A.I. Enshittifies Everything  🔗 **Source:** [A.I. Enshittifies Everything](https://share.snipd.com/episode/357ebc04-cfaa-4881-bd32-83e79540fccb) ## Highlights - **Reverse Centaurs Make Humans The Bottleneck** - Cory Doctorow coins "reverse centaur" to describe machines that direct humans to do tasks the machine cannot, reversing the ideal assistive centaur model. - He uses the I Love Lucy conveyor scene to show humans become the bottleneck when machines dictate human work patterns. Transcript: Lizzie O'Leary Doctorow is a journalist, a sci-fi author, and he coined the term inshittification. The idea that goods and services online degrade over time in the service of profit. Now, Doctorow's written about AI in his book, The Reverse Centaur's Guide to Life After AI. Cory Doctorow The human can't work as long or as fast or lift things as heavy or what have you. And so when you pair the human and the machine, the human gets worked very hard, as opposed to the other way around, when you pair the machine with the human and the human's in charge. And that brings us to reverse centaurs. Yeah. So a centaur in automation theory is a person assisted by a machine. You know, riding a bicycle, you don't just look like a centaur. You are one in some figurative sense, right? You're a human in control of a machine. The machine can do things you can't do, move faster than you, but it doesn't tell you what to do, right? My notes right now. Yeah, your notes, a spell checker, right? Of times that a human just can't pay attention in the same way or can't lift the same weight or can't run as fast or what have you. So we use a machine. The reverse centaur is the opposite. It's when there's a thing the machine can't do. And so the human is directed by the machine to do the things that the machine needs the human to do so the machine can do its function. So Ethel and Lucy, they're the reverse centaurs. ([Time 0:02:51](https://share.snipd.com/snip/63d26b6c-0de8-4799-8ac7-374bcb187f85)) - **Investor Growth Narratives Drive Harmful AI Use** - Doctorow argues AI deployments are shaped by a growth narrative aimed at investors, forcing reverse-centaur arrangements that prioritize cost-cutting over useful automation. - He links corporate incentives to firing workers and making remaining staff mark AI outputs, creating blameable human overseers. Transcript: Cory Doctorow Every arrangement of AI that is deployed in service to that narrative has to be a reverse centaur one, right? It has to be one where you're like firing nine-tenths of your workers, putting the remaining tenth in charge of marking the AI's homework, and then making them bear the blame when things Go wrong. Which means that that's the destructive power of the bubble. And so a material critique of it has to talk about why these growth stories keep getting thought up, these sort of ever more outlandish growth stories, and also what the problems are With this idea that we can just replace humans with software. ([Time 0:04:35](https://share.snipd.com/snip/80f2dc7e-e8cc-4036-8cf5-1818197d5e19)) - **Trucking Example Shows Automation Numbers Mislead** - Doctorow uses trucking to show automation hype: claims autonomous trucks will replace millions ignore fleet diversity and on-the-ground realities. - He notes BLS categories lump florist vans with long-haul drivers and drivers' precarious pay makes the automation story misleading. Transcript: Lizzie O'Leary I wonder if you could explain that, because I think these really tangible examples are helpful to people if the big concepts are hard to get your fingers on. Cory Doctorow So you and I have been around a bit. You probably remember about 10 years ago, there was a lot of truck driver discourse. Yes. Because the Bureau of Labor Statistics category for commercial drivers, I forget what it's called. It's got a long name. It's got a number. Says that it's the number one job in every state, right? That truck driver is the number one job in every state. And so you get these people just as like the AI is achieving liftoff. And I want to say there is like another material basis for the AI bubble, which is that the software is interesting. Lizzie O'Leary There are 2.2 million heavy tractor trailer drivers. That's right. Based on 2024 numbers. Continue. Cory Doctorow So the software's taking off. All these dividends to scale are pouring in, right? They do the same trick that they did to get more performance out of these existing machine learning techniques. And instead of getting diminishing returns to scale, they get flat returns to scale. Every time they do more of it, they get more out. So people are getting really excited and they go, guess what? We could take trucks and we could give them their own lane and we could have AI coordinate among them to create like a constant flow uh where you wouldn't need a driver and we would offset All of the wage bill for those drivers and then the number one job in every state would go away and that's a huge automation labor crisis can we believe how crazy this is going to be so it Turns out the number is deceptive in many ways. Well, first of all, this is a bad truck, right? It's not just a bad truck. It's an even shittier train, right? A thing with its own lane that is just a bunch of container cars together. It's just like a less fuel efficient train, right? We don't need that. But also the BLS category encompasses everyone who drives heavy goods vehicles or commercial vehicles, including like florist vans. And so now we've got these florist vans that are being incorporated into this number that was being bandied about at the time. And, you know, the number of people you need in the florist van, whether or not it's driving itself is one. Because you can't just like catapult the flowers onto someone's porch or into someone's apartment. Lizzie O'Leary I guess the theory is a little robot could do it. Cory Doctorow But now we're citing facts, not in evidence, right? Where is that robot and how reliably does it work? So this idea that we've invented the shitty train and that it's going to put everyone into work when in fact the number of workers you need hold steady is silly enough. But then you actually get granular on the way that heavy goods vehicles drivers operate. And like illustrators, they're among the most immiserated workers in America. So they are basically gig workers. They are not paid for time when they're waiting. So if like waiting at the Port of Los Angeles for 16 hours, that's unpaid wages. They buy their trucks on higher purchase through these installments sale agreements, where if you miss a single payment, it's a full repossession. So you might have paid like $90,000 of the $95,000 in and you miss a payment and then they take it back and sell it to a different driver and start all over again. So they're like the worst treated workers in America whose wages have been eroded to the point of nothing. And so now we've got this thing where it's like, well, we're going to spend billions and billions of dollars and also have a national freak out about the coming robot apocalypse because We think maybe we can make a shitty train. ([Time 0:15:44](https://share.snipd.com/snip/acba0e42-7eb1-4ba8-9a3c-8bb05ef4f1a3)) - **AI Bubble Rewards Hype Over Fundamentals** - Doctorow highlights the bubble's scale: massive valuations and hype make fundamentals irrelevant and create Ponzi-like dynamics. - He explains AI's unit economics worsen with scale: prolific AI customers lose more money, unlike prior web cycles. Transcript: Lizzie O'Leary Right? But here we are. It is late June. As you yourself say, seven giant AI companies account for 35 percent of the U.S. Stock market. We are talking in a week where major tech stocks have freaked out and indexes have bounced around a little bit. We don't totally know where that's going. Do you understand the way public markets are reacting to this now? And I guess the place we are in in this arc or bubble or like, where are we? Cory Doctorow Corey Robin wrote something really good about this, that we can see the crazy valuations of the AI companies, especially now the SpaceX IPO as as reflecting something that started To happen in the cryptocurrency era where it's really just vibes based. The fundamentals have gone out the window. Firms that lose money are getting crazy valuations and they don't just lose money. They also have bad unit economics. ([Time 0:20:31](https://share.snipd.com/snip/c67dee93-c230-4db8-8136-1113f7a3da49)) - **Cancer Example Shows Risk Of Automation Blindness** - Doctorow and Lizzie, both cancer survivors, discuss AI in radiology: sales pitches often promise assistants but aim to cut staff and have humans rubber-stamp AI outputs. - He warns this creates automation blindness where remaining radiologists click through and bear blame for missed diagnoses. Transcript: Cory Doctorow Okay. Lizzie O'Leary Because there's this part in the book where you say, look, this is not about AI taking your job. It's more that an AI salesman can convince your boss to fire you and then the AI will not be able to do your job. Cory Doctorow Sure. Lizzie O'Leary So you and I are uniquely qualified to talk about this. I've had cancer. You've had cancer. Indeed. You seem to be doing well. Cory Doctorow I got my cancer-free diagnosis four days ago. High five. Yeah, five now. Okay. Yeah. Lizzie O'Leary But companies love to say that AI can detect cancer. Sure. So I think who better to like try to puncture this a little bit than like we have some moral authority. Yeah. Yeah. Okay. So there's this example that you can use AI to find solid mass tumors and that maybe people can get rid of some radiologists. Talk me through the example. Yeah, sure. Cory Doctorow So, you know, if I imagine a sales call down at the Kaiser Hospital in LA where I get my treatment, where there's a sales guy in there talking to the CEO of the hospital system, and they're Like, look, you've got 10 radiologists on staff at $300,000 a piece. That's $3 million in your wage bill. I'm going to sell you a chap-op for another million dollars a year, and it's going to sit on their shoulders. And twice a day, it's just going to whisper in their ear, take another look at that one. Because sometimes it can detect things that humans miss. Then I'd be like, this sounds great. I mean, you know, I don't want my premiums to go up, but this seems great. If on the other hand, there was a pitch that was like, we've got the software that works so well that we can fire all the radiologists and they weren't lying. And they were saying, let's get rid of all the radiologists. We'll have software do it. It'll be too cheap to meet her. I can imagine a potential cancer patient or someone who loves someone with cancer who says, look, the point of radiology is to fight cancer and not to pay the mortgage of radiologists. And so let's find a soft landing for those people. But I think that what the sales call actually consists of is something that that person wouldn't like, even if they don't give a damn about the radiologist. I think the sales call is fire nine tenths of your radiologist. Save $2.7 million a year on the wage bill. Split it between my boss, Sam Altman, and your shareholders or whoever, however you want to distribute it. Have the remaining radiologist mark the homework of the AI at a speed that no human being could ever pay attention to. Have the x-rays be more or less correct most of the time so that all they're doing is clicking OK over and over and over again and become inured to it, get automation blindness. And then when people die of cancer, we're going to blame them because it's their signature at the bottom of it. They're the accountability. Lizzie O'Leary And it's the automation blindness because most of the time it's right. Yeah. I mean, just like, yeah, yeah, yeah, yeah. If you've been through a TSA checkpoint, you know how this works, right? Cory Doctorow So we have, can I swear on this podcast? Of course. We have created- You're literally wearing a poop. I am wearing a poop. ([Time 0:25:36](https://share.snipd.com/snip/d8a58c94-1b88-4efb-b2c3-47738c8f2bf9)) - **Do Politics Not Just Shopping To Shape AI** - Do political organizing and local engagement rather than relying on individual consumer choices to influence AI deployment. - Doctorow recommends joining groups like EFF, influencing local policy on data centers, surveillance, and municipal AI systems. Transcript: Cory Doctorow Well, I have good news and bad news. The bad news is that shopping isn't politics. And as much as it would be really nice if you could resolve all of the world's political problems just by shopping really carefully. I like shopping, right? It's fun. You don't need to get anyone else to come along and do it with you. It's quite solitary. Hell is other people. You don't have to convince anyone else to come and shop with you. I think that would be very nice if you could just vote with your wallet. You can't vote with your wallet. Billionaires love it when you try to vote with your wallet because they can vote with their wallets because their wallets are thicker than all of our wallets put together. But we can't vote with our wallet. So we always get outvoted. So the way that you do politics is by doing politics. So even like a boycott, you know, the Montgomery bus boycott was not like a bunch of people waking up every morning for 380 days and deciding to walk to work, right? It was a cohesive movement where they had like meetings, but those meetings were to organize carpools and also to put pressure on bosses that were leaning on workers who were late because They were in the carpools and to articulate a crisp set of demands and to get the whole community behind them and while um this is the good news while it was a pain in the ass to have to do all Of that work rather than just shopping on the other hand the residue of that effort was the backbone of the civil rights movement but that is also a group of people who could very easily Lizzie O'Leary Identify and see sure the material harm to them their families society etc this is much more diffuse it is well okay i don't think there's any shortage of people who are correctly worried Cory Doctorow About ai but maybe incorrectly understand maybe they're not bound together yeah And so that's the, that's how you get a movement, right, is by like articulating a kind of agenda and A set of causal explanations for things that a lot of people are worried about, but don't really understand what to do with. And so, you know, I've worked for the Electronic Frontier Foundation for 25 years now. And, and EFF, we do a lot of community outreach. We organize a lot of local groups. If you go to EFF.org, you can find ways to do that. There's also a lot of local politics intersection with this. Because of data centers? Because of data centers, because of AI usage in municipal politics. So, you know, flawed cameras, automated decision making, you know, your kids being taken away from you by CPS or flagged as disciplinary problems by their school because of AI or the Cops deciding where to police or the local courts deciding who gets parole. And, you know, if there's one thing we learn from Moms for Liberty, it's that like the stupidest people in the world can take over unregarded local offices and completely upend the material Conditions for people around them for the worse, which means that people who aren't just total idiots could do stuff for the better. And, you know, a failure to involve yourself in this ends up resulting not just in data centers and like all the farmland being taken over by eminent domain for data centers and, you know, All your water and energy being siphoned off and so on. It also determines whether or not there's like flock cameras that are used by chuds to kidnap your neighbors and send them to Salvador and slave labor camps. So, you know, this is like getting involved in local politics is good. And you can move the needle on local politics a way that's much harder than moving state or federal politics. ([Time 0:30:01](https://share.snipd.com/snip/2349b3bd-692a-469b-9a81-0cb2fc8d0f89)) - **Use Unions And Antitrust To Counter Abusive AI Cuts** - Unionize and pursue antitrust measures at workplaces to give workers leverage against harmful AI-driven reorganizations. - Doctorow ties historical labor movements and antitrust origins to modern needs for collective bargaining and trust-busting. Transcript: Cory Doctorow And then the last place to get involved is in your workplace. Right. Lizzie O'Leary I was going to say, you are a big advocate of unionization. Full disclosure, I am a WGA member. I'm unionized. But that also does seem like a very old fashioned answer to what feels like a modern problem. Cory Doctorow Well, it's funny because we used to not have a lot of the problems that we have now. We used to have like ways of addressing them. And then we stopped doing things like trust busting and organizing workers in unions. And then the problems came back. And now we say maybe we should do trust busting in unions. It's sort of like, well, those are very old fashioned. But, you know, it's kind of like we used to not have a rat problem because we had all this rat poison. And then these neoliberal economists from the Chicago school said, you don't need the rat poison. And now rats are eating our faces off. And they're like, why are you blaming us for the rat problem? This is the great forces of history creating the time of the rat, making them more fecund than at any time in history. And while the rats did buy and shut down all the rat poison factories, that's just economically rational when you're not using rat poison. Rat poison is so old fashioned. What would we use the rat poison for? I mean, I don't think a union in 2026 is the same thing as a union in 1936. Yeah. But I also think that it shares a lot of DNA with it. And the problems that we have are not dissimilar to the problems we had then. And in fact, like the material conditions are not a lot dissimilar. ([Time 0:33:22](https://share.snipd.com/snip/d4739965-2346-4d14-979f-9515077d7d14)) - **Free AI Services Are A Bubble Subsidy** - Doctorow says current AI freebies are unsustainable subsidies: companies sell high-value services at a loss to buy users and will stop when the bubble bursts. - He warns users may face paywalls or service collapse once investors demand profitability or funds dry up. Transcript: Cory Doctorow Well, I think that we might be able to have that stuff. I mean, right now, the reason you're getting all that stuff and it feels great to you is because the AI companies are selling you $100 bills for a dollar a piece. What do you mean? They're just losing gigantic amounts of money. There's this whole question about what Mythos, the Mythos model was. I was speaking to a friend of mine who's quite a well-known security expert who said, and also a big AI guy. And he said, what I think they've done is they figured out how to make an AI that can produce a $20,000 security audit for about $100,000. And that's why they don't want to release it to the public, because they are going to lose a lot of money on it. And you may remember a couple of weeks ago, they tried to clean up their balance sheets preparatory to these IPOs that are being teed up. And they said, how about if we charge everyone $5 for these $100 bills? And all their major customers were like, that's too expensive. Lizzie O'Leary We are not getting $5 worth of value from these well we're now starting to see companies be in these token crises where they're like wait we've used all our tokens yeah yeah and and they Cory Doctorow Can buy more right but they're like well yeah these hundred dollar bills were a bargain at a dollar a piece but they're not at five dollars a piece so you might like claude but do you like Claude well enough to pay a thousand dollars a month for it? That's one question. But the other question is like, what do we do when the bubble's over? Like what AI will we have? Because there's this bedrock of finance Stein's law that anything that can't go on forever eventually stops. They're either going to unload on normies and we're going to stop subsidizing them and then we'll all go broke. Or they won't find suckers and then they'll stop subsidizing them and the companies will fail. But one way or the other, this selling $100 bills at a dollar apiece just can't go on forever. You can't keep shoveling money into the money furnace. ([Time 0:35:42](https://share.snipd.com/snip/291ae3b7-e9db-42c4-ad77-8b12b6309074)) - **Bubble Rooted In Growth Stories Not Inevitability** - Doctorow's main takeaway: the AI bubble's material basis is corporate growth narratives, not inevitable job-replacing tech. - He urges stopping current delivery models quickly to avoid larger future bubbles and deeper damage. Transcript: Cory Doctorow That the bubble has a material basis. And that material basis is not that AI is going to take your job. That material basis is that companies need a growth story and that stories about replacing workers are very attractive to financiers and to bosses. And that the worse the bubble gets, the worse the reckoning will be. And that whatever you think about AI, that this way of delivering AI is the worst way we can imagine. ([Time 0:37:47](https://share.snipd.com/snip/bb4f8fa4-5fdc-4cac-83f3-4e0a5d9d8e9d))