The research shows that models like ChatGPT express the ideological preferences of their creators, displaying less affinity for supranational organizations and welfare policies.
Marteen Buyl states, 'Our results confirm that there is no such thing as a large language model that is completely politically neutral,' highlighting the inherent biases of AI.
Jose Hernandez Orallo elaborates that various models reflect different ideologies when confronted with identical prompts, showcasing alignment with creators' values rather than neutrality.
The researchers employed a unique method by asking AI models to describe 4,000 famous individuals, thus revealing underlying ideological biases in responses.
Collection
[
|
...
]