“We don’t have to choose between AI and no AI. The bigger choice is what type of AI we want, and who we want in control of it.” — Ben Green
In our age of AI, do we want to be ruled by all-knowing engineer kings? Not according to Ben Green, a professor of information and public policy at Michigan, whose new book, Algorithmic Realism, presents “data science practices to promote social justice”. Unsexy? Yes. And that, according to Green, is the whole point of algorithmic realism. Such a down-to-earth use of AI is super unsexy by design, he explains.
Green’s critique is of Silicon Valley utopians who promote an algorithmic formalism. So he’s dismissive of the idea of “superintelligence” with its promise of curing cancer at the snap of a finger. He borrows from the sociologist Erik Olin Wright the idea of real utopias reached concretely, step-by-step. So he’s not exactly against algorithmic idealism. Just the lazy type that sees the world as an infinitely scalable laboratory.
Green describes working as a data scientist for the city of Boston and being asked to cut EMS response times. Instead of building some sort of super sexy algorithm, he went and talked to people, and found what was actually driving the call volume: mental illness, addiction and homelessness. Shaving two minutes off an ambulance would have done nothing for any of those people. What they built instead — and he is careful to stress the they — was a team trained to handle someone in a crisis and connect them to counselling and shelter. Super unsexy; and super useful.
FIVE TAKEAWAYS
1. His argument is with the engineer’s mindset, not engineering: the assumption that problems have neat optimal solutions and that more powerful tools will of themselves improve society. What it leaves out is power — who decides, and what their incentives are.
2. He does not think you can fix it from inside the labs. They are pushing as hard as they can for superintelligence, and in that setting safety is a PR and liability problem rather than an ambition. Plenty of well-intentioned people there; the business decisions lead.
3. Asked for a concrete example, he gives an ambulance story. Hired to cut Boston’s EMS response times, he talked to people instead of building the algorithm, and found the call volume was being driven by mental illness, addiction and homelessness. They built a crisis-response team instead.
4. He wants narrow AI over general. A system that tops a benchmark and a system that is useful to someone are barely related. His unglamorous advice to cities: clean up your data first. On the scaling strategy: our sausage factory is more powerful than your sausage factory.
5. He is not against utopian thinking, only utopias with no route. Erik Olin Wright’s Envisioning Real Utopias shaped the book: what the labs’ promise lacks is accessible way stations. Radical ambitions, deliberately small steps.
ABOUT THE GUEST
Ben Green is an assistant professor at the University of Michigan School of Information and the Gerald R. Ford School of Public Policy, and a faculty associate at the Berkman Klein Center for Internet and Society at Harvard. He has worked as a data scientist for the City of Boston and writes on algorithms in government for Tech Policy Press. His previous book, The Smart Enough City, was published in 2019. Algorithmic Realism: Data Science Practices to Promote Social Justice is published by MIT Press.
ABOUT KEEN ON AMERICA
Keen On America is a daily conversation about the ideas, books and arguments shaping the United States, hosted by Andrew Keen, author of The Internet Is Not the Answer and How to Fix the Future. New episodes every day.
CHAPTERS
00:00 — Intro
00:31 — The Engineer of Human Souls
02:22 — An age in which engineers reorder society
03:33 — Are the big tech companies doing their best?
06:16 — Can you do good from inside the labs?
09:30 — What the engineer’s mindset leaves out
11:06 — DOGE and reforming government with AI
14:10 — Bannon and Sanders on the same stage
16:49 — What kind of AI would you like?
18:20 — General AI versus narrow AI
19:26 — Clean up your data first
21:28 — Algorithmic realism is unsexy
22:56 — Erik Olin Wright and real utopias
27:37 — The human outside the algorithm
30:22 — A concrete example, please
31:45 — The Boston ambulances
34:39 — Can it scale?
39:03 — AOC versus Bernie Sanders on AI
42:36 — Sausage factories