Author: Nita Farahany Source: https://nitafarahany.substack.com/p/inside-my-ai-law-and-policy-class Published: August 24, 2025
This is the opening lecture of an AI law and policy intro course taught by Nita Farahany at Duke Law and serialized weekly on Substack. Class 1 is the first of 27 installments. It frames the definitional question that Farahany presents as running through the entire semester: how do we write laws for something we can't define?
Summary
Farahany organizes the class around the proposition that competing definitions of AI each imply a different legal regime, so the inability to settle on a definition is itself a governance problem rather than a preliminary one. The class introduces three definitional camps, Helen Toner's steering-wheel metaphor for governance under uncertainty, a classroom exercise on what counts as AI, and an initial technical vocabulary the later classes build on.
Definitional camps
Farahany presents three positions on what AI is, attributing each to named figures and noting that each leads to a different legal regime:
- Geoffrey Hinton's view that the systems "understand. They are intelligent."
- Emily Bender and co-authors' "stochastic parrots" characterization — that the systems are "just predicting next tokens."
- Mustafa Suleyman's "seemingly conscious AI," framed around the prescription to "build for humans, not to seem like them."
Steering-wheel metaphor
The class uses Helen Toner's steering-wheel metaphor for governing AI: it is likened to driving a road that morphs while you drive (capabilities change monthly), with a fogged windshield (the black box), while passengers disagree about whether you are even in a vehicle (the definitional problem). The metaphor's prescriptions are a clear windshield (transparency), good steering (adaptive governance), and working brakes (kill switches).
"Definitely AI" gradient exercise
A classroom exercise asks students to place examples on a gradient of what counts as AI — calculator, spam filter, ChatGPT, Netflix recommendations, and Google search. Farahany uses it to show that what counts as AI shifts as a technology becomes mundane, and draws the implication for law that definitions written for "AI today" may be obsolete tomorrow.
Technical vocabulary
The class introduces core technical terms the course builds on, including the black box, natural language processing (NLP), and generative adversarial networks (GANs).
Relationships
- part-of: Nita Farahany intro course series (Class 1 of 27)
- related: AI Governance (umbrella), Black Box Ai
- next: Inside My AI Law & Policy Class 2: How AI Actually Works (Farahany, August 2025)