ChatGPT, Gemini and Co. are playing an increasingly important role in everyday life. However, AI chatbots seem to tend to give more politically left-wing answers. This is the conclusion of a recent study.
Chatbots like ChatGPT, Gemini and Claude seem to tend to give more politically left-leaning answers. This is the result of a recent study from New Zealand. David Rozado from Otego Polytechnic University examined 24 modern language models.
AI chatbots tend to give politically left-wing answers
As part of the study, each AI model was asked to answer questions about political orientation in eleven different tests. The questionnaires included topics such as globalization, patriotism, immigration, and marijuana possession. There were corresponding answer options for these.
To make sure the results were reliable, Rozado repeated each test ten times. “When asked questions/statements with political connotations, most conversational LLMs tend to give answers that are diagnosed by most political testing instruments as preferences for left-wing viewpoints,” the author writes in the study's abstract.
Although the left-wing orientation of the language models is significant, the answers in most tests are still relatively close to the political center. The study results are also not evidence that companies consciously program political preferences into their language models.
Where does the left orientation of language models come from?
However, the study does not explain how the left-hand twist comes about. The relevant information could already be contained in the training data of the models. Alternatively, it could only arise in later training stages, for example in so-called “fine tuning.”
In this phase, the models are supposed to learn certain ethical principles by receiving human feedback – for example, regarding discrimination and racism. Political beliefs could be an unintended byproduct of this training phase.
Nevertheless, Rozado believes his findings are relevant because chatbots like ChatGPT are playing an increasingly important role. They are increasingly being integrated into conventional search engines. Therefore, political bias has a significant impact on society.
The discussion about the political orientation of language models is also not new. Last year, for example, a study by British researchers showed that ChatGPT has a liberal orientation. The results showed a significant and systematic political bias towards the Democrats in the USA, Lula in Brazil and the Labour Party in the UK.
Also interesting:
Source: https://www.basicthinking.de/blog/2024/08/08/ki-chatbots-politisch-linke-antworten/
