Mystery AI Hype Theater 3000

Artificial Intelligence has too much hype. In this podcast, linguist Emily M. Bender and sociologist Alex Hanna break down the AI hype, separate fact from fiction, and science from bloviation. They're joined by special guests and talk about everything, from machine consciousness to science fiction, to political economy to art made by machines.

All Episodes

Mystery AI Hype Theater 3000

Working 9 to Hype, 2026.08.10

September 02, 2026 • Emily M. Bender and Alex Hanna • Episode 84

0:00 | 54:23

In the past year, tech companies have laid off thousands of employees. But when CEOs blame "AI innovation," they're redirecting attention from something much more predictable: job discrimination and gig-ification. Emily and Alex examine some narratives of AI-streamlined workplaces — and why they don't match up with reality.

References:

Block press release: "From Hierarchy to Intelligence"
"Lawsuit claims Meta's layoff decisions were made by AI, not humans"

Fresh AI Hell:

Check out future streams on Twitch. Meanwhile, send us any AI Hell you see.

Find our book The AI Con here, and MAIHT3k merch here.

Subscribe to our newsletter via Buttondown.

Emily

Bluesky: emilymbender.bsky.social
Mastodon: dair-community.social/@EmilyMBender

Alex

Music by Toby Menon.
Artwork by Naomi Pleasure-Park.
Production by Ozzy Llinas Goodman.

Alex Hanna: Welcome, everyone, to Mystery AI Hype Theater 3000, where we seek catharsis in this age of AI hype. We find the worst of it and pop it with the sharpest needles we can find.

Emily M. Bender: Along the way, we learn to always read the footnotes, and each time we think we've reached peak AI hype, the summit of Bullshit Mountain, we discover there's worse to come. I'm Emily M. Bender, professor of linguistics at the University of Washington.

Alex Hanna: And I'm Alex Hanna, director of research for the Distributed AI Research Institute. This is episode 84, which we're recording on August 10th of 2026. This week, we're looking at automation and labor.

Emily M. Bender: In the past year, we've seen huge layoffs at tech companies like Amazon, Google, and Meta, but the reasons that employers are giving for these decisions don't match up with what's actually happening. Tech executives talk about, quote, "The potential of AI to change how we work together," as though the mathy maths are actually replacing all these jobs.

Alex Hanna: And one study found that so far this year, AI has been referenced in nearly 10% of layoff plans across industries. In reality, companies are often hiring back the same workers for lower pay and fewer benefits and calling it, quote, "AI innovation." So let's get into it.

Emily M. Bender: All right, here we go. We are starting with a piece from block.xyz, from March 31st of 2026, and the title here is "From Hierarchy to Intelligence." And right off the get-go, this is written by Jack Dorsey and Roelof Botha, and it has the tags technology, AI, purpose, and Block.

Alex Hanna: I love that Block is one of these, as if- it is right after purpose, as if the purpose is to block you, but of course, this is the name of the company. We should also say that this was released as Block had laid off a bunch of employees, thousands of employees, and I believe Block was the successor to Square.

Emily M. Bender: Yes. I was really sad when I put that together, 'cause Square is the device you use to pay people when you're at your local farmers' market, and they either don't wanna take cash anymore, or they have an option instead of cash. It's that little white thing that allows you to do credit card payments kinda out, away from a landline.

Alex Hanna: Yeah, and it's this little plugin. I think that maybe the device is still called Square, and then the company is called Block. And I'm sure, once they reach AGI, they're gonna change it to Hypercube. It's just going to keep on evolving.

Emily M. Bender: Yeah, and for me, it's like, I hadn't yet associated that device with Silicon Valley weirdness, although it's obvious that it must have been. And for me, it's very much associated with these really lovely in-person, in-my-community interactions. And to realize that's also Jack Dorsey, it was like, "Oh, no."

Alex Hanna: Yeah, there's nothing pure in this world. And we already have great takes into this, including, of course, the next order, which is Tensor, and then also, the facility from the film Cube, because Torment Nexus. So let's get into it.

Emily M. Bender: Yeah. This is long. We're gonna do the top part and then skip on down. So this is a blog post from Block at block.xyz, but it starts with, "At Sequoia, we see that speed is the best predictor of startup success." So in other words, Dorsey's also a major VC, I think, and doesn't quite know which position he's talking from. "Most companies are focused on AI as a productivity enhancer. Few are focused on the potential of AI to change how we work together. Block is showing what it looks like to fundamentally rethink organization design, ultimately harnessing AI to increase speed as a compounding competitive advantage."

Alex Hanna: Yes. So we are going from a square to a block, so we are actually reducing the distance by going from two-dimensional to three-dimensional.

Emily M. Bender: I think they're reducing the distance from speaking to the venture capitalist across the room to licking their ears with this paragraph.

Alex Hanna: Yeah, 100%.

Emily M. Bender: So gross. All right, you want this next bit?

Alex Hanna: Yeah. "2,000 years before the corporate org chart, the Roman army solved the problem that every large organization still faces: How do you coordinate thousands of people across vast distances with limited communication?"

Emily M. Bender: The what now?

Alex Hanna: Yeah, this is the Roman army, which, every middle-aged cis white man likes to look into Roman history or what causes the fall of the Roman Empire. I think this is- when we did that Nonzero podcast, that's also one of the, the podcast's host had referenced as something that he wanted to know about. And you can't make it up that this is the paradigmatic, example for folks.

Emily M. Bender: Yeah, not beating the allegations here. All right.

Alex Hanna: So, yeah, it's worth reading this just 'cause it's silly, and I'm gonna, I don't... Actually, maybe you should read it, because I don't know if I'll say these words correctly.

Emily M. Bender: Oh, great. Thank you. All right. "Their answer was a nested hierarchy with a consistent span of control at every level. The smallest unit was the contubernium, eight soldiers who shared a tent, equipment, and a mule, led by a decanus. 10 contubernia formed a century of 80 men under a centurion. Six centuries made a cohort. 10 cohorts made a legion of roughly 5,000. At each layer, a named commander held defined authority, aggregated information from below, and relayed decisions from above." I think that's probably enough of that.

Alex Hanna: Yeah, and it just keeps on going on. Sorry, I just... My internet brain read, "Eight soldiers who shared a tent," and I just thought, "And they were roommates!"

Emily M. Bender: Yeah. For sure. All right. So then he's basically talking about this as the history of the world, but tracing through Prussia, and then American railroads, and then we get to Frederick Taylor. Would you like to do this one?

Alex Hanna: Yeah, Frederick Taylor, everybody's favorite quantitative scientist. So they say, "Frederick Taylor, often called the father of scientific management, optimized what happened within that hierarchy. Taylor broke work into specialized tasks, assigned them to trained experts, and managed through measurement rather than intuition. This produced the functional pyramid organization, a structure optimized for efficiency within the information routing system that the military had pioneered and the railroads had commercialized." So first off, just jumping off and using Taylor uncritically, and as someone that did the optimization- someone whose name we associate with the degradation of work, but just saying, this is how we went ahead and built on organizational structure.

Emily M. Bender: Yeah. All right. We've got some fun riffs on your earlier shared a tent jokes, Alex. So abstract_tesseract says, "Eight soldiers who shared a tent makes it a men loving men MLM LLM MLM."

Alex Hanna: Oh my gosh, yeah. This is breaking my brain a little bit, but I do love to see new formations.

Emily M. Bender: Yeah, abstract_tesseract is prompting us to do the vocal warmups that we didn't do before the podcast.

Alex Hanna: Ah, yeah. Okay, we'll do that when we return.

Emily M. Bender: MLM, LLM, MLM. Got it. All right, so then we get to the Manhattan Project, which I think we're gonna skip over- McKinsey coming in. "Over time, other frameworks emerged to address complexity, rigidity, and bureaucracy." And then here, "In more recent decades, technology companies have experimented aggressively with organization structure. Spotify popularized cross-functional squads with short sprint cycles. Zappos attempted holacracy, eliminating management titles entirely. Valve operated with a flat structure and no formal hierarchy. Each of these experiments revealed something about the limitations of traditional hierarchy, but none solved the underlying problem." Any comments?

Alex Hanna: Is it holacracy or olacracy? 'Cause it's...

Emily M. Bender: Ola? I don't know, yeah.

Alex Hanna: I don't know. And also, I was gonna say something snarky about Valve, that that's why it's taking it so long for Half-Life 3 to come out. But then they also criticize these folks. And irateLump also says, "Valve still operates like that." Which is interesting. Anyways, they're criticizing these folks, and they're saying, "Spotify moved back towards conventional management as it scaled. Zappos saw significant attrition-" and I believe they were acquired by Amazon. "Valve's model proved difficult to scale beyond a few hundred people." And so then they're basically saying that the traditional, more structured McKinsey-oriented organizational structure won out.

Emily M. Bender: Yes. So they then move on to promoting their great idea. They say, "So what's different now? At Block, we're questioning the underlying assumption that organizations have to be hierarchically organized with humans as the coordination mechanism. Instead, we intend to replace what the hierarchy does. Most companies using AI today are giving everyone a copilot, which makes the existing structure work slightly better without changing it. We're after something different, a company built as an intelligence-" in parentheses- "or mini AGI."

Alex Hanna: Yes, incredible. Just a little co-pilot in their pocket. In the chat, folks are clarifying some of the questions. sjaylett says you pronounce the H, thank you. And twoangstroms says, "Yes, nothing else happened in economics to affect Zappos." So, yeah, very good. Then they go past other types of management structures, whatever. But then talking about what the difference is with AI, so I'm gonna jump into the middle of the paragraph where it says, "AI is that technology. For the first time, a system can maintain a continuously updated model of an entire business and use it to coordinate work in ways that previously required humans relaying information through layers of management."

Emily M. Bender: That's some magical thinking there.

Alex Hanna: Yeah, but the thing about this is that- about this whole document, I should say- what this really shows is that the use of these technologies that get called AI are about command and control. Like, it's a real "yes, and?" So it's really about, oh, we have something that effectively suggests it has a full accurate mapping of the organization, and can effectively direct human resources, in this case workers, to do what needs to be done. And what it also does is it really takes out the manager from any kind of way that they are someone that makes demands, that puts unreasonable expectations on people, and then becomes this kind of accountability shield or dodge. And so it's like, "Yeah, this is about command and control, and that is what this is about, and that's why we're laying off, 30,000 of you."

Emily M. Bender: Yeah. And it's gonna get even more bananas as we go through this, because, they need a world model, fine. And basically what they really mean is you need a whole lot of data. And so they get into how they've got that data. This paragraph says, "Block is remote first. Everything we do creates artifacts. Decisions, discussions, code, designs, plans, problems, and progress all exist as recorded actions. It's the raw material for a company world model. In a traditional company, a manager's job is to know what's happening across their team and relay that context up and down the chain. In a remote first company where work is already machine readable, AI can build and maintain that picture continuously. What's being built, what's blocked, where resources are allocated, what's working and what isn't. That's the information the hierarchy used to carry. The company world model carries it instead."

Alex Hanna: Absolutely bananas. And it's throwing me off a little bit- this is just a tiny nit- that I thought Block was a positive word here, and they're saying "what's blocked." I don't know. It's weird.

Emily M. Bender: Oh yeah, no, I missed that.

Alex Hanna: It's a weird thing. And what magical thinking, though, that there is any kind of ingestion of data, that the data are not problematic in their own right- and they're gonna get into that in another paragraph- and that there is some kind of way that there is a decision-making structure that is happening, that is probably done in a way here where it is trying to push workers as hard as possible. So maybe not as different as Jack Dorsey would do in his own actuality, but now it is then suggesting that it's taking him out of the equation.

Emily M. Bender: Maybe. Although where they get to is weirder than that. But two things I see here. One is the absolute commitment to wall-to-wall surveillance of your workers. Which Amazon, for example, has perfected, and Meta, which we'll see in a moment. But also the idea that the artifacts are the work, and that just having that data is enough to know what went on. And this is what we see over and over again, when people take the synthetic text extruding machines and say, "This can do your job, 'cause your job is just to produce artifacts that look like this."

Alex Hanna: Yeah. And it's really helpful to read the next paragraph, too, the one that says "people lie on surveys." Because then it is getting at what is, I think, one of these main epistemic beliefs in this, and I think machine learning and something like computational social science, where they say, "People lie on surveys. They ignore ads. They abandon carts. But when they spend, save, send, borrow, or repay, that's the truth." And like, what a fucked up thing to say.

Emily M. Bender: And also, what an uber capitalist way of looking at the world. If money doesn't touch it, it's not real.

Alex Hanna: Yeah, 100%. Or if it's a fiscal, financial transaction, that's the truth of what people's, whatever, interstates are, or whatever. And they continue, "Every transaction is a fact about someone's life. Block can see both sides of millions of these transactions every day, the buyer through Cash App and the seller through Square, plus the operational data running from the merchant's business. That gives the customer world model something rare: a per customer, per merchant understanding of a financial reality built from honest signal that compounds. The richer the signal, the better the model. The better the model, the more transactions. The more transactions, the richer the signal." And it's just, to me this is fucking wild insofar as it's ignoring the idea of the artifact itself as being something that structures the data signals, the people that interact with these technologies because of real world constraints, and what that is supposed to say about the customer. So for instance, a lot of people- I'm thinking about Jay Cunningham's work on Cash App in certain Black communities. And Cash App will often be used a lot because a lot of folks are unbanked. And so then you can get on Cash App without some of the other existing types of identification. And so, okay, Cash App is there, and in some ways is actually quite parasitic, because it's taking advantage of a community that's unbanked. You're just basically structuring the world, and then you're saying, "I know the truth of these people's financial situation, because they're using cash." I'm like, no, you don't know shit. There's so much that's going on there, even in financial transactions. And it even gets back to your initial example about people using Square at the farmers' market. Yeah, people use Square at the farmers' market because no one carries cash anymore. But cash also has a certain way of structuring these types of interactions about how we interact with local producers, and it is not as traceable, everything like this.

Emily M. Bender: Yeah, absolutely. The other thing that I wanna say in here is that this is where this document starts getting really muddled between, "here is what Block has access to as our data moat," and, "here is how we think companies should be organized," and they're not differentiating. They're munging that all together, which is wild. So they talk about, in order to build this new kind of company, you need capabilities, a world model, an intelligence layer, and interfaces. So the intelligence layer we need to dig into, because it sets up this next thing, which is bananas. So they say, "Third, an intelligence layer. This is what composes capabilities into solutions for specific customers at specific moments and delivers them proactively. A restaurant's cash flow is tightening ahead of a seasonal dip the model has seen before. The intelligence layer composes a short-term loan from the lending capability, adjusts the repayment schedule using the payments capability, and surfaces it to the merchant before they even think to look for financing. A Cash App user's spending pattern shifts in a way the model associates with a move to a new city. The intelligence layer composes a new direct deposit setup, a Cash App card with boosted categories for their new neighborhood, and a savings goal calibrated to their updated income. No product manager decided to build either solution. The capabilities existed. The intelligence layer recognized the moment and composed them."

Alex Hanna: Oof. Yeah, this is wild stuff. There's so much here. If you are suggesting that there is going to be automated product development because of the way that their people have spending patterns, how long will it take for some kind of a LLM, a synthetic media generation machine to be like, "Hey, you seem to be a little cash poor. How about this short-term loan with really exorbitant interest rates? Don't worry, you can repay it back, you're gonna repay it back in a few months." Or the same thing, like somebody is moving- "Wow, it looks like you're moving to a certain area. Why don't you engage in these other types of local economies that are making very particular presumptions about how you will spend, and what your behavior is gonna be like?" So it's, yeah, Emily, what you're saying is that it's this fantasy of this complete capture. And this is at the customer side. They're not even talking about the customer related to what the product manager is doing, hence, no need for this person that has any kind of input into what is being built. So just complete fantasy stuff.

Emily M. Bender: Yeah, absolutely. So a couple great comments from twoangstroms. The first one is, "Oh, great, invisible opt-in to possibly predatory loans." And then also, "New definition of man-in-the-middle attack. We slip you into a fee relationship with someone and take a cut. See also, crypto brokers." So, what they're imagining here is evil enough as it is, and then they're imagining it being put together automatically because of their intelligence layer, so they don't need a product manager. So with that intelligence layer, we get to how they imagine their company working without people. So they say, "If this is what the company builds, then the question becomes what do the people do? The org structure follows from this, and it inverts the traditional picture. In a conventional company, the intelligence is spread throughout the people and the hierarchy routes it. In this model, the intelligence lives in the system. The people are on the edge. The edge is where the action is." Can we talk for a second about the way they're talking about intelligence here? It is so gross.

Alex Hanna: Like a virus, I don't... What is intelligence supposed to be here?

Emily M. Bender: Some sort of substance that it's bad when it's spread out through the people and better when it's all clumped, glommed together into one big something.

Alex Hanna: Yeah. It's intelligence, it has become something that lives in the brain rather than something that is just a green goo of neurons that seems to live at the heart of the company or something.

Emily M. Bender: Definitely lights up from time to time in different places, right?

Alex Hanna: I'm thinking about the robot boss in Portal or something.

Emily M. Bender: I was thinking of the giant brain in Madeleine L'Engle's series.

Alex Hanna: Oh, Wrinkle in Time?

Emily M. Bender: Wrinkle in Time, yeah. That's controlling all the people, so the kids are all bouncing their balls at the same time.

Alex Hanna: Yeah, it's a very interesting, very bizarre, very science fiction version of this. Oh, you know what I was actually thinking about? I was thinking about the weird tumor that is in Legend of Zelda: Ocarina of Time in the water level. Sorry, I'm just making very deep video game cuts today.

Emily M. Bender: All right. Let's keep going here.

Alex Hanna: Yeah. abstract_tesseract knows the name, Morpha.

Emily M. Bender: We are talking to your people, Alex.

Alex Hanna: Yes, incredible. I knew the chat was gonna get this.

Emily M. Bender: Yeah. All right, you wanna read this next bit?

Alex Hanna: Which one? Okay, so, "The edge is where the intelligence makes contact with reality. People reach into places the model can't go yet-" so they're getting into those nooks and crannies. That's me editorializing. "They sense things the model can't perceive: intuition, opinionated direction, cultural context, trust dynamics, the feeling in a room." Sure. "They make the calls the model shouldn't make on its own, especially ethical decisions, novel situations, and the high-stakes moments where the cost of being wrong is existential."

Emily M. Bender: So those low-cost loans, or whatever, maybe high-cost loans, that's not an ethical decision? Sorry, I couldn't resist.

Alex Hanna: No, I was gonna say the exact same thing. "A world model that can't touch the world is just a database." But I thought you had access to the world through your objective data points. The metaphor is just falling apart. I don't need to read the rest of this.

Emily M. Bender: Yeah, I love that line, though. "A world model that can't touch the world is just a database." Yes, you have a database. It's a big database. And?

Alex Hanna: Yeah. Yes.

Emily M. Bender: All right, so then they have, talking about how we get rid of having to go "up and down a chain of command." There are three roles: "Individual contributors who build and operate capabilities that model the intelligence layer and the interfaces. They are deep specialists and experts in a specific layer of the system. The world model provides the context that a manager used to provide, so ICs can make decisions about their layer without waiting to be told what to do." And this is the part that I thought was entirely bananas. So when you're talking about this as command and control, but also, it's a synthetic text extruding machine. So now you've got a company of people who are just gonna be, in quotes, "taking orders" from, in quotes, "the intelligence," and going off and doing their own thing with no actual coordination.

Alex Hanna: Yeah. That's fantastic management. It's absolutely great management. "Go do this," and then they go off and do this, and don't be surprised if what people do is not what you wanna do. But also, it's this kind of world where the vision is one in such that actions are so specialized- going back to Taylorism- and they're so atomized, and they're so de-skilled, that there's no wiggle room for messing it up. This is the vision of people who manage data work platforms. This is just so specific. But I'm like, okay, there's actually so much there, and that is skilled work, and you're trying to suggest that it's not.

Emily M. Bender: Yeah, absolutely. And I don't know that we need to go into the other roles here, but there's a little song for you in the comments, Alex, if you want it. coliar_mauve, "The world model's connected..."

Alex Hanna: Oh, "The world model's connected to the blockchain. The blockchain's connected to the galaxy brain!" That's very good. Shout out.

Emily M. Bender: Scans perfectly, too. Thank you for that. All right, so this is all bananas. "Block is in the early stages of this transition. It'll be a difficult one, and parts of it will likely break before they work." No shit. But not before they work. They'll just break.

Alex Hanna: Yeah. I want to read these two and then maybe transition to the next artifact. 'Cause they say, "We're writing about it now because we believe every company will eventually need to confront the same question we did. What does your company understand that is genuinely hard to understand, and is that understanding getting deeper every day? If the answer is nothing, AI is just a cost optimization story. You cut headcount, improve margins for a few quarters, and eventually get absorbed by something smarter. If the answer is deep, AI doesn't augment your company. It reveals what your company actually is." And I should say again, I want to reiterate, this is a message that they sent out publicly when they laid off many people. Actually, I want to see how many layoffs that they actually did. And yeah, so March 2nd, 2026, laid off 40% of their workforce. So 4,000 people. And so they lay them off sometime in February, and then release this absolutely bananas- I don't know what to call this, it's not a manifesto- weird philosophical tract. And then to read this after you got laid off, this is just like, what an insulting thing to read.

Emily M. Bender: Yeah, absolutely. A little bit of fun from the chat here. So coliar_mauve says, "What if it reveals that your company is a weird garbage cult?" And twoangstroms replies, "Weird garbage cult at scale, please." And yeah, absolutely. So this is just a really pointed example of, "we're laying off a bunch of people to juice our stock price, and we're gonna say we're doing it because we've figured out how to get the AI-" or the intelligence, as they're saying here- "to do all of the work so we need fewer people." And it's never true, but it's dressed up, and here it is so clearly speaking to investors, too, in this whole thing.

Alex Hanna: Oh, yeah. And twoangstroms says, "I heard Dave Karp say that one lie that keeps getting pushed is that 'hard things are easy.'" And I'm like, no, hard things are hard. And you're trying to techno solution it, and you're not going to actually find anything deeper. There's no there there. You're just being a weirdo.

Emily M. Bender: Yeah, and the question that I have is, when this starts falling apart because they have laid off so much of their workforce and they can't actually maintain their product anymore, who's gonna be bearing the brunt of that? And my guess is that we're gonna see a lot of it falling on those small merchants who all of a sudden can't get transactions processed.

Alex Hanna: Well, my sense is that it's going to look a lot like post-Musk Twitter, where it's going to be people who are these critical engineers who might be on types of visas that keep them wedded to a certain workplace, and that it's going to just put more and more stress on those people, and it's going to be this kind of Musk model of firms, where you are effectively going to put more work on those individuals and then hire contractors that then have to fill in when the product just absolutely doesn't function as you need it to.

Emily M. Bender: Yeah. All right. With those bleak predictions, we're gonna go to something else bleak. So how are companies actually using, in quotes, "AI" in connection with layoffs? We have a bunch of new information now from a lawsuit against Meta. The headline here, so the sticker is "AI and Layoffs," and it's interesting to me that that has now become a presumably not one-off sticker in Ars Technica, which is where this is from. This is by Jon Brodkin on July 14th, 2026, and the headline is, "Lawsuit claims Meta's layoff decisions were made by AI, not humans." Subhead, "Meta denies using AI to terminate workers with disabilities and medical problems."

Alex Hanna: Yeah. Very dark stuff here. And really honestly, again, not surprising. So it behooves us to start with the first two para- we should read a lot of this, but the first two paragraphs really hammer it home. "Meta's AI-fueled layoffs of 8,000 employees targeted workers with disabilities and those who took protected medical or family leaves, alleged a lawsuit filed by 26 employees who were selected for termination. Meta used internal AI tools to select employees for layoffs, according to the complaint filed yesterday by 26, quote, 'Doe' plaintiffs in US District Court for the Northern District of California." And then the quote here from the lawsuit says, quote, "'Meta did not assemble the termination list through the considered judgment of managers who knew the work. Instead, Meta used a constellation of internal artificial intelligence systems, including a system referred to internally as, quote, "MetaMate," employee-trained, quote, "second brain agents," keystroke and activity monitoring data, AI token usage dashboards, and algorithmically assisted performance ranking and calibration to score, rank, and select employees for inclusion on the list." Wow.

Emily M. Bender: Yeah. So what's interesting to me here is that the headline suggests that the "AI," in quotes, was designed specifically to target people with disabilities or people on medical leave or family medical leave. But getting into it further, basically, we've known for a while that Meta was doing all kinds of worker surveillance. So this thing about mouse and keyboard usage, for example, and then of course the whole tokenmaxxing thing. And then whatever algorithm they were using for ranking people to consider for termination was basically using those metrics. So if you've been on disability leave, if you've been on parental leave and so on, your numbers are going to be lower. And the allegations here are that Meta did not account for that at all, therefore basically violating labor law, I think.

Alex Hanna: Yeah, and I think the thing to also highlight are the ways in which workers with disabilities are going to be using technology differently. Are going to be using computer-human interaction interfaces much differently than people without disabilities. And that's obviously discrimination. Who could have guessed that was going to happen if you were measuring these, trying to infer some kind of elements of productivity from keystrokes and mouse movements. So just some of the most... Ugh, it's just infuriating that this is the kind of stuff that was even internally justifiable, and that it's, of course, bearing in this lawsuit.

Emily M. Bender: Yeah. Absolutely. And there's relatively little commentary from Meta in this. Maybe we should try to find it. Yeah, so subhead here, "Meta says people, not AI, made layoff decisions." "Meta says that people made the layoff decisions. Quote, 'These claims lack merit and are not based on facts. Workforce management and organizational decisions were and are made by people, not AI,' Meta said in a statement provided to Ars today. Meta did not provide any other comment on the lawsuit. The lawsuit alleged that Meta management did not take steps to adjust scores for employees who took leave or who requested reasonable accommodations for disabilities." So basically we've got Meta saying- and there's probably some human in the loop who looked at the output and, in quotes, "made the decision." And that's the basis on which they're saying that. But I really hope they do not win this lawsuit.

Alex Hanna: Yeah. And there's a comment from supportlivemusic in the chat, who says, "I worked for a place that tracked my keystrokes, and I'm physically disabled and use voice-to-text. I had to explain to my boss why my, quote, 'productivity score' was likely so low." Ugh, that's so awful. I'm really sorry.

Emily M. Bender: That really is terrible. And we don't see that much unionization yet in the tech sector, but this is the kind of thing that I think you would really want to be getting workplace protections against. This degree of surveillance just can't be good.

Alex Hanna: I should also say that this is the kind of stuff that there is draft legislation to try to protect against. So in California last session, there was a no robo bosses bill that I think actually made it out of the Assembly and Senate, and then was vetoed by Governor Gavin Newsom. And so now there are efforts to try to pass similar types of legislation on the city level in different jurisdictions. And one thing that I think that's important for that legislation is basically to ensure that there's no way that there is this kind of accountability dodge that Meta is making here, effectively, that the decisions were made by people, as if it was fully automated. They sent the layoffs via some kind of a message or operation in Workday. It should be like, even if there's a quote-unquote, "human in the loop," if you're making these determinations, those are not valid. Those cannot be used according to these types of regulations.

Emily M. Bender: Yes. So good regulation to be pushing for, I think, and hopefully one day you will no longer be saddled with Governor Gavin Useless.

Alex Hanna: God, I hope so, but it's not looking great. Our options for our next governor are not fantastic. So there's some pretty terrible stuff here.

Emily M. Bender: Yeah, do you want to keep on going, too? So there's terrible stuff here, but also I think interesting connection to what went before. So the terrible headline, "Employee told layoff, quote, 'day before her water broke-'" told of layoff. "The May 2026 layoffs came after an internal memo in which Chief People Officer Janelle Gale told staff that Meta would cut about 10% of employees and stop hiring for about 6,000 open roles. Quote, 'We're doing this as part of our continued effort to run the company more efficiently and to allow us to offset the other investments we're making,' Gale's memo said." And then the next part, so, "The lawsuit said that, quote, 'Meta announced these cuts even as it reported record revenue the prior month and committed to spending between 125 billion-'" with a B- "'and 145 billion, more than double its 2025 expenditure on artificial intelligence in 2026, prompting employees to question why the job cuts were necessary." So this whole thing of, "we are doing this because of AI," isn't that they really think they can automate these people's work. It's that they've decided they have to put all of this capital into data centers, basically. And then also there's a whole thing about reassigning employees to AI projects. "And that's why we really- I'm sorry, it's so bad, we have to do these layoffs." It's like, no you don't!

Alex Hanna: Yeah. This is part of the play, right? It's the, "oh, we're gonna save a bunch of money." And then, the revenue's only gonna keep on going up if they can cut those jobs, right? Ugh.

Emily M. Bender: Revenue should be able to be independent, but net receipts... And I'm sorry about all these ads. They're annoying. But do you wanna do this last paragraph here? 'Cause they managed to add on a... it's not even the end, but it gets to some of those really terrible details.

Alex Hanna: Yeah. So, "They also alleged violations of various laws imposed by states and the District of Columbia. For example, an update to California's Fair Employment and Housing Act, quote, 'forbids the use of an automated decision system that produces disparate impact discrimination on the basis of disability or sex, including pregnancy,' the lawsuit said."

Emily M. Bender: Yeah. And then I think the last thing that's of interest here to me is that this is not a class action lawsuit because the employment contracts that people sign prohibit class action lawsuits.

Alex Hanna: Oof.

Emily M. Bender: But the 26 plaintiffs are working together because they all want to have this recalculation of the employee scores. They're trying to get an injunction, basically keep their jobs, while this goes through the courts.

Alex Hanna: Yep. And I guess the coda here is that these companies are already doing massive discrimination according to protected characteristics. And I will say, as a former Google employee, the number of lawsuits that I have been implicated in based on race and gender is at least three. They didn't even need the mathy maths to be doing this kind of discrimination. But they are certainly supercharging it.

Emily M. Bender: Yes. And I just wanna end by shouting out the reporter here, Jon Brodkin, whose tag is senior IT reporter, because the people working on this beat who are actually looking at what's happening as opposed to just transcribing the breathless announcements from the big tech companies, are few and far between. And when we were planning this out, Alex, you were saying this is more of a hell artifact than a hype artifact. And it's true. This is good reporting about a terrible situation, as opposed to the sorts of things we usually use as main course.

Alex Hanna: Yeah, absolutely.

Emily M. Bender: All right. So musical or non-musical for the transition?

Alex Hanna: What did we do last time? I forgot. I think we didn't do musical last time, and we already have music- we already warmed up with our MLMs.

Emily M. Bender: Yes. So you ready to do this musically, then?

Alex Hanna: Let's do it.

Emily M. Bender: Okay. And do you have a genre that you want to use?

Alex Hanna: Last time I think we did Scatman. And by the way, our producer Ozzy got us a desk copy, for quote-unquote "research," of a recent biography of the Scatman, who actually had a fantastic life. He actually had a stutter, and he turned it into an electro Europop career. So shout out to the Scatman. But let's do something else. I don't have a preference.

Emily M. Bender: Okay. I want sort of '60s psychedelic slow, all right? Because-

Alex Hanna: What's that?

Emily M. Bender: You are the intelligence trying to access the edge through the people.

Alex Hanna: What's a '60s psychedelic slow... this may be a generational thing. Are you thinking about Grateful Dead? 'Cause I don't know what any of that kind of music-

Emily M. Bender: I'm reaching back before my time, too. I'm picturing... I have more of a visual here, of lava lamps and the visual aesthetic that goes with that. But-

Alex Hanna: I don't know if I could do that without having just a massive reverb on a Stratocaster or anything. So, not quite sure how to do that, but maybe, I'm trying to think of- Oh, I got it. Hold on. I'm thinking of "Purple haze." Like, "purple haze up in the sky." Purple- oh, purple brains. Excuse me while I intelligence the sky. Da-now-now, da-now-now, da-now-now. I don't know the lyrics to that song. Anyways, that was not one I was prepared for. "Purple Bayes." Thank you, in the chat. Very good. Zubenelgenubi17 said that.

Emily M. Bender: Yeah. I think we said, Zubenelgenubi was what we decided that one was. Okay, so here we are in Fresh AI Hell. I've got some nice reporting in Scientific American by Joseph Howlett. August 6th, 2026. Headline is, "OpenAI's Latest Math Breakthroughs Commit Research Misconduct, Experts Say." So OpenAI released this thing, saying, "we've got 10 AI-generated results to hard unsolved math problems." And then you've got folks like a mathematician at Yeshiva University named Steven Miller, who says, "'They are running roughshod over the work of others who came before them in a deliberate way. It seems completely systematic to me, and it points to research misconduct.'" So basically, OpenAI was claiming, "no one's been making progress on these for a decade," and then the actual solutions they're publishing pair together recent work by people without citation to that.

Alex Hanna: Yeah. So I'm glad that they're calling them out here.

Emily M. Bender: Yeah, and I just want to remind people, we talked about this last week, but there's this wonderful thing called the Leiden Declaration on AI and Mathematics, where a bunch of mathematicians say, "look, here's what our field has to say about this, and here's what it means to be doing research in math and how we do this as a community." And so it's a nice counterpoint to what's happening there. Okay. You can have this one, Alex.

Alex Hanna: Yeah, so this one is on Bluesky, and the original skeet is an article from 6ABC. And it's by friend of the pod Hypervisible, who says, "'Researchers said the goal of CAMI-'" which is the name of the, tool, I'm assuming- "'is not to replace physicians, but to provide clinics, schools, and families with a faster diagnostic tool.'" And the chyron on the news thing is "AI-assisted autism screening," and the headline in the link says, "University of Pennsylvania researchers develop AI tool to help speed autism evaluations." And then the person who had quote-tweeted it is Miss Mira, Born Fool, who says, "This is the most obvious eugenics ploy I've ever seen."

Emily M. Bender: Yeah. And so it's, I like how the sort of subhead here says, "Researchers said the goal of the program is not to replace physicians, but to provide clinics, schools, and families with a faster diagnostic tool." It's like, this is actually not the only problem with this kind of tech, right?

Alex Hanna: Exactly.

Emily M. Bender: It is a labor issue, but it is not only a labor issue.

Alex Hanna: Yeah, the problem is that we aren't having a backlog of- maybe this is one thing that they're running into, but there's so much wrong with this.

Emily M. Bender: Yeah, absolutely. And somewhere down in this article, I think it said- No, I'm thinking of a different article, but another medical screening thing saying, "previously you had to rely on talking to patients and understanding their behavior and talking to caregivers, and now we can..." It's like, oh, come on. Ugh. Okay. So this one, this is from Forbes, and the sticker is "Innovation and Consumer Tech," published December 3rd of last year, so a little bit old here for Fresh AI Hell, but I still think this is hilarious. The journalist- not journalist, sorry- the senior contributor is John Koetsier, and their tag is "Journalist, analyst, author, podcaster." And the headline is, "New Data: AI Is Almost Green Compared To Netflix, Zoom, YouTube."

Alex Hanna: This is such a funny thing. I'm like, yes, those things 100%, just great uses of carbon as well. And they also have this image of these nuclear cooling stacks, which is not really, you know, aiding any fears.

Emily M. Bender: Yes. It's a pretty sunset, though. And the caption on the photo says, "AI is taking more and more power, but many other digital activities are growing as well and take even more." And then there's this list further down, where it's all apparently quantified. "Here is the relative impact to our environment of common digital activities. YouTube or Netflix, comma, one hour, parentheses, HD," and this is "approximately .12 kilowatt hours, right arrow, 42 grams of CO2, tied for the dirtiest single activity in the study." And then "text-to-video generation, 6 to 10 seconds." And then it's glossed as roughly the same as an hour-long Zoom call. "Zoom, one hour. Short email, comma, no attachment, and this one is approximately 0.0133 kilowatt hours, which is the same as 4.7 grams of CO2." And then a little bit of text that says, "One email is tiny. Billions per day are not." Next one, "AI image generation." Smaller than that, goes to "one gram of CO2." There is no, "one image is tiny, billions per day are not." Like-

Alex Hanna: Yeah, this is ridiculous- Oh, and there's also something, if you scroll a little down, too, where they actually say, they say "Google search or AI chatbot prompt." And then they give a metric for that. But I'm like, oh, so now you are... So first off, do you have an estimate of what a Google search is prior to the AI overviews? Because are they making an equivalence there? Because that is- and yeah, the chats also say it's just that the framing is just so weird. And I'm kinda like, who is this guy again?

Emily M. Bender: Oh yeah, before we get to looking at him, I just wanna point out that after the Google search or AI chatbot prompt, there's a separate category of two Gemini prompts, which is apparently much more efficient than other AI chatbots- according to what, I don't know.

Alex Hanna: Who did- is there, what is this study? I'm curious on-

Emily M. Bender: Oh, is there a link for it? So, "TRG Data Centers."

Alex Hanna: TRG, okay. What is this organization?

Emily M. Bender: Am I gonna be able to get to it?

Alex Hanna: It's going to an archive page. But that's because the original-

Emily M. Bender: Original's in the archive, yeah.

Alex Hanna: Yours is in the archive, yeah.

Emily M. Bender: I might not be able to pull it out. But yeah.

Alex Hanna: Yes. Source probably needed.

Emily M. Bender: BoxoMcFoxo, "Did he use Andy Masley's calculator to work this out?"

Alex Hanna: Probably, yeah.

Emily M. Bender: Yeah. Ah, okay. So you can have this one.

Alex Hanna: Okay. So this is from, oh gosh, from Futurism. The sticker is "Hoosier Daddy," and the Hoosier as in the referent of a resident of Indiana. And the title is, "Google Data Center Announces Plans To Pave Over Protected Wetlands." And it's happening in Fort Wayne, Indiana, which is why it's Hoosier Daddy. And then the subhead, which is a quote, "'It was determined that meeting the project purpose and completely avoiding water resources was not feasible.'" This is Joe Wilkins, journalist, August 7th. And yeah, then there's a picture of some wetlands, which is, I think, just a stock photo. And the lede, "Residents of Fort Wayne, Indiana, may have to say goodbye to over four acres of federally protected wetland, at least if Google has its way. New reporting by local publication WANE-" W-A-N-E- "revealed that a company called Hatchworks LLC, a data center developer for the tech giant..." For which tech giant?

Emily M. Bender: Google, I think.

Alex Hanna: Ah, okay. Google, the prior. "...Is looking to fill in several acres of federal wetlands as part of the third phase of Project Zodiac, a $2 billion hyperscale data center project." Yes, the Zodiac Killer indeed.

Emily M. Bender: Yeah. And magidin in the chat says, "'If the choice is between our fantasy and your environment, we're going with our fantasy.'" In quotes, of course. Yeah, just, gross. Okay, one more bad news before our chasers. This isn't exactly bad news. This is just hilarious. So this is a skeet on Bluesky, posted by Veerender Singh Jubbal, but it is a screencap of a tweet by Sam Altman over on X. Altman writes, "Cool use case of ChatGPT work I heard last night. Connect your family calendars and explain your kids' interests. Every morning for the drive to school, have it make a podcast that talks about one kid's soccer game that afternoon, one kid's upcoming birthday, some news, et cetera." And Alex Hirsch replies to this, "What if you just talk to your children?"

Alex Hanna: Yeah. This is also Sam Altman who once said, "I can't imagine being a parent without using ChatGPT." And he's either had a kid or is, his family is about to.

Emily M. Bender: I'm just looking at this avatar, 'cause I can't make that look like Sam Altman. So maybe this is fake.

Alex Hanna: No, but it's Sam Altman 'cause this is sama. I'll double check, but yeah.

Emily M. Bender: Yeah, yeah. That's probably real. Who knows? Okay. Onto some chasers.

Alex Hanna: Yes. So the first one, this is by Zoe Tillman, who I believe is a journalist, on Bluesky. And the quote is, "'The government cannot escape liability for DOGE's work by scapegoating ChatGPT,'" end quote. And the text says, "A US judge slammed DOGE's use of AI in a decision today, blocking the termination of $100 million in humanities grants, finding DOGE lacked authority and unlawfully based decisions on race and viewpoint." And then it's linking to, looks like a court filing. So, good news there.

Emily M. Bender: Yes, I do like that. And that first sentence of the court filing quote is, "The government cannot escape liability for DOGE's work by scapegoating ChatGPT." So basically, if you're gonna use the system and use its output as your rationale, you are responsible for what it said, which is some nice accountability. All right. abstract_tesseract, "Uppences, they're coming." Yes. And here we've got some more uppences. This is another skeet, by Jesse Felder this time. This is from August 9th, with a link to something from The Information. And it's this map of the US with a heading that says, "Banned Together: A wave of data center bans and moratoria hit the US this year." And then the quote that the poster on Bluesky pulled out says, "'In July alone, more than 150 towns and counties passed temporary or permanent bans on data centers, many adopted in emergency meetings. This brings the nationwide total to more than 500.'"

Alex Hanna: Yeah, it's very nice. And there's a lot coming. And I know in the full article, I believe they also called out specifically New York State, which did pass a statewide moratorium, and also Texas, which I think is starting to be more aggressive. I don't think they have a statewide moratorium, but weirdly enough, they are not very happy.

Emily M. Bender: Yeah. Now, I forget where I was listening to this 'cause I've been listening to a lot of podcasts about data center activism. But somebody was pointing out that it matters how long the moratorium is. 30 days, 60 days isn't gonna do it. This needs to be really long enough that you can then start working on legislation and other regulations that really pin down people's rights to air, power, water, and so on.

Alex Hanna: Yes, 100%. All right, cool. That's it for this week. Our theme song is by Toby Menon. Graphic design by Naomi Pleasure-Park. Production by Ozzy Llinas Goodman. And thanks as always to the Distributed AI Research Institute. If you like this show, you can support us in so many ways. Order "The AI Con" at thecon.ai or wherever you get your books, or request it at your local library.

Emily M. Bender: But wait, there's more. Rate and review us on your podcast app, subscribe to the Mystery AI Hype Theater 3000 newsletter on Buttondown for more anti-hype analysis, or donate to DAIR at dair-institute.org. You can find our merch store there, too. That's dair-institute.org. You can find video versions of our podcast episodes on Peertube, and you can watch and comment on the show while it's happening live on our Twitch stream. That's twitch.tv/dair_institute. Again, that's dair_institute. I'm Emily M. Bender.

Alex Hanna: And I'm Alex Hanna. Stay out of AI Hell, y'all.

People on this episode

Alex Hanna

Co-host

Emily M. Bender

Co-host