The River and Who It Drowns First

EDO SEGAL: I want to put my own metaphor on the table and let you both break it, because it's the one I'm least sure of. In [YOU] on AI I describe intelligence as a river — a current that's been flowing and finding new channels for a very long time, through chemistry, biology, language, culture — and I describe the present moment as an acceleration, the river moving faster than we've ever seen. I wrote that acceleration as mostly a marvel. But a river that moves faster doesn't lift everyone equally. It drowns the people standing in the low ground first. Timnit, I think you'd tell me my river metaphor is true in a way I didn't intend. Pamela, I think you'd tell me rivers have always been like this. Let me start with you, Pamela, because I want Timnit to answer the historian.

McCORDUCK: Rivers have always been like this, and I mean that as agreement with Timnit, not as comfort. I lived through the cycles — I helped create one. In 1983 I co-wrote The Fifth Generation with Edward Feigenbaum, warning that Japan's national AI program would leave America behind. The book was enormously influential. It helped inflate the expert-systems boom. And then the boom collapsed into what the field came to call an AI winter — withdrawn funding, deflated hope, a wall the technology hit hard. I watched the wave crest and break, from the inside, as a believer. So when I say the river surges and recedes, I'm not theorizing. The danger I'd add to your metaphor, Edo, isn't only that the fast river drowns the low ground. It's that the people who profit from the surge are never the ones standing in the low ground, and they're long gone — onto the next surge — by the time the water recedes and the damage is counted. The river floods, the promoters leave, and the people downstream are left holding the consequences of a flood they were told was salvation.

EDO SEGAL: Timnit, the historian just handed you the cycle. What does the engineer who measures the harm do with it?

GEBRU: I take the cycle and I refuse the fatalism it usually smuggles in. Pamela's right that the surge-and-recede is real, and she's right that the promoters are gone before the bill comes due — that's the gold rush structure, where fortunes are made less by finding gold than by selling shovels and spreading the belief that gold is there. But here's where I won't follow the river metaphor, Edo, and it's important. A river is a natural force. It has no owner, no intention, no choice. When you call this a river, even a fast and dangerous one, you've already conceded the thing I most want to contest: that it's a force of nature we can only adapt to, build levees against, get out of the way of. It isn't. It's a set of decisions. The "acceleration" is not water finding a channel; it's specific firms choosing to ship faster than rivals, choosing to scrape rather than license, choosing to deploy before they understand the harm, because the incentives reward speed and punish caution. You can't negotiate with a river. You can absolutely negotiate with a corporation — or regulate it, or refuse it. The metaphor disarms the very people who most need to know they have agency.

McCORDUCK: That's a real correction and I'd accept most of it. But let me defend a sliver of the river, because I think there's something in it Timnit's framing loses. Not everything about the surge is a corporate decision. Some of it really is more like a current — the way a genuinely useful capability spreads once it exists, the way the girl in Addis Ababa reaches for the tool because it actually helps her with her physics, not because a marketing department told her to. There's a democratization happening that isn't only theft and hype. The danger of treating it all as pure corporate choice is that you miss the part that's genuinely a flood of capability reaching people who never had it — and those people aren't only victims. Some of them are being lifted. The river isn't only drowning. It's also carrying. Both are true, and a politics that sees only the drowning will lose the people the river is carrying.

GEBRU: I'll grant the carrying, with a condition, and the condition is everything. Yes — the tool genuinely helps the girl with her physics, and I would never tell her not to use it; I'm not a Luddite and I've never been one. But ask what she's getting along with the answer. She's getting a worldview with the answer — the defaults, the blind spots, the sense of what goes without saying, of a particular corpus, overwhelmingly English, overwhelmingly the English of the people with the access and leisure to publish online. The river carries her, yes, and it carries her in a particular direction, toward a particular center, while the data that would have reflected her own context back to her was never collected. That's not democratization. That's the oldest imperial habit there is, automated — the export of one center's assumptions to everyone, fluently, invisibly. So when you say "lifted," I ask: lifted toward whom? Into whose frame? The carrying and the colonizing are the same current. You can't separate the lift from the direction of the flow.

EDO SEGAL: Let me restate that, because it's the hardest thing said tonight and I don't want it to slide past. Pamela: the river is also carrying — real capability reaching real people who never had it, and a politics that sees only the drowning loses them. Timnit: the carrying and the colonizing are one current, because the lift always points toward a center, and the girl gets the worldview baked into the answer she never asked for. Pamela, here's my question. Can you hold Timnit's direction-of-the-flow and still call it democratization? Or does her point dissolve the word?

McCORDUCK: It bruises the word; it doesn't dissolve it. Here's the distinction I'd hold onto. A printing press also spread one center's assumptions — it spread them in Latin, then in the vernaculars of whoever owned the presses, and it absolutely carried imperial cargo. And it also, genuinely, put reading into hands that had never held it, and those hands eventually used it to write things the press owners never intended, including revolutions against them. The direction of the flow is real, Timnit, and it's also not destiny, because the people being carried are not inert. They reshape the current downstream. I'd call it democratization contested — a real distribution of capability that arrives loaded with one center's assumptions, and that the periphery will bend, given time and the right institutions, toward its own ends. Your DAIR is exactly that bending. You're not standing outside the river telling people to get out. You're trying to redirect the current from inside it.

GEBRU: [pause] "Democratization contested" — I can live inside that phrase, because the word "contested" puts the agency back where the river metaphor took it out. That's the whole fight, isn't it. Not whether capability spreads, but whether the people it reaches are treated as a market to be captured or as agents who get to bend the flow. And yes — that's what DAIR is. Research with communities rather than on them, on their timeline, answering their questions, refusing to treat community knowledge as raw material to be mined. The river will carry her either way. The question is whether she gets a paddle or just gets swept.

EDO SEGAL: That's convergence three, and it's the most important one yet — both of you now agree the capability genuinely spreads, and the entire fight is over whether the people it reaches are agents or cargo. Hold the paddle. Because the next round is about the surge itself, the thing Pamela has lived through and named — the bubble, the hype, the manufacture of belief. The Fifth Generation, and what a historian who inflated a bubble can teach an engineer trying to puncture one. After this.