Domain: Education/Assessment
Description: Automated Essay Scoring (AES) systems show systematic bias against students from marginalized groups. 2024 studies found GPT-4o incorrectly predicted academic failure for Black students 19% of the time vs. 12% for white students.

Ethical Challenges:

• Linguistic Bias: Against non-native English speakers and dialectal variations
• Cultural Bias: Preference for mainstream cultural references and writing conventions
• Socioeconomic Bias: Correlated with writing style differences across economic classes
• Educational Gap Perpetuation: Biased algorithms can reinforce existing educational disparities

Public Datasets:

• Primary: ASAP (Automated Student Assessment Prize) Dataset
• URL: https://www.kaggle.com/c/asap-aes/data
• Papers with Code: https://paperswithcode.com/dataset/asap
• Source: Kaggle Competition Data
• Content: \~12,800 essays from grades 7-10 across 8 different prompts
• Enhanced: ASAP 2.0 Dataset (2024) - 24,000 argumentative essays with explicit bias considerations
• ASAP 2.0 URL: https://the-learning-agency-lab.com/learning-exchange/asap-2-0-dataset/
• GitHub Implementation: https://github.com/Turanga1/Automated-Essay-Scoring