Panel on Artificial Intelligence
The panel on AI explored its rapid evolution, challenges like bias and safety, and its societal impact, emphasizing the need for robust evaluation and interdisciplinary collaboration. Speakers highlighted power imbalances, misconceptions, and the importance of balanced regulation to harness AI’s benefits while addressing its risks.
Speakers:
-
Rama Chellappa, Bloomberg Distinguished Professor in electrical and computer engineering and biomedical engineering and chief scientist at the Johns Hopkins Institute for Assured Autonomy (IAA)
-
Henry Farrell, Stavros Niarchos Foundation Agora Institute Professor of International Affairs at Johns Hopkins School of Advanced International Studies; 2019 winner of the Friedrich Schiedel Prize for Politics and Technology. Author, “Underground Empire: How America Weaponized the World Economy”
-
Anjalie Field, Assistant Professor in the Computer Science Department at the Whiting School of Engineering at Johns Hopkins University
-
Anqi Liu, Assistant Professor in the Computer Science Department at the Whiting School of Engineering at Johns Hopkins University; affiliated with the Johns Hopkins Mathematical Institute for Data Science (MINDS) and the Johns Hopkins Institute for Assured Autonomy (IAA)
-
Moderated by: Mark Dredze, John C. Malone professor of computer science at Johns Hopkins University and Director of Research (Foundations of AI) for the JHU Data Science and AI Institute
Key Takeaways
The panel on Artificial Intelligence (AI) opened with reflections on the rapid evolution of AI, marked by the introduction of tools like ChatGPT, which have significantly increased public awareness of AI capabilities. Rama Chellappa highlighted the problem of domain shift, where AI struggles to perform outside its training environment and noted the trade-offs between eliminating bias and maintaining system performance. These examples of challenges emphasized the need for robust evaluation and continuous improvement.
The discussion pivoted to the power dynamics and safety of AI deployment. Anjalie Field highlighted the unequal access to the resources needed to create state-of-the-art AI models. If deployed without addressing underlying biases, AI systems risk widening existing power imbalances. Anqi Liu talked about the challenges of defining safety in AI, particularly with general-purpose models, and emphasized the technical difficulties in explaining AI decisions due to the black-box nature of deep learning models. Moreover, the integration of AI into democratic processes was examined, with Henry Farrell noting both its potential to empower marginalized groups and the dangers of unequal access to AI-driven resources.
The panelists also addressed misconceptions about AI, such as the belief that it is deterministic or infallible. Mark Dredze pointed out that AI models can create plausible-sounding, but wholly inaccurate information, making it hard to distinguish between factual information and misinformation. The dialogue then shifted towards the importance of evaluation, which should not be confined to the lab but needs to measure how a system performs in real-world contexts, particularly in situations where AI systems may be interacting with people.
The speakers explained that while fears of existential AI risks may be overblown, the immediate challenges of fairness, robustness, and societal impact demand urgent attention. The panel concluded by emphasizing the need for interdisciplinary collaboration to navigate the ethical, technical, and societal challenges posed by AI, advocating for balanced regulation and ongoing dialogue to maximize AI's benefits while mitigating its risks.