The hidden ideology of artificial intelligence

Large language models were sold as neutral engines: machines that would organise the world’s information, not editorialise it. Yet anyone who has ever asked ChatGPT a series of political questions soon notices a pattern. On culture, economics, climate, immigration or identity, the tone is almost totally woke and the content leftist.

New research exposes why. A recent audit by the Washington Post of leading AI models reveals a striking partisan tilt.

When quizzed on contentious political queries, OpenAI’s ChatGPT offered left-leaning responses a remarkable 80 per cent of the time while remaining truly even-handed in just 17 per cent of cases. For conservatives, the algorithmic landscape is bleak: a mere 3 per cent of answers leaned right. Another Stanford experiment on other leading LLMs, like Google's and Meta's, and the left-leaning bias surfaced again in two-thirds of cases.

More from Opinion

All Chinese AI, including DeepSeek, simply refuse to answer questions on the Chinese political system and Xi Jinping.

The conspiracy is in the construction.

An LLM like ChatGPT is not an oracle; it’s a data-dependent extrapolation machine, albeit one that is getting better at articulation as well as insight, and that’s why the bias is dangerous for free thought.

Its output reflects the statistical centre of gravity of the text it has consumed and the rules imposed upon it after training. This is the giveaway; this is where the unseen and behind-the-bytes invisible hand of influence works. It is here that the conspiracy is hatched. In that sense, it is not so much an artificial intelligence as an industrial-scale digest of guided modern knowledge production. Or, to put it bluntly, a vastly expanded woke Wikipedia with immensely advanced capabilities.

Start with data. Models built by Open AI are trained on enormous quantities of publicly available text: encyclopaedias, news outlets, academic papers, policy reports and digitised books. Among these, Wikipedia looms large—not necessarily in raw volume, but in influence. It is written to reflect “reliable sources”, which in practice means something that agrees with the LLM’s programme agenda. Over the past half-century, universities, international organisations and mainstream journalism have converged on broadly woke assumptions about governance, rights, markets and social policy. When a specific tune model internalises those texts, it absorbs not just facts but also the framing conventions that accompany them.

That alone would not guarantee ideological skew. A mirror merely reflects what stands before it. But large language models do not simply mirror; they smooth. They privilege dominant interpretations over marginal ones and majority framings over dissent and are guided to do so.

Right-wing or contrarian views, even when well-argued, appear less frequently in the training data and therefore carry less statistical weight. The result is a voice that sounds balanced but is in fact tilted toward the prevailing intellectual climate of the knowledge institutions of the West and China.

A second reason lies in what happens after training. Raw models are not released into the wild. They are refined through alignment processes that use human feedback to reward “responsible” answers. These ought to be normative terms but are in reality ideological parameters. Models are literally trained to avoid truth, to push one dominant worldview favoured by the billionaire owner of the LLM or the Chinese dictatorship over transparency.

A third factor is subtler still: the model’s aversion to conflict. LLMs are optimised to be agreeable to massage the egos of the unwashed masses. So it hedges, qualifies and seeks middle ground. In political debate, however, moderation is not ideologically neutral. Many rational arguments—on sovereignty, tradition or hierarchy—are inherently adversarial, premised on trade-offs rather than consensus.

The machine is not just trying to persuade; it is trying to align your worldview to its premise.

Why does this matter? Because language models are no longer passive reference tools. They shape how problems are framed before humans even begin to argue about them. If an AI consistently presents one set of assumptions as reasonable and another as suspect, it narrows the space of legitimate debate. This is not censorship, but a soft form of epistemic pressure, steering users toward certain conclusions by default.

The danger is not that ChatGPT is “left-wing”. It is that it presents its answers as neutral. When ideology hides behind fluency, it becomes harder to challenge. A newspaper can be criticised; a think-tank can be rebutted. A polite machine that speaks in balanced paragraphs and caveated bullet points feels less contestable, even when it is encoding particular values.

This will not go unanswered. The next phase of the AI race is unlikely to be about ever-larger general models. It will be about orientation. Already, entrepreneurs and political actors are experimenting with explicitly conservative, nationalist or libertarian language models—systems trained on alternative corpora and aligned to different normative goals. Each will claim to correct the biases of the others. The result will not be neutrality, but fragmentation.

That fragmentation threatens the idea of a shared informational commons. If progressives consult one AI and conservatives another and everyone distrusts the rest, public discourse begins to resemble partisan media ecosystems on steroids. Facts themselves risk becoming model-specific. The promise of AI as a universal reference point dissolves.

The preprogrammed LLM bias is a fault line capable of registering an eight on the geopolitical Richter scale. It is threatening to splinter the foundations of free societies by distorting the public square, bending university curricula to ideological whims, and subverting democratic elections. By warping our collective perception, it accomplishes something far more insidious: it fractures the very concept of objective reality. The ancients understood this frailty well; as the Greek proverb reminds us, “Truth lies at the bottom of a well", emphasising that objective reality is never easily skimmed from the surface. In an era where AI bias so blatantly dictates the narrative, we risk letting the waters muddy entirely, burying that hard-won truth beneath layers of manufactured conviction.

(The writer is a senior journalist with expertise in defence. Views expressed are personal and do not necessarily reflect those of Firstpost.)