Dissertation Title: A Crown vs A Question Mark: Integrating Liberatory AI Principles into Responsible AI

Abstract:

The democratization of AI has expanded access to powerful technologies while simultaneously amplifying risks of misuse, particularly when AI development occurs without domain expertise or critical social awareness. In response, tech companies have adopted Responsible AI (RAI) frameworks. However, these efforts frequently prioritize technical safeguards over the reflexive, intersectional, and socio-emotional practices necessary to address systemic inequity. Without historically grounded approaches that center positionality and power, RAI risks reproducing rather than challenging structural injustice.

This dissertation introduces “liberatory AI”, a pedagogical framework built on five pillars: sound racial identity, critical consciousness, liberation centered academic identity, collective obligation, and activism skills. This framework is designed to equip African American youth with the tools to audit AI systems and mitigate algorithmic oppression. I applied this framework across three Data Activism Programs that I created and implemented from 2022 to 2024. Together, my research integrates critical qualitative and quantitative methods to evaluate both students and industry professionals critical data and AI literacy skills.

Three research questions guide this inquiry: (1) What type of training can support Black youth, who have been historically minoritized in the field of computing, to integrate justice-oriented frameworks into the data science and AI development process? (2) How do industry professionals, some of whom come from historically minoritized backgrounds, understand and enact reflexivity in their AI work? (3) What are the long-term effects of data activism training on Black youth's engagement with justice oriented data science and AI development?

To address these questions, three interrelated studies are presented. Part 1 evaluates how grounding AI education in youth's lived experiences and historic proximate knowledge enables students to reason as ecological auditors of AI and resist AI solutionism. Part 2 presents an interview study with tech employees, revealing how practitioners incorporate reflexivity into AI development and usage. Findings further illuminate how practitioners seek greater institutional support for collaborating with communities with diverse lived expertise in the creation of AI, all in service of embedding reflexivity more deeply into AI risk assessments. Part 3 examines the long-term outcomes of 3 Data Activism Program cohorts through a mixed-methods design incorporating surveys, interviews, and arts-based research with 27 former participants. Results show that the program produced lasting improvements in students’ embodiment of liberatory AI principles. This includes the integration of data science and critical AI skills into their daily lives, a deepened commitment to social justice, and meaningful growth in their academic and career trajectories. Overall, these findings demonstrate that without centering reflexivity, identity, and justice, there can be no truly Responsible AI. When these practices are meaningfully integrated, both youth and professionals gain the agency to transform AI from a site of harm into a tool for liberation.

Cynthia Breazeal
Professor of Media Arts and Sciences
MIT Media Lab

Gretchen Brion-Meisels
Senior Lecturer on Education; Faculty Co-Chair, Identity, Power, and Justice in Education Concentration
Harvard Graduate School of Education

Catherine D'Ignazio
Associate Professor of Urban Science and Planning
Massachusetts Institute of Technology