New A.I. Chatbot Intentionally Expresses Biased Opinions to Spark Debate on Prejudice in Technology

Researchers in Berlin created an A.I. chatbot called OpinionGPT that intentionally expresses human biases based on region, demographics, gender, etc.

The technology is based on Meta's Llama 2 language model and can generate biased text responses.

The researchers asked it questions about art to demonstrate how biases emerge.

It gave questionable responses about topics like cultural appropriation, Duchamp, and Warhol.

The researchers say the goal is to promote discussion about bias in A.I., not promote any specific bias.