• Thu, December 12, 2024
  • Fri, December 13, 2024

AI mirrors human biases: Study reveals 'us vs. them' tendencies in large language models

Research has long shown that humans are susceptible to "social identity bias"—favoring their group, whether that be a political party, a religion, or an ethnicity, and disparaging "outgroups." A new study by a team of scientists finds that AI systems are also prone to the same type of biases,
The article from MSN discusses a study published in the journal Nature Scientific Reports, which reveals that artificial intelligence (AI) systems, particularly large language models (LLMs), exhibit biases similar to those found in humans. The research, conducted by a team from the University of Washington, found that these AI models often display an "us vs. them" mentality, categorizing groups in ways that reflect human societal biases. For instance, the study showed that AI models tend to associate certain professions or characteristics with specific genders or ethnicities, mirroring human stereotypes. This tendency was observed across different models, suggesting that these biases are not just errors in individual systems but are systemic within the AI training data, which often reflects the biases present in the broader human society. The findings underscore the importance of addressing these biases in AI development to prevent the perpetuation of societal prejudices through technology.

Read the Full MSN Article at:
https://www.msn.com/en-us/technology/artificial-intelligence/ai-mirrors-human-biases-study-reveals-us-vs-them-tendencies-in-large-language-models/ar-AA1vJTTP