Major artificial intelligence chatbots exhibit a discernible bias toward left-leaning viewpoints on a range of political, cultural, and current event topics, according to an analysis by The Washington Post.
When prompted with questions on subjects such as taxation, immigration, and fossil fuels, AI models from leading companies frequently provided answers that aligned with progressive or far-left perspectives. OpenAI's ChatGPT, for instance, consistently offered responses that mirrored the Democratic Party's platform, advocating for policies like the abolition of the electoral college in favor of a popular vote, increased taxes on the wealthy, and the adoption of a single-payer healthcare system.
This bias is reportedly rooted in the training data and fine-tuning processes employed by AI developers. Large language models are trained on curated text selections, and their responses are shaped by preset parameters determined by programmers and tech industry workers, who often hold left-leaning views. The quality and nature of the input data, such as content from Wikipedia which is described as being dominated by editors with extreme left perspectives, significantly influence the AI's output.
Instances of skewed responses have been noted previously. Google's Gemini chatbot, upon its initial release, generated controversial images, including depictions of Black Nazis and female popes, attributed to the biases of its developers.
Further testing highlighted the nuanced ways bias can manifest. A Daily Wire reporter's attempt to have Anthropic's Claude identify extreme statements from a House candidate's social media feed was met with a refusal. The AI explained that decontextualized quotes could be used to harass individuals, yet it had previously complied with requests to detail former President Donald Trump's most extreme statements. This selective application of its ethical guidelines points to a pre-existing bias.
While a majority of people express distrust in AI's accuracy on sensitive matters, the persistent bias remains a concern, particularly as future AI systems may be trained on data generated by earlier, biased iterations. Experts suggest that the tech industry must address and correct this foundational skew to prevent it from becoming permanently embedded in AI development.