This is a very rough syllabus/bibliography of different articles about the legal issues raised by AI.
What happens when the models break?
- Negligence and AI’s Human Users (Andrew Selbst, 2020). How might negligence law apply when humans use/rely on AI tools for decision making assistance? This Article argues that the use of AI will create several holes in negligence law, which mean that deserving plaintiffs will go uncompensated.
- When AIs Outperform Doctors: Confronting the Challenges of a Tort-Induced Over-Reliance on Machine Learning (A. Michael Froomkin, Ian R. Kerr, Joelle Pineau, 2019). At the current rate, we should expect AI systems to outperform most doctors for a number of medical diagnostic tasks. At that point, relying on the AI system's predictions will become the standard of care. This Article investigates the consequences of such a future.
- Crashworthy Code (Bryan H. Choi, 2019): Drawing on lessons learned from the development of the crashworthiness doctrine in automotive litigation, this Article suggests a similar standard should be applied to negligence in software. In short, thought "[r]equiring all code to be perfect is impossible... expecting code to be crashworthy is reasonable."
What are the implications when governments use AI for adjudication, planning, or other decision making processes?
- The Automated Administrative State: A Crisis of Legitimacy (Ryan Calo & Danielle Keats Citron, 2021): Agencies are increasingly relying AI to automate tasks previously performed by agency officials. But if agencies were originally delegated the power to perform those tasks because of percieved expertise in those areas, doesn't their reliance on AI diminish the need for the delegation? Citron and Calo pose this question, provide an answer, and then suggest a future for a technologically informed administrative state that retains legitimacy.
- Glass Box Artificial Intelligence in Criminal Justice (Brandon L. Garrett, Cynthia Rudin, 2022): This Article suggests that claims about the accuracy vs transparency tradeoff for AI in criminal justice contexts is overblown, and suggests that "Glass Box" AI (i.e. more interpretable methods), offer a more just and lawful path forward.
- Tragedy of the Data Commons (Jane Yakowitz, 2011): The idea that anonymization is insufficient to protect privacy is commonly cited in justifications to limit the release of large public datasets. This Article argues that that this idea is flawed, and threatens to foreclose potentially valuable research findings from these datasets.