Towards Understanding and Practices of Ethical Artificial Intelligence
Abstract:
The successes of machine learning (ML) and artificial intelligence (AI) models encourage their widespread deployments in high-stakes domains -- from public transportation to social decision making such as autonomous driving, criminal justice, and company hiring. Such widespread deployments call for assessing and addressing the ethical concerns of AI systems. The thesis aims to develop practical techniques and theoretical understanding for building ethical AI systems, focusing on the two ethical AI principles--fairness and privacy. To this end, our preliminary works can be divided into two lines. The first line of works focuses on automatic information extraction using natural language processing (NLP) from privacy policies, a type of documents about how different stakeholders (e.g., users, ML services providers, etc.) agree on how the services providers commit to using, sharing, and protecting users' data. Such developed NLP techniques could be extended to other natural language law documents describing other ethical AI principles and help improve mutual trust among different parties. The second line of works focuses on the theoretical understanding and development of algorithmic intervention under ethical AI principles for different tasks of interest, including (1) studying the costs and benefits of adversarial representation learning for information obfuscation; (2) understanding and mitigating accuracy disparity in regression; (3) achieving return parity in sequential decision making. In this proposal, we will dive deeper into the second line. Specifically, we aim to explore further whether it is feasible to use emergent machine learning techniques (e.g., contrastive learning) to ensure particular ethical AI requirements (e.g., fairness) and understand their implications in different real-world challenging scenarios, including but not limited to few-shot learning and learning under noise. The expected contributions of the thesis will facilitate the practices of building ethical machine learning systems and help increase the understanding and trust among different stakeholders towards the machine learning systems.
Committee:
- Aidong Zhang, Committee Chair, (CS/SEAS/UVA)
- Yuan Tian, Advisor, (CS/SEAS/UVA)
- Yixin Sun (CS/SEAS/UVA)
- Cong Shen (ECE/SEAS/UVA)
- Han Zhao (CS/UIUC)