Improving Ethical Outcomes with Machine-in-the-Loop: Broadening Human Understanding of Data Annotations
Project Overview
The document explores the role of generative AI in education, focusing on the creation of a machine-in-the-loop (MITL) pipeline designed to mitigate bias in natural language processing tasks. Central to this initiative is the LivedX platform, which empowers minoritized students to share and document their lived experiences as a pathway to earning micro-credentials in essential 21st-century skills. By emphasizing the importance of these personal narratives, the platform aims to enhance fairness and equity in educational outcomes. The iterative process of refining data annotations and model predictions enables a more accurate representation of students' unique contributions, fostering an inclusive educational environment. The findings indicate that integrating generative AI in this way not only improves the accuracy of educational tools but also promotes a deeper understanding of diverse student experiences, ultimately leading to a more equitable learning landscape.
Key Applications
LivedX platform for micro-credentialing
Context: Online platform for minoritized students to document experiences and earn micro-credentials
Implementation: A machine-in-the-loop model is used to refine predictions of micro-credentials based on student-submitted essays, iterating between human annotators and machine learning models.
Outcomes: Improved fairness in awarding micro-credentials and recognition of the unique skills of minoritized students.
Challenges: Initial model exacerbated biases against minoritized students, requiring ongoing adjustments to annotations and model training.
Implementation Barriers
Bias in Data Annotation
Annotations often reflect societal biases, leading to unfair outcomes in micro-credentialing.
Proposed Solutions: Iterative review of annotations based on model predictions, engaging annotators in reflecting on their biases.
Project Team
Ashis Kumer Biswas
Researcher
Geeta Verma
Researcher
Justin Otto Barber
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Ashis Kumer Biswas, Geeta Verma, Justin Otto Barber
Source Publication: View Original PaperLink opens in a new window
Project Contact: Dr. Jianhua Yang
LLM Model Version: gpt-4o-mini-2024-07-18
Analysis Provider: Openai