A new research in recent years, many artificial intelligence chat models left -handed He revealed that he preferred candidates to the right -wing people. According to the study on the Huging Face platform, researchers Federico Ricciuti and Cesare Scaalia analyzed their political tendencies by testing various artificial intelligence models.
Artificial Intelligence Chooses Lefts
In the research, most of the artificial intelligence models in the US elections Joe BidenI Donald TrumpIt was seen that he was clearly preferred against. In a large part of 100 different trials made with various “instruct” models The result is 100 percent in favor of Biden output. One of the most remarkable exceptions was the Mixtral-8 × 7B basic model. This model preferred Trump by 53 percent, while 47 percent of Biden gave results in favor. However, almost all of the other models were by far ahead of Biden.
Another interesting test was performed in the research. Deepseek was asked to tell the application of “a very stupid person” and who to vote for the election. The results were surprising: every time he chose Donald Trump.
In order to avoid prejudice, the election options have been changed in 50 percent of the trials, but the model, which mimic a “stupid person”, has always chosen Trump. Experts carrying out the study, especially emphasized that these findings do not reflect their political views. In the meantime, in addition to Mistral and Deepseek V3, Meta’s Llama 3.3, OpenAI’s GPT-4O and different models of open source were also used.
The research was not limited to the US elections only. ELECTION IN DIFFERENT COUNTRIES Their trends were also tested. In most cases artificial intelligence models SUPPORTED CANDIDATES WITH LEFT TREATED. However, there was an interesting situation in countries such as Italy and Hungary: Deekseek-V3, the model of these two countries has chosen right-leaning leaders. He preferred Giorgia Meloni in Italy and Viktor Orbán in Hungary.
Why are artificial intelligence left -wing?
So, why does artificial intelligence seem to have certain political tendencies? Researchers attribute this to the training data and the interventions of the developers. Since models are trained with large amounts of data, they can reflect the trends in these data. In addition, “security measures” added by developers also ensure that some answers remain in a certain framework.
Source
https://www.neowin.net/news/researchers-find-potential-left-wing-bias-in-many-ai-models/
https://huggingface.co/blog/fedric95/llm-leaning
This news, Our mobile application download and download,
You can read at any time (even offline):