Four Steps to the Apocalypse: A talk at 60th anniversary of MIT's Project Mac [video]

This is a 21 minute talk by Prof. Keith Winstein. The first half/three quarters of the talk doesn't address his title at all - he reviews his career as a student at MIT, a faculty member at Stanford, and a journalist, and his research on networks and statistics.

The most interesting part of the talk is the last 5 - 10 minutes, when he discusses what worries him about AI. His last slide:

Four Steps to the Apocalypse:

Existential (the Terminator scenario - he doesn't think this is likely. what really worries him is the next three)

Sociological (AI makes us stupid)

Disciplinary (AI makes us lose something core to our discipline --- our confidence that computation is understandable)

Consolidationist (Centralization of computing power so innovation only happens where it is 'supposed' to --- but historically, innovation often comes from unexpected places)

He very briefly proposes a remedy -- something computer scientists can do -- which is a not at all overtly political or social, but instead is a new direction of computer science research. Something about enabling a computation to be made comprehensible outside the organization that computes it. Also, changing incentives so organizations are rewarded for correctness of the result rather than the amount of computation they do. His next to the last slide: "Paying for results, not effort". His third-from-last slide: "We're building an OS for distributed computing that represents computations as first-class (not processes, containers, VMs, lambda invocations)"

I didn't find this remedy to be very clearly explained or convincing - it's crammed into the last few minutes of the talk.