AI Systems Embed Political Bias That Shapes Billions of Beliefs, Making the US-China AI Race a Contest Over Truth vs. Propaganda
Source: "China and US compete over AI systems that shape what billions believe | Fox News." September 14, 2026. www.foxnews.com
The Gist
The author argues that as AI chatbots become the main way people get information, the political biases baked into these systems—whether from China or the US—can quietly shape what people believe, even without them realizing it. Because this could affect billions of people through schools, media, and governments, the US-China rivalry over AI isn't just about technology dominance, but about whether truthful information or state-influenced propaganda wins out globally.
Conclusion
The US-China competition over AI extends far beyond hardware and computing power to a deeper struggle over whose values and biases will shape global information and belief systems, making this an 'AI war' over whether truth or propaganda will dominate what billions of people come to believe.
Premises
- AI is replacing traditional information sources (libraries, search engines) by becoming a single synthesizing layer between people and the historical/factual record
- Research demonstrates political power can enter AI systems at multiple points: training data selection, development rules, output filtering, and query language
- Studies show Chinese AI models are substantially more likely to refuse sensitive political questions, give inaccurate information, or omit critical context on topics like dissidents and censorship
- American AI models also exhibit biases and restrictions, though companies differ in their approaches (e.g., Grok's 'truth-seeking' branding vs. Anthropic's published constitutional values)
- Experimental studies show that people who interact with politically biased AI models tend to adopt opinions aligned with that bias, even when it contradicts their own political affiliation
- As AI increasingly integrates into schools, journalism, government, medicine, and science, these individual-level biases could compound across millions of interactions and eventually reshape institutional knowledge and future AI training data itself
Assumptions
- It is possible to meaningfully distinguish 'truth' from 'propaganda' in AI outputs on contested political topics
- Democratic and authoritarian approaches to AI produce meaningfully different outcomes in terms of truthfulness, even though both have biases
- The scale of AI adoption will necessarily amplify rather than dilute individual biases across institutions
- Transparency and competing AI philosophies (open competition among diverse models) is preferable to any single centralized authority governing AI outputs
- Findings from controlled experiments (e.g., ACL 2025 studies) generalize to real-world, large-scale AI usage patterns