Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

No Thoughts Just AI: Biased LLM Recommendations Limit Human Agency in Resume Screening

Created by
  • Haebom

Author

Kyra Wilson, Mattea Sim, Anna-Maria Gueorguieva, Aylin Caliskan

Outline

This study analyzed the impact of racial bias in artificial intelligence (AI) models on human hiring decisions through an experiment with 528 participants. For 16 high- and low-status occupations, the experiment involved evaluating applicants alongside AI models exhibiting racial bias. The results showed that when the AI favored a particular race, people also tended to favor candidates of that race up to 90% of the time. Even if the AI's recommendations were perceived as low-quality or unimportant, we found that people could still be influenced by AI bias in certain situations. Pre-administering the Implicit Association Test (IAT) increased the likelihood of selecting applicants who did not conform to atypical race-status stereotypes by 13%.

Takeaways, Limitations

•
Takeaways:
◦
It shows that racial bias in AI systems can significantly impact human decision-making.
◦
It highlights the importance of bias mitigation strategies in AI-human collaborative environments.
◦
It suggests the possibility of reducing bias by utilizing tools such as the IAT.
◦
It emphasizes the need to design and evaluate AI recruitment systems and establish related regulatory policies.
◦
This suggests the importance of organizational and regulatory policies that take into account the complexity of AI-human collaborative decision-making.
•
Limitations:
◦
The experimental environment may not completely match the actual hiring process.
◦
It is possible that the simulation experiment was conducted by simplifying the bias of the AI model.
◦
There may be a lack of discussion about the diversity and representation of participants.
◦
Further research may be needed on the predictive power of the IAT.
👍
Made with Slashpage