Summary of Large Language Models Are Geographically Biased, by Rohin Manvi et al.
Large Language Models are Geographically Biased
by Rohin Manvi, Samar Khanna, Marshall Burke, David Lobell, Stefano Ermon
First submitted to arxiv on: 5 Feb 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); Computers and Society (cs.CY); Machine Learning (cs.LG)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary Large Language Models (LLMs) inherently carry biases from their training corpora, which can perpetuate societal harm. To evaluate and achieve fairness and accuracy, understanding LLMs’ knowledge of the world is crucial. Our approach focuses on geography, leveraging ground truth for aspects like culture, race, language, politics, and religion. We demonstrate accurate zero-shot geospatial predictions with strong correlation to ground truth (up to 0.89). However, we also show that LLMs exhibit common biases across various topics, including socioeconomic conditions (e.g., Africa) on sensitive subjective topics (attractiveness, morality, intelligence; up to 0.70 Spearman’s ρ). We introduce a bias score to quantify this and find significant variation across existing LLMs. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large Language Models carry biases from their training data, which can cause problems. To fix these issues, we need to understand what the models know about the world. One way to do this is by looking at how well they can predict geographic information. We found that some models are really good at making predictions (up to 0.89 correlation with the truth). However, we also discovered that many models have biases against certain places or groups of people based on sensitive topics like beauty, morality, and intelligence. |
Keywords
* Artificial intelligence * Zero shot
