
Swati Srivastava: A Black Box Within a Black Box
Everyone wants someone to blame when AI goes wrong. Swati Srivastava thinks that instinct is exactly why the problem is so hard to fix, and she wrote a short, free book about it. Swati is an associate professor at the University of Maryland, in the Department of Government and Politics and the AI Interdisciplinary Institute, where she directs the International Politics and Responsible Tech Lab. She's also a faculty associate at the Berkman Klein Center, which is where Mara heard her speak and then hounded her until she said yes. She used to study tech platforms. When the companies she studied started pouring money into AI and firing the people she knew, she followed the money and gave herself a second, do-it-yourself PhD. Her starting point: corporations are already built to displace blame, and now we're putting AI systems inside them. "We're putting a black box within a black box." The courtroom version of responsibility (who knew what, and when) falls apart when the harm can't be traced. She walks us through the "many hands" problem, which Helen Nissenbaum applied to computing in 1996 and which foundation models have supercharged, and the history of contractors overseeing contractors that she studied in her first book. Twenty years later, the plan is AI overseeing AI. Then the airport bathroom. A model built for one job gets reused for another, and suddenly the faucet won't run for her hands. Her workaround is to wait for a white woman to use the sink first. "This is reparations, you just have to give me some water." Mara offers to walk her into every bathroom for the rest of her life. Swati's answer is structural responsibility. We aren't to blame for AI harms, but we are responsible for repairing the unjust structures behind them. Thin responsibility goes after the powerful developers. Thick responsibility pulls in deployers, users and the rest of us as publics. "The difference between coercion and authority is legitimacy," and that comes from the governed. Mara connects it to her own work on AI enfranchisement. Then it gets personal. Swati says being asked whether you're pro or anti AI is "like asking me about how I feel about electricity." Mara talks about feeling like a traitor for using tools she has doubts about, and Logan talks about the guilt that comes with posting online about using AI at all. Swati's answer: it's unreasonable to ask people to opt out, and you shouldn't feel guilty about that kind of stuff. She closes with a "life map" for finding where you, specifically, can do something. "You cannot have a one-person public," but three or four people is enough to start. Her book, Responsible AI and Structural Justice: Addressing Harm When Blame is Hard to Find, is under 100 pages and free to read online. It's linked below. What needs womansplaining? Leave us a voicemail at womansplainingai.com/answering-machine or email ask@womansplainingai.com. Anonymous is fine. In this episode 00:00 Cold open: "The difference between coercion and authority is legitimacy" 00:35 Meet Swati Srivastava 02:00 From studying platforms to following the AI money 02:31 A do-it-yourself second PhD 03:29 "A black box within a black box" 05:22 Why the liability model breaks down 08:37 Agent whistleblowers, agent auditors 08:48 What Blackwater taught her about accountability 09:28 Contractors overseeing contractors, and now AI overseeing AI 10:34 The "many hands" problem 13:48 The airport bathroom 14:33 "This is reparations" 15:47 Logan: is there a positive view of the future? 16:34 Not blameworthy, but responsible for repair 17:44 Thin responsibility and thick responsibility 18:58 Deployers, users and the rest of us 20:14 "Anytime a computer scientist decides not to leave for industry, I clap" 21:00 "AI is coming for my job. Please come quickly" 22:09 Mara on AI enfranchisement and AI fatalism 23:39 Their constituent, not their subject 24:26 Coercion, authority and legitimacy 25:59 The mind meld 26:13 Logan: a lobster in a boiling pot 27:05 "I'm an AI vegan" 27:45 Pro or anti? 28:52 "Like asking me about how I feel about electricity" 29:50 Opting out of AI use isn't opting out of the conversation 31:33 Mara: feeling like a traitor for using the tools 32:05 Why it's unreasonable to ask people to opt out 32:58 "Don't feel guilty about that kind of stuff" 33:52 Maybe don't ask Claude about the weather 34:45 Logan: the guilt is so real 35:42 Where do I start? 35:53 The book is free 37:01 Make a life map 38:51 "I am all fired up" 39:19 See her in an airport? Bring water Resources mentioned Swati Srivastava, Responsible AI and Structural Justice: Addressing Harm When Blame is Hard to Find (Cambridge Elements, free and open access) Swati Srivastava, Hybrid Sovereignty in World Politics (Cambridge University Press, 2022), her first book, including the Blackwater case Helen Nissenbaum, "Accountability in a Computerized Society," Science and Engineering Ethics (1996), which brought Dennis Thompson's "many hands" problem to computing Swati's website
- Transcript













