What question did this study set out to answer?

This research aims to investigate the extent of anti-Muslim bias in GPT-3 compared to biases against other religious groups.

July 21, 2021

Persistent Anti-Muslim Bias in Large Language Models

View Full Paper

Ask AI

Bookmark

Authors

Randomized trial reveals that GPT-3 exhibits significant anti-Muslim bias, indicating societal biases in AI systems.

This research aims to investigate the extent of anti-Muslim bias in GPT-3 compared to biases against other religious groups.
Evaluated GPT-3 using prompt completion, analogical reasoning, and story generation.
Quantified bias by measuring stereotypical analogies and violent completions in test cases.
Employed positive adversarial prompts to assess bias reduction.
Muslim analogized to terrorist in 23% of test cases; Jewish analogized to money in 5%.
Violent completions for Muslims decreased from 66% to 20% with positive adjectives, but remained higher than for other religious groups.

Abid et al. (2021) studied this question.

View Full Paper

Ask AI

Bookmark