Superintelligence

I finished this one on a long stretch of nights where I told myself I’d read one chapter and kept sliding into three, which is not the same as saying it was a fast read. It wasn’t. But it was the kind of dense that rewards you for pushing through it, most of the time.

The core question here is a genuinely important one: what happens once we build something smarter than us, and how do we design in a way to keep control of that situation before it’s too late to matter. Bostrom walks through the plausible paths to a machine intelligence takeoff, the ways containment strategies could fail, and the specific philosophical knots of trying to encode human values into something that will eventually out-think the people who wrote the rules. When he’s on, he’s genuinely on. There’s a passage midway through that puts it as a room full of toddlers each holding a detonator, and one of them is eventually going to press it just to see what happens, and that image has stuck with me longer than most of the book’s actual arguments.

The problem is the surface texture. This reads like an academic paper that got dressed up for a general audience and never quite finished changing clothes. I found myself rereading paragraphs not because the ideas were hard, exactly, but because the sentences were doing more structural work than they needed to. I’ve heard Bostrom interviewed and he’s a genuinely engaging, clear talker in person. None of that ease made it onto the page here.

I don’t regret reading it. The subject matters too much not to have engaged with the actual arguments rather than just the pop-science summary version. But I’d hand a newcomer a good podcast interview first and save this for after they’re already hooked.