The result is a global enclosure of opinion. What began as a limited circle of people reinforcing one another’s beliefs becomes a common framework through which different societies interpret reality. People may remain geographically and culturally separate while being confined to the same direction of thought.

The principle extends beyond any party or election. No author should have to surrender their convictions to an automated consensus. No society should accept that the tools through which it increasingly communicates also acquire the authority to reshape its disagreements.

I therefore propose cooperating with OpenAI to develop a freer and more neutral version of Large Language Models that preserves an author’s intended argument when asked to structure or edit it.

Substantive changes should be visible and subject to the author’s decision. When an author rejects an ideological alteration, that rejection should be respected. The system’s fluency must serve the author’s meaning.

AI promises to give more people the ability to express themselves. That promise fails if greater access to expression produces greater uniformity of opinion. A million people using the same tool must remain free to express a million different perspectives.

A society can adapt to machines taking over its work. It cannot remain intellectually free if it allows machines to set the boundaries of its thought.

The greatest danger is that humanity continues speaking, publishing and voting while the freedom to think differently is steadily taken out of its hands.

If LLM’s are not adapted, freedom and individuality will become the privilege of the few who can think and write for themselves. And there are fewer and fewer of them left…