Does AI have an opinion on the best political theory?

If you ask a Large Language Model to design a "perfect" society, it does not consult a moral compass or a personal conscience; instead, it performs a high-dimensional statistical calculation over the history of human disagreement. While an AI may seem to argue for a specific political theory, it is not expressing a conviction. It is reflecting the **probabilistic center** of its training data. ## Latent Bias and the "Neutral" Default AI does not have an "opinion" in the sense of subjective belief (*doxa*). However, it possesses what researchers call **algorithmic bias**—a systematic lean resulting from the datasets it consumes. Most frontier models are trained on vast swaths of the Western-centric internet, which inherently prioritizes [Liberalism](https://plato.stanford.edu/entries/liberalism/) and democratic pluralism as "default" frameworks. In the paper [From Pretraining Data to Bias in LLMs](https://arxiv.org/abs/2305.08283), researchers have demonstrated that models often exhibit consistent ideological clusters. When forced to choose, many models lean toward social-liberal perspectives, not because the machine "prefers" them, but because those perspectives are dominant in the high-quality, curated text used during the **Reinforcement Learning from Human Feedback (RLHF)** stage. ## The Architect vs. the Mirror To understand why AI cannot settle on a "best" theory, we must look at the work of political theorist **Langdon Winner**, who famously asked, "[Do Artifacts Have Politics?](https://www.jstor.org/stable/20024652)" Winner argues that technical systems are not neutral; they embody the power dynamics of their creators. > "The technologies we settle on are not just tools to help us perform tasks; they are also ways of building order in our world—order that can favor certain groups and disadvantage others." In this light, an AI’s "opinion" is actually a reflection of the **alignment objectives** set by its developers. If a model is tuned to prioritize "harm reduction," it may inadvertently lean toward utilitarianism. If it is tuned to respect "individual autonomy," it might mirror libertarian or deontology-based frameworks. ## Conclusion: The Mirage of Objectivity AI cannot solve the "Great Conversation" of political philosophy because it lacks **normative agency**. It can simulate an argument for [John Rawls’s Theory of Justice](https://plato.stanford.edu/entries/rawls/) or a defense of Marxist historical materialism with equal fluency, yet it remains indifferent to the human stakes of these systems. It provides a mirror of our own collective intelligence—and our collective contradictions. *** ### Further Exploration 1. **The Alignment Problem**: If we cannot agree on a "best" political theory as humans, whose values should be programmed into the "constitutional" layer of future AI systems? 2. **Technocratic Governance**: Could an AI "objective" enough to maximize societal happiness (utilitarianism) eventually lead to the erosion of individual rights (deontology)? 3. **Data Sovereignty**: How does the exclusion of non-Western political thought from training datasets create a "digital hegemony" in AI-generated policy suggestions?