Everyone I recommend on AI is optimistic. Except Cory Doctorow.
He punctures hype for a living — and he’s usually right about how things break.
When I started blogging in 2002, one of the first signs it was working was a link from Boing Boing. Cory Doctorow was co-editing it then.
I’m sure he has no memory of me. But I’ve read him ever since, and I’ve heard him speak a few times along the way.
I bring it up because he’s the odd one out on my list of people worth following on AI. Nearly everyone else I recommend is some version of optimistic. Doctorow is the one arguing that most of what you’re hearing about AI is a sales pitch aimed at investors.
Why I trust him
He punctures hype. He’s rigorous. And he’s never put himself in a position where he’d have to soften what he believes to be true — his years at the Electronic Frontier Foundation are a fair marker of that.
He also isn’t in it for the money, which is rare among people who write about technology. He watches where money is flowing, because that’s usually where the next problem starts. That’s what sits behind his argument that the AI build-out is a bubble. He isn’t putting a date on it; he’s describing how the money works. Whether or not it pops on any particular schedule, it’s a reason to think twice before you build your core workflow on a single vendor’s platform.
Here’s what makes him useful rather than just contrarian. He doesn’t make predictions. He points out the things that could come unraveled and shows you the mechanism — and a good share of the time, they do come unraveled.
Centaurs and reverse centaurs
His most useful idea is a distinction between two ways of working with a machine.
A centaur is a person helped by a machine. You stay in charge, the tool adds speed, and you do better work than you could alone.
A reverse centaur is a person serving a machine. The software sets the pace and you scramble to keep up.
That gives you a decent question to ask about any tool you’re considering: does this let me do better work, or does it just make me keep up with it? The same software can go either way. What decides it is how you use it.
The accountability sink
He borrows this term from Dan Davies. His example is a radiologist reviewing scans that AI has already flagged. The AI makes the call, the radiologist signs it, and when the call is wrong, it’s the radiologist’s signature on the record.
The human in the loop isn’t a safeguard there. They’re insulation for whoever bought the system.
You can see where that lands for lawyers. When you sign a brief, the liability is yours. The vendor’s terms of service make sure of it.
Why careful people still miss things
He pairs that with a related problem. When a system is right most of the time, the person checking it loses the ability to spot the times it isn’t. Attention drifts. That’s a feature of the task, not a flaw in the reviewer.
Which means “I always review it” isn’t much of a plan on its own. It’s why a fabricated citation can slip past a careful lawyer, not just a sloppy one.
This isn’t a new finding, by the way. Lisanne Bainbridge described the same problem in a 1983 paper about industrial control rooms — automating the routine work makes the human’s job harder, not easier.
Where to start
pluralistic.net. He posts almost every day — a long essay plus a set of links — with no ads, no tracking, and no paywall. You can read it on the site, get it by email, or follow him on Mastodon. He isn’t on X.
His new book is The Reverse Centaur’s Guide to Life After AI. I have a copy and I’m partway in.
Fair warning: he’s relentless, and the tone won’t be for everybody. He isn’t going to talk you out of using AI. He’ll make you more careful about who benefits from the version you’re being sold.
;-)
Ernie
P.S. In the Inner Circle we spend as much time on what these tools do to your practice as on what they can do for it
→ https://ernietheattorney.net/