Research

On this page, you can find my research motivation & statement and below that a list of my publications. You can also find a list of my papers on my Google Scholar profile.

Motivation

Inspired by the potential of intelligent technologies to improve our lives, I began my M.S. in Computer Science at the University of Pennsylvania (UPenn) in 2020. I took the course CIS 522 Deep Learning, which repeatedly raised the question of how we – as developers – can design more ethical algorithms. Interested in how the industry approaches the topic, I reached out to former colleagues and contacts, asking about how their companies manage the ethical implications of AI. Surprisingly, no one had an answer. They weren’t managing adverse outcomes at all because they didn’t know how. Given the increasing use of such systems throughout society, this sparked my interest in contributing to research in responsible AI.

The possibility that AI systems could be exploited, misused, or inadvertently harm global stability and political, social, and human rights further calls for significant collaboration, supervision, and technically robust regulation, particularly by leading AI nations. This is why, in additional to technical aspects of responsible AI, I got into AI governance. AI governance more broadly requires an interdisciplinary lens, including expertise from, e.g., computer and political science. Coordinating and integrating insights from these multifaceted fields is complex and requires an understanding of both areas, specifically in the field I am focussing on: technical research for AI governance, i.e., designing and developing tools and mechanisms that are technological in nature and are instrumentally valuable through informing or enabling effective AI governance. This line of work makes it necessary to deeply understand the current capabilities and limitations of AI systems while also being aware of the (geo-)political constraints, dynamics, and mechanisms in a policy setting to design governance instruments that are robust, effective, and enforceable.

My research agenda is structured around three main areas. First, I am exploring the factors contributing to the responsible development and deployment of AI systems. Research questions I’m interested in include exploring the quality and validity of model evaluations and how we can operationalize abstract concepts like ‘robustness’ on a technical level.

Second, I am analyzing national AI policy aspirations and the technical tools needed to fulfill them. My goal is to highlight gaps and encourage the AI research community to address these gaps, some of which I am tackling directly. For instance, although numerous jurisdictions mandate evaluations of generative AI systems pre-deployment to ensure their safety, from a technical standpoint, there is a lack of clarity on how to perform these assessments both comprehensively and reliably (Chang et al., 2023; Zhou et al., 2023).

Third, I have been working on a project comparing regulations across countries to find a consensus for international AI governance, focusing on technical provisions like privacy, security, and transparency. In conjunction with my prior work on international AI governance (Trager et al., 2023), the goal is to lay the groundwork for international negotiation and determine which technical aspects of AI development can be governed globally.

Gen ZEO Top Talents Under Awards 2019, winners of the category ‘Entrepreneurship’

Publications

Reuel, A., Hardy A., Smith, C., Lamparth, M., Kochenderfer, M. (2024). BetterBench: Assessing AI Benchmarks, Uncovering Issues, and Establishing Best Practices. Under review at 2024 Conference on Neural Information Processing Systems.

Reuel, A., Bucknall, B., Casper, S., Fist, T., Soder, L., Aarne, O., Hammond, L., Ibrahim, L., Chan, A., Wills, P., Anderljung, M., Garfinkel, B., Heim, L., Trask, A., Mukobi, G., Schaeffer, R., Baker, M., Hooker, S., Solaiman, I., Luccioni, A. S., Rajkumar, N., Moës, N., Ladish, J., Guha, N., Newman, J., Bengio, Y., South, T., Pentland, A., Koyejo, S., Kochenderfer, M. J., & Trager, R. (2024). Open Problems in Technical AI GovernancearXiv preprint arXiv:2407.14981.

Reuel, A., Soeder, L., Bucknall, B., & Undheim, T. A. (2024). On The Importance of Technical Research and Talent for AI Governance2024 International Conference on Machine LearningAccepted as oral – top 1.5% of papers

Reuel, A. & Ma, D. (2024). Fairness in Reinforcement Learning: A SurveyAAAI AI, Ethics & Society 2024.

Rivera, J.-P.*, Mukobi, G.*, Reuel, A.*, Lamparth, M., Smith, C., & Schneider, J. (2024). Escalation Risks from Language Models in Military and Diplomatic Decision-Making. In Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency (FAccT ’24), 836-898.

Reuel, A.* & Undheim, T. A.* (2024). Generative AI Needs Adaptive Governance. Under review at Digital Policy, Regulation and Governance.

Undheim, T. A. & Reuel, A. (2024). A Literature Review of AI Governance Trends, 2020-2024. Under review at AI & Society.

Trager R., Harack, B. Reuel, A., Carnegie, A., Heim, L., Ho, L., Kreps, S., Lall, R., Larter, O., Ó hÉigeartaigh, S., Staffell, S., & Villalobos, J. (2023). International Governance of Civilian AI: A Jurisdictional Certification Approacharxiv:2308.15514.

Nie, A., Reuel, A., & Brunskill, E. (2023). Understanding the Impact of Reinforcement Learning Personalization on Subgroups of Students in Math TutoringInternational Conference on Artificial Intelligence in Education, pp. 688–694.

Schuett, J.Reuel, A. & Carlier, A. (2023). How to design an AI ethics boardAI & Ethics.

Lamparth, M., & Reuel, A. (2023) Analyzing And Editing Inner Mechanisms Of Backdoored Language ModelsProceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency (FAccT ’24).

Reuel, A., Peralta, S., Sedoc, J., Sherman, G., & Ungar, L. (2022). Measuring the Language of Self-Disclosure across CorporaFindings of the 60th Annual Meeting of the Association for Computational Linguistics 2022.

Reuel, A., Koren, M., Corso, A., & Kochenderfer, M. (2021). Using Adaptive Stress Testing to Identify Paths to Ethical Dilemmas in Autonomous SystemsProceedings of the AAAI-22 Workshop on Artificial Intelligence Safety.

*Equal contribution.