GabyChatbots — Artificial intelligence is becoming an increasingly accessible way to get answers and advice. Large language models are being consulted even though they are known for racial and gender biases.
A new study provides evidence that political bias should be added to that list. Researchers warn it could influence society’s political orientation.
What did the research show?
Computer scientist David Rozado, a professor at the Otago Polytechnic Institute (New Zealand) and the study’s author, raised concerns about how chatbots shape our worldview and which information we trust. He said that artificial intelligence should be impartial and objective.
Professor Rozado used 11 standard political questionnaires to analyze the political orientation of 24 leading language models, including ChatGPT, Gemini, Claude, and Grok. He found that the political stance of these AI systems is far from neutral. Rozado told Science Alert that most of the models exhibited left-leaning views.
As we increasingly turn to AI bots for information, there is growing concern that the answers we receive may influence our thinking.
It’s not entirely clear how this bias seeps into the systems. Some suggest developers may introduce it intentionally. These models are trained on vast amounts of online text, so a prevalence of “left-leaning” material in the training data could tilt their outputs.
Professor Rozado suggested that ChatGPT’s dominance in training data for other models could also be a factor, since earlier analyses have shown that this chatbot tends to adopt left-leaning positions.
LLM-based bots use probabilities to choose which word should follow another in their responses. That means they can produce errors even before various types of bias are considered.
Tech giants like Google, Microsoft, Apple, and Meta are actively pushing AI-based chatbots, trying to expand their influence into as many areas of life as possible. So perhaps it is time to reassess where these technologies help and where they do not.
Rozado said it’s crucial to critically examine this issue and eliminate potential political bias inherent in language models, “to ensure a balanced, fair, and accurate presentation of information in their responses to user queries.”
The results of the study were published in the journal PLOS ONE.
