The dark side of AI regulation is unrestrained government control - The Washington Post

Georgy Egorov and Konstantin Sonin have peered into the future of AI regulation and discovered, writing in The Washington Post's Ripple section, that it arrives with a considerably larger surveillance apparatus attached. Their claim is that any regulation genuinely effective at slowing frontier development would demand "an extraordinary expansion of the government's surveillance and investigative powers." Well, yes. Enforcement requires enforcement; that is rather how enforcement works. One assumes the tuition at whatever institution employs them comfortably covers this insight.

Note, though, the verb deployed against that expansion: the risks of it "might" outweigh the risks of unregulated development. Might. Two scholars, a national newspaper, an entire column of considered prose, and the firmest word available is a conditional. Meanwhile every lab in Christendom now gets to print out "history shows" and staple it to the front of the lobbying deck it was going to send anyway. The history in question — governments, powers, expansions — is not a ledger that flatters the optimistic reading, which is presumably the point being delicately not made.

The essay notes that momentum is building in the United States for a framework that slows progress at the technological frontier. Momentum, in this town, tends to arrive roughly four years after the thing it intends to regulate. And if Washington ever does acquire the investigative muscle described, it will first have to rebuild what it has spent two years cheerfully shedding — the capacity problem we flagged when CISA lost roughly a third of its workforce has not improved by being ignored. The essay sets out two doors: an unregulated frontier, or a state that reads your email. Nobody sketches the third door, the one where modest powers are competently used, because that would require professing faith in institutions, and faith is not something you can footnote.