AI has become a broad cultural concept. Those same two letters could refer to Instagram animal video slop, Claude code, a science-fiction antagonist, or the algorithm that doctors use to spot skin cancers. It’s a big bucket, and one that consistently treats the technology as both autonomous and opaque.
In his new book Humans of AI, anthropologist Joseph Wilson pulls back the curtain on that narrative to expose the most important part of the AI story: the people. Wilson embedded himself in a Toronto AI chipmaking lab and global data annotation networks, and chatted with prominent boosters and critics to understand the human forces behind the technology.
BetaKit managing editor Sarah Rieger sat down with Wilson to chat about why he wanted to study the customs and culture of AI developers, the hidden labour behind the tech, and what he hopes readers will learn from his studies.
The following interview has been edited for clarity and length.
Other than journalists, most people writing about AI are either in tech or are critics of tech. Why did you want to bring an anthropologist’s perspective to this topic?
There’s this feeling in science that science is somehow outside of culture. That it tries not to be tainted by the messy social world. But of course it is; everything is.
AI in particular was this intoxicating mix of scientific investigation, public-facing apps, and rhetoric about what it is to be human. So there were these big social questions, and these big questions of science and technology that were nicely entangled, and that’s the confusion that anthropologists love.
You write that there’s nothing artificial about artificial intelligence. What do you think are some of the risks of treating AI as alien, or omnipotent, or even anthropomorphized?
The first is that if we mistake the metaphor for the thing, we start to assign capacity for ethical decision-making. We start to assign characteristics to these systems that they’re just not capable of. We start to think, even subconsciously, that if it’s modelled after a human, then maybe it can do these other human things, including feel emotion, or make ethical decisions, or self-reflect.
It’s a really interesting misinterpretation of how it works technically, right? You often hear people say, “Oh, AI is always learning,” but that’s really not how these models are built: they’re trained, and then they’re locked.
The second thing that worries me about anthropomorphizing these systems too much is that it, by design, obfuscates and erases the very real human labor that goes into making the systems. It’s people that did this.
So it’s a human stack, not just a tech stack. One of the layers in that human stack that you write about is ghost workers. Who are they?
Ghost work is a term coined by another anthropologist, Mary Gray, who worked with a computational sociologist to look at labour patterns around the world. She looked at how people, mostly in the Global South, are working behind the scenes to make it look like our apps and our interfaces run smoothly and automatically. These are people who are doing the sometimes very unpleasant work of filtering toxic data, like violence or sexual imagery. Or they are tagging data so that AI can get better at image detection and video detection for self-driving cars and security videos.
All the big tech companies use these systems because they essentially chop up this data work into small tasks, and people get paid by the task. So it’s kind of like an assembly-line model. That work is precarious at best, and potentially really traumatizing, and because they don’t have job security, a lot of them don’t have healthcare; they sign NDAs to say that they’re not even allowed to talk about what they’re doing.
You spent 18 months embedded in a Toronto company making AI chips that you refer to in the book under the pseudonym NextChipAI. I’m going to skip past what I really want to do, which is to try to guess which company that is. Can you tell me what it was like to be in that environment?
It was so interesting because AI is sold not only as autonomous, but also as this kind of incorporeal cloud. The cloud metaphor is apt here. We know with this furor over data centres that there’s very real physical infrastructure that needs to be built to run these big models.
RELATED: America loves Canadian chips
So, I was an intern. This company makes computer chips. They’re about the size of a soda cracker, and they are designed to run the neural nets that generative AI is composed of. I just spent time hanging out in the kitchen and by the coffee machine and in meetings to try and figure out how these engineers collaborated, how they went from an idea on paper to this almost magical object with billions of transistors etched into the surface with light.
Some of the physical details of the work they were doing really struck me. You detail the type of sand that’s used in silicon, or how the engineers would smell the chips to sniff out errors in production.
It’s embodied labour; It’s embodied cognition. These guys are using their whole bodies and their expertise. It’s quite elegant to watch it all unfold, this collaborative dance that they do that results in this remarkable product.
It sounds like that elegance was mirrored a bit in the mindset of the engineers you observed: the focus on skepticism and the art of problem-solving. You describe feeling a disconnect between witnessing that and the sometimes bluster of people hawking the AI products.
It’s funny. The engineers that I worked with, and I don’t think this is universal, because in Silicon Valley there’s a different kind of tone, but here in Toronto anyway, and especially the engineers that are working on the hardware, they have little to no interest in talking about AI as this potentially conscious or even intelligent system.
They know how the sausage is made, and they are skeptical of the claims made by the marketing department. But they need to raise money, and the way you raise money is you tell stories. But often they felt hurt that they get pushed out of the story of AI. Their work, which is hard and complex—these guys are working weekends and evenings and don’t see their families—their work is written out.
Do you think the obfuscation of the work that goes into AI is intentional?
Yeah. And not in a conspiratorial way. The process of science often tends to erase labour because the idea is you want the product to stand on its own, right? But with AI, there’s an added layer where it’s purposefully done to sell these systems as autonomous and as capable of superhuman intelligence. That’s part of the pitch: No human needed.
A term I was introduced to while reading your book is the concept of “criti-hype.” Can you explain what that means?
The idea is that there are two arguments. First, that AI can lead to this glorious future where problems are solved by these super-intelligent systems. The flip side is that we could lose control of it, and it could lead to the destruction of the human race.
On the surface, they seem like opposites. But I argue, and this concept of criti-hype shows, that criticizing technology as something that will end the human race is admitting it is powerful enough to do so.
RELATED: Waving the AI bubble red flags
So these criticisms from Anthropic and OpenAI and this extinction narrative, it just serves to whip the conversation into an even bigger frenzy. The people who have called for a pause on AI development still have their IPO in the works.
I don’t think it’s necessarily 100-percent cynical or ill-intentioned—I think some of them genuinely believe it, and they’re kind of worried—but it has that effect. These guys are entrepreneurs, but they’re not experts in social impacts.
You mention an attendee at an AI conference grumbling that “we didn’t have to think about ethics before all this.” Which is so funny, because ethics have always been a part of tech. But it feels like perhaps awareness of public interest in tech ethics wasn’t this acute until now?
Yeah. I think that caught a lot of them off guard. You hear engineers talk all the time about their obsession with puzzles and really hard problems. But the ethics problem is, for many, too hard. It’s essentially that these are questions without answers, or at least without clean, clear, identifiable answers.
It would be nice if it was that tidy.
Well, it would be nice and easy for the platforms. But part of what makes life interesting is the dynamism. Our friends and families surprise, and anger, and delight us.
What’s one thing you hope readers will take away from this book?
A sense of how AI is made: intentionally, by a baffling diversity of people distributed across the world.
The more fundamental thing I hope my book can do is ease people’s fear and their anxiety about this thing they don’t often understand. This complex phenomenon was made by people.
People not understanding is part of the sales pitch. It’s like, “You’re scared. You don’t understand. Let us do it. We know better.”
That’s not a fun place to be, sitting in fear. We can demand better. We can demand more transparency around data practices, around labour practices, around regulations. We can move forward together here and figure out what we wanna do intentionally.
Feature image courtesy Joseph Wilson.