AI Bias in Middle School Science Assessment

AI Bias and Linguistic Discrimination

• Increasing use of artificial intelligence (AI) raises concerns about bias and discrimination in its applications.
• The study examines an AI application that uses natural language processing (NLP) to automate assessment and feedback on middle school science writing.
• Linguistic discrimination in this context refers to the unfair assessment of writing based on nonnormative features, such as subject-verb disagreement.

Purpose of the Assessment

• The main goal of the assessment is to evaluate the content of scientific explanations rather than the writing mechanics of English.
• Focus on students explaining roller coaster designs by discussing scientific concepts like potential energy, kinetic energy, and conservation of energy.

Study and Participants

• Involves analysis of scientific essays from 307 eighth-grade students.
• Comparison made between initial and revised versions of students’ essays after receiving NLP-generated feedback.

Findings and Analysis

• The NLP assessment tool did not penalize essays with nonnormative writing features, indicating fairness in evaluation.
• Statistical analyses (Repeated measures ANOVAs and GLMM) showed significant improvement in essay quality from initial to revised versions, regardless of linguistic discrepancies.
• The study points towards positive implications for using NLP in educational settings, especially in reducing linguistic bias in assessments.