Ideas - 33,000 Horsepower Gamechanger | The Greatest Numbers of All Time

Episode Date: June 25, 2026

There is nothing random about featuring 33,000 in our number series. It's very powerful. So much so, that the number put millions of horses out of work. Inventor James Watt used “33,000 foot-pounds a minute" to measure the capabilities of a horse when trying to market his new and improved steam engine. The engine was a big success, saving horses from the drudgery of manual labour. Now, a similar process is underway with artificial intelligence — but are we the horses, or the steam engine? *This episode is part of our series, The Greatest Numbers of All Time.For more in the series:Listen to The Curse of 13Listen to 12 is SublimeListen to 27 Club LoreGuest in this episode:Stephanie Dick is a historian of mathematics, technology, computing, and AI, and an assistant professor at Simon Fraser University.

You know that feeling when you reach the end of a really good true crime series? You want to know more, more about the people involved, where the case is now, and what it's like behind the scenes. I get that. I'm Kathleen Goldhar and on my podcast Crime Story, I speak with the leading storytellers of true crime to dig deeper into the cases we all just can't stop thinking about. Find crime story wherever you get your podcasts. This is a CBC podcast. Around 1780, a Scottish engineer needed a number to represent a horse. After watching, measuring, and calculating, that number turned out to be 33,000.

Welcome to Ideas, I'm Nala Ayyed, and to another installment in our series, exploring the most significant numbers in the world. The series examines unique, revolutionary, or as in this case, a number that doesn't get much fanfare. The number, 33,000. A critical number in the Industrial Revolution. A defining number of energy, labor, and power. And a step on the way to redefining work, intelligence, life, and ourselves. 33,000 opens into... the remaking of the world in the context of industrialization.

The people have always tried to understand natural systems by analogy with artificial ones because we've made them. The world is itself pretty violent and surprising sometimes. We ourselves have some of that structure as well. We're calling this episode 33,000 and the machines we think we are from producer Matthew Laysen Rider. Matthew, why of all numbers did you land on 33,000? So as you mentioned, 33,000 is an important number in the industrial revolution.

And there's another potential revolution going on in the world right now with artificial intelligence. So what I'd like to do is look at that number, how it changed horses into machines about 250 years ago, and what that process means for humans and humans. human intelligence in the 21st century. So help me understand how exactly does 33,000 relate to horses? So in the 18th century, it was the number that inventor James Watt gave to a brand new unit of measurement horsepower. So James Watt, Scottish inventor, sometimes credited with inventing the steam engine, but he didn't invent it. He did vastly improve it. And that improved steam

mentioned drove the industrial revolution, mechanized the workplace, helped create the modern world, everything we learned in social studies nine. But that number tells a story. And it's about how machines give people new ways to think, starting with what it means to be a horse and today what it means to be a human. And that story starts around 1780. And that particular number, one horsepower, 33,000, emerges in the context of the industrial revolution in Europe at a moment when really incredible transformations in how work is being done are happening. And work was done before these machines by horses and people and wagons and wheels. So we needed a way to think about these really new technologies in terms of the world we already understood. This is Stephanie Dick. She's a historian of mathematics, technology, computing, and AI, and an assistant professor at Simon Fraser University in British Columbia. And she sees 33,000 as a kind of hinge moment. Before horsepower, you had a horse and after you had an abstract idea of what a horse can do.

So James Watt, who was an innovator in the development of steam engines, starts watching what horses are doing and trying to measure the rate at which horses can do work. How fast can they move, you know, some number of pounds in a minute, for example. And through that observation, he comes up with this idea of horsepower, this abstract concept, which is the power required essentially to lift 33,000 pounds one foot in one minute, which is what he observed horses to be able to do. But the reason I think the number is interesting and exciting is because it represents this longstanding desire that we have to come up with these abstract concepts that remove things we care about like work from their material specificity. Material specificity, Matthew, what does that mean?

So this is a subtle point, but it becomes really important for thinking about what comes after. So on the one hand, actual steam engines replace actual horses, right? thing physically changes in the world. On the other, there's a new way of thinking about work that removes it from the physical world. Work becomes a number that can be scaled, managed, marketed instead of like a hairy beast that has to be fed and brushed and shod and so on. The term Dick uses for that stuff is messiness. When Watt was marketing the steam engine to factories, part of the deal was he was going to help get rid of all that. messiness. Of course, factory owners want to get rid of messiness. What kind of messiness,

though, is he talking about? Well, horses are messy things. Right before horsepower, if you said, hey, how much work can a horse do? The answer has all kinds of variables. Well, I don't know, is it a big horse or a small horse? Is it well fed? Is it thirsty? How much work can a horse do is a messy question? Now that you've got horsepower, no mess. So all of a sudden, work has nothing to do with the muscles and the grains and the wheels and the wheelbarrows and the mud that horses work in, it becomes this abstract idea. All right, that abstract idea and mechanization of the factory totally change everything, create the wealthy and prosperous world that we live in today. Job done. And then, around the same

time watt is abstracting work from the horse, others begin asking if we can do the very same thing but for the human mind. Brain power instead of horsepower. Yes. If you can 33,000 work, can you 33,000 thought? There's another piece of this story that leads into the work that I've been doing as a historian of technology and computation and mathematics to try to understand what is happening to our idea of intelligence at that same moment.

And there are people in Europe at this time, like mathematician Charles Babbage or a French mathematician, Gaspar de Prony, who are looking at the factory. They're looking at how work is being reimagined through these machines. And they are asking a question, could we do for the mind what these machines have done for the body? is intelligence something like work that could be abstracted away from its material conditions. And to do this, these thinkers, Duproni and Babbage, working at the end of the 1700s, borrow the logic of the factory. Maybe the same way that work could be understood independently of horses and muscles and grains and mud and ropes. Intelligence could be understood as something independent of the huge. human body, the human brain, the human face, the human hands, to try to reimagine through

industrial terms a lot of things about ourselves. Charles Babbage is famous today for his designs for early mechanical computers. And Gaspar de Prone, he's working in France after the revolution. And he wants to produce these massive mathematical tables, logarithmic tables and trigonometric tables. And like a factory instead of a craftsman, instead of having... having one brilliant guy solve mathematical problems with creative thinking, maybe you can reduce it all, find the smallest unit of thought, break it up into different steps. And DeProni figured out a way to take one type of mathematical equation, namely a logarithmic equation, and solve it in principle by doing a whole bunch of additions and subtractions in a row. So DeProni thought this was achieving

what the factory had achieved, but for thinking reimagined as work. Here, thinking is thought of as a type of work that needs to be done and it's broken down in the same way basket weaving was broken down into more basic elementary steps, in this case, basic arithmetic. And at that moment, which is actually the moment the word computer comes into being, computer comes into being as a word to describe these people who could never, be hired to carry out the additions and subtractions. And they did not have the mathematics education that aristocratic philosophers and theologians

would have had. These would have been lower paid, lower educated workers who are now being hired to do work. So if the ability to pull 33,000 foot pounds a minute was the mechanical essence of a horse, maybe the ability to add and subtract is the. mechanical essence of a mind. And there's a story about DeProni's human computer Nala, and that is that it was made of hairdressers. That's a surprise.

Hairdressers? Yeah, so this is a colorful story told over the years, but truth is unclear. But so under the monarchy, there's huge demand for hairdressers all the time. All those rich people need their hair done and their wigs made. So there's tons of women who had work as hairdressers. Then the revolution comes along, and the monarchs are. and the aristocracy are overthrown, there are masses of

unemployed hairdressers. There's just not so much hair to do. Or heads to put wigs on. Right. And so DeProni needs people to do the pluses and minuses. Hey, let's hire all the hairdressers. And there you go. You have a computer

made of hairdressers. It's a bit of a tall tale, but it represents a couple of things. Neither of them very kind to hairdressers, actually. But one is, this is when boring, rote computation work becomes seen as women's work. It's a view that lasts up to the 1960s or so. But also that this is work that kind of anyone can do. Thinking has been abstracted from the brain and the mind, removed from its material specificity and broken down

into easy, repetitive tasks and mechanized. Still, not mechanized enough for some. So there's a really famous quote, and I can't say whether it's true or not, but famously or apocryphally, we have a quote from Babbage, looking at Gaspar de Prone's tables of logarithms that had been produced by this first exercise in labor-divided mathematical calculation and said, I wish to God these calculations had been carried out by steam. And by that he meant, so I see this exercise in having these people do. all the additions and subtractions, but there were still tons of errors, there were still tons of, you know, zeros didn't get carried, and this, you know, hundreds of them, thousands of them. And he thought, what if instead these human computers could be replaced by steam engine powered machinery? If only, right? If only there was a machine that could do the thinking for us. So this desire

to do away with the fallibility of humans goes way, way back. So just like, Horses get grumpy and tired and dump waste everywhere, humans are messy too. We make mistakes. We have bad days. We get sad and distracted and sick. That messiness starts to be seen as the enemy of scientific rationality, basically, where all of those things that we might have imagined actually at some level constitute the human condition, right? That we are messy. Our days are all different. There's so much going on with us that all shape what we're capable of, that was seen as a problem to be solved. And Nala, just to show you this continuity of thought from Babbage to today, let me play you

some recent tape. Okay. And this is from the Joe Rogan show. You've got 20 AI bots that are all as good as the best programmer in the world that are doing exactly what you tell them to do on every project you've ever wanted to do. And they're running 24-7. And the only thing you have to do is be there every 10 minutes to be able to give them feedback on what they're doing. Oh, my God. So this is Mark Andresen, inventor, software engineer,

venture capitalist, something of an analog to Babbage today, maybe. He's also on U.S. President Trump's Council of Advisors on Science and Technology. So here's him briefly explaining to Joe Rogan as a business owner, the benefits of artificial intelligence. You tell a normal person, you tell, you know, you hire somebody over here, and you tell them you want a screen display and you want it to be an animated version of your thing you got back here. Okay, they spend two weeks doing it. They bring it to you. They animate it. It's like, okay, that's pretty good, but I actually want the whole thing to be whatever, purple and green. And they spend a week doing that and they come back. And you're like, I actually preferred the old version. The guy gets like pissed at you

because he's like, I just wasted my time. The bot's like, no problem. You know, no sweat. Like whatever you want. Right. And by the way, never gets drunk. It never gets sick, never gets high. Right. Never gets depressed because his girlfriend broke up with him. Never files HR complaints. Right, right. Mark Andreessen, speaking on the Joe Rogan experience. What if there was this perfect intelligence that was just rational deduction without all of that humaneness getting in the way? And I've always thought, like, how alienating at some profound level to be told that what you are fundamentally at some level is a huge problem that needs to be solved in the workplace in decision making.

and that we can only be as good as workers or knowers as we can be inhuman and suppress all of those facets about us. But to a large extent, that is what the history of industrialization, scientific management, and ultimately artificial intelligence are born from is this desire to generalize away all of these remarkable things about human beings and the ways we work and learn and know. And it's a fascinating story that's born out of this moment of abstraction and generalization across really messy material realities. There's another part of this story, and that is, it seems like whenever we invent a new thing, we compare ourselves to that thing. The idea that maybe humans are a math factory is part of a long trend of human thought. technology gives us new ways to think about human life. In the ancient world, philosophers reached for the most sophisticated technology they knew, pulleys, levers, and wheels, and found a new language for the body.

I really think it goes all the way back as far as you want to go. I mean, I think you find that in antiquity. You find it throughout the early modern period. So I think it goes all the way back. and, you know, our current sort of machinery that we turn to is information machinery, but that's pretty recent. This is Jessica Riskin. She is a historian of science at Stanford University in California. She's the author of the books The Power of Life and The Restless Clock, both of which examine and challenge a recurring metaphor in science academia and out in the public that life can be explained mechanically. I think people have always tried to understand that.

natural systems by analogy with artificial ones because we've made them. There's a term for this, and it's called the mechanistic worldview. And a really good example is, in the 1600s in Europe, clocks start becoming commonplace household items. And before you know it, we're made of clockwork. So Descartes, 17th century French philosopher, he sort of declared that as the right methodology for philosophy. You know, we should try to understand the natural world and the cosmos in the same terms that a clock maker understands a clock. That is moving parts, material moving parts that you can, you know, catalog, you can see how they move and how they cause one another to move. And that represented for him a kind of total intelligibility.

And so I think that, you know, it's certainly in the modern period that has a lot to do with it. You know, we really, really understand there are no, we don't have to assume any occult forces or any kind of, any kind of mysteries. We know what the parts of a clock are, and we can take them apart, look at them, put them back together. And so, you know, that's the model of modern science. I think the kind of core model. A machine is knowable. You can understand every part of it. Passion and memory and imagination are messy concepts. But if you can reduce them to some kind of mechanical metaphor, perhaps they are understandable. These functions, including passion, memory and imagination,

follow from the mere arrangement of the machine's organs, every bit as naturally as the movements of a clock. And Descartes wasn't just using machines as examples. He argued that at its core, the human body is a machine. And after the clock,

the factory, the steam engine, in the late 19th century, came the telegraph. And all of a sudden, philosophers and scientists find a new understanding, for the mind. Nerves have often and not unsuitably been compared to telegraph wires. According to the different kinds of apparatus with which we provide its terminations,

we can send telegraphic dispatches, ring bells, move magnets, and so on. In the early 20th century, the telephone replaces the telegraph, and now the brain is a switchboard. The brain is no more than a kind of central telephonic exchange. Its purpose is to allow communication. It really constitutes a center where the elements get into relation with this or that mechanism. In the mid-20th century, transistors and carbon resistors and circuit boards come along, and that's what we are. Regarding the anatomy of the nervous system, as if it were a wiring diagram, and the physiology of the neuron, as if it were a component relay of a computing machine,

we shall describe the brain in terms thoroughly familiar. The electrical engineer. And naturally, now that we live in a world of powerful computers and artificial intelligence, we have a brand new metaphor. Maybe our minds are like artificial intelligence. One of Riskin's students recently asked her, how does he know his brain doesn't work just like chat GPT? He did.

Yes, he did say that. And yet, I don't know, my response to him was basically, but you're you. You clearly have a stuff. that's the difference between you and chat chippy T is that you have you in there. And, you know, the fact that we can't explain how that works. I agree. We certainly cannot explain how selfhood works. But to deny that it exists because we can't explain how it works seems to me to be kind of crazy. And to acknowledge that something exists that we don't fully understand how it works is not, to me, a kind of mystical departure from science. It's, right? I mean, to accept that. that there are limits to our science is not to become a mystic. So it just seems crazy to deny the existence of anything we don't understand. There's a little problem here. Remember, progress relied on reducing a horse in all its organic messiness

to simply a machine for moving 33,000 foot pounds a minute. Good for factory owners, good for consumers, probably good for horses too. reducing the human to clocks and telegraphs and switchboards was useful. It helped scientists develop working models of the body and mind. But all of those models left something out. And before artificial intelligence is everywhere, embedded in everything, maybe we need to ask, what if there's more to intelligence than just what we can model? Right.

I mean, it's a little bit like the old parable. of the person, you know, looking under the lamppost for his keys because that's where the light is. I mean, it's true that that's intelligible, but it doesn't necessarily cover all the possibilities in the world. 33,000 is the subject of this episode of ideas and part of our ongoing series about numbers. Loud, important numbers like 13 and 3, and the quiet ones that might not get the attention they deserve. You can find the rest of our series on cbc.ca.ca. and wherever you get your podcasts. This is Ideas. I'm Nala Ayat.

A history of the United States in 100 Objects is a brand new podcast from 99% Invisible and BBC Studios. Each week, we're looking at a different object from across American history with a unique story to tell about who we've been, what we've built, and what we've allowed ourselves to forget. Some of these objects are well-neutral. known, many are not, but all of them carry the story of how we got to this moment. Find a history

of the United States and 100 objects on the 99% of visible feed wherever you get your podcasts. We're calling this episode 33,000 and the machines we think we are from producer Matthew Laysen Rider. Hello, Matthew. Hi, Nella. So we've gone from The Horse is a number, to the world is a clock, to the brain is a switchboard, and maybe any. AI. Where do we go next? I want to look a little more deeply at that computer metaphor. So as you mentioned, we started with the idea that a horse can be abstracted into a machine for pulling 33,000 foot pounds a minute. And then maybe you could abstract thinking into something really simple, like a machine for doing pluses and minuses. Well, in the mid-20th century, just at the dawn of computers, as we

sort of know them today, another idea came about, which is maybe you could reduce enormously complex parts of life and society and politics into machine terms as well. And to set the scene, I'm going to throw another number at you here. 99. You and a show set them free at the break of dawn to one by one back at base, box in the software, flashed. The message Something's out there Floating in the summer sky 99 red balloons go by

Classic This is the song 999 zigg Luftballoons Also known as 99 red balloons Yes, from 1983 And the West German band Nina or Naina One of many,

Believe it or not, New Wave dance songs About Nuclear Anxiety during the Cold War Right. And in the song, balloons set off a nuclear apocalypse. Yeah, so the song is about some people minding their own business, playing with some helium balloons. They let them go and off they fly into the sky. But, uh-oh, some radar system picks them up and determines, oh no, the enemy is attacking.

And here's the follow-up from the English version. So this whole automated system kicks in. And pre-made plans are set in motion, and planes are launched and bombs are dropped, and the whole world is destroyed. All because of some red balloons. Yeah, so then it's Mad Max, and the song ends with finding some red balloons in the dust. So this was a big cultural fear in the night. So this was a big cultural fear in the 1980s that humans had been so removed from the process of war that computers could make a mistake and trigger an apocalypse. You can also think about the movie War Games, right, with Matthew Broderick and Ali Sheedy.

And part of that was true in a sense. There was a lot of research during the Cold War into how to automate not just mechanical systems, but strategic and diplomatic and intelligence systems. It is in that context at that moment that the first so-called artificial intelligence program is developed. Here's SFU historian of mathematics and AI, Stephanie Dick. I can talk about this maybe most concretely in terms of mutually assured destruction. You won't bomb me because I'll bomb you and neither of us can handle the consequences of that. But in practice, mutually assured destruction is really complicated if you are going to reliably retaliate against a nuclear attack. You have to either know that it's going to happen or be able

to respond after it happens. You need to know where all of your enemy's nuclear resources are so that you can monitor them and be able to give an order to retaliate after you've been struck, which means you need communications infrastructure in place, you need military decision-making in place. And there was a lot of concern that the weak link in that entire network or system of safety and support and surveillance and decision making were the people. And that we might not be able to respond fast enough or clear-headed enough. We might not be able to process enough information in order to be trusted with the maintenance of safety in this nuclear world. A lot of defense research in the post-war period are sort of. of, it's like a laboratory for trying to feel in control of an uncertain future.

And so many of those approaches to research and planning and decision-making involve abstracting intelligence, judgment, decisions away from their messy human and material realities in order to be something like horsepower that we feel like we can get a handle on and optimize and manage in rational ways. A perfect intelligence, one that doesn't. get scared, get angry, no panic, no fatigue. And in the 1950s, a lot of this research comes together at a place called the Rand Corporation. And that's a think tank born out of the U.S. Air Force to try and predict the nature of

future warfare from an abstract mathematical perspective. And the idea was, if you want to build a replacement for human decision making, you need a model of how humans make decisions? There were two men there, Alan Newell and Herbert Simon, who are two of the largest names in early artificial intelligence, and they were both working at Rand in the 1950s. And they proposed that the mind and the computer were both what they called symbolic information processing systems.

So for them, both the computer and the mind takes symbolic information as input, so like words, numbers, numbers, pictures of the world, representations of things. And then in the mind, according to them, we manipulate that symbolic information according to a set of rules. They were insistent that human thinking is a rule-bound exercise. And that when we do intelligent behavior, like solving math problems or formulating a judgment or making an evidence-based decision of some kind, could be represented as rules for the manipulation of symbolic information. This is the computer metaphor formalized. The mind and the computer have the same sort of rules.

And then more than that, they said if you can write those rules down and make them explicit, and you can figure out what are the rules for symbolic information processing that a human is engaging in when they make a medical diagnosis or write a story or, you know, interpret the Bible or whatever it is we're doing. If we can write down the sequence of basic symbolic information processing that are involved, we would have a program that describes the human behavior and that could be run on a computer such that it would produce that same intelligent behavior in the same way. And Matthew, did it work? Did they write that human decision-making program? They did, in a sense, and it's a pretty cute story. Herbert Simon sort of really famously said that he and Alan Newell wrote a computer program that was intelligent over Christmas, but he simulated it in operation using his family. So he had his kids, his wife, sort of holding little pieces of paper and moving to different positions to sort of simulate the symbolic information processing that Newell and Simon believed constituted logical theorem proving, which was the first program. that they developed. So here are these two guys. They're in the middle of this Cold War crucible

of anxiety, control sciences, this desire to abstract away from a messy and uncontrollable, scary world. And it's an eraser of the human from the question of intelligence in that moment, which I find so fascinating and so in keeping with the same move of horsepower to to see if we can imagine work as something that can be carried out, not by horses and carts, but by anything. Any hybrid mix of human animal machine can carry out work. Similarly, any mix of human animal machine, child paper bucket could be intelligent so long as it was capable of doing the appropriate abstract symbolic information processing operations. So listening to this from the present standpoint, it's hard not to ask this question, which is, did Newell and Simon create a kind of artificial intelligence?

Yes and no. So to be clear, their work was groundbreaking. The programs, they went on to build, proved mathematical theorems that had gone unproven by humans. They helped show that computers could be treated as machines for reasoning, not just calculation, but as a complete, model of human intelligence, it didn't scale into what we'd think of as AI today. Something resisted the model. Some people start to say maybe the path to intelligence and machines doesn't go through us. Perhaps our own intelligence is resistant to this type of abstract, rule-bound automation. And in my academic work, I've studied a lot of these critics who just reject the rejection of human messiness. And we are not intelligent in spite of being tired or upset or angry or hungry or wrong or irrational or religious or superstitious. You know, what if our intelligence is a product of the fact that we are that too?

That it's negotiating being both rational and irrational, being both, you know, tired and awake, excited in it. Maybe it's the dichotomy in us and the navigation of all these facets of the human condition that leads to our intelligence. And if that's true, then any attempt to distill our knowledge or our thinking into some kind of rule-bound general abstract procedure that the computer could carry out instead of us would fail to capture us at all. So now as successful as their work was, in a sense, they were doing what had been done for hundreds of years. approaching the latest technological marvel and trying to use it to capture the essence of human experience. Right. Just like with the clock, then the telegraph, the telephone. Right. So in the, in the 50s, the computer joins the parade and is arguably still a dominant metaphor for how the mind works. And it's baked into our language in a way. We have mental processes, right, and cognitive

bandwidth. We talk about the brain as if it's the hardware and the mind as if it's the software. And we do it the other way around too, right? Computers have memory and sleep and your desktop talks to your printer. In psychology and neuroscience, it's a thing called the computational metaphor. How accurate or inaccurate or even useful is that metaphor? Hotly debated among academics. So on the one hand, The computational metaphor is a good way to get the public and students to start thinking about the brain. How do its components add up to a mind? On the other, maybe it gives us the totally wrong impression of what makes us human. Science training requires metaphors.

And so the metaphors really were sort of always the thing I came up against when trying to make sense of the research that I was sharing that I was trying to answer and, respond to from other folks. Tell me how the mind is representing. Tell me how the mind is processing. This is Damien Kelty Stephen. He is an associate professor of psychology at the State University of New York. He researches perception and cognition and once ruffled some feathers by making a blog post comparing those components of the brain, not to a computer, but a milkshake. What I was trying to get out with the milkshake was that, in other words, way to look at the brain and ask whether it's a computer is to see, well, what is it got? It's got mostly fat water and sugar. And I was perhaps frustrated or exasperated one day. And I was like,

this is not sound like something that you make a computer out of. So Kelty Stephen thinks the computer metaphor for the brain, long, popular in academia, may have hit its limit. The brain is not really a circuit board. We're not constantly calculating. And besides, even things without brains can do some amazing things. There are these conversations where lots of people will say when they read or hear about some of my stuff, they're like, oh, yeah, of course, the mind's not a computer. But it's the, it's the sort of the best way to talk to a lot of people who aren't specialists to help them understand what's going on.

It's a wonderful way to start getting people into it. Like, I see it, I feel it, I know it. But it's not hard for some of the thoughtful students to go, like, well, wait a second. like if that's really how it is, then what about this other? And like, and they start bringing to me the textbook and they start finding all these sort of like,

you know, unplugged wires that seem to be straggling out from, you know, this or that part of the explanation. And they're like, what, how does that work? And I'm like, right. You know, there's more to this.

There is a little exercise you can do that Kelty, Stephen argues, shows one way we are unlike a computer. It's something we can do that doesn't fit the computer metaphor. One of the classic computer challenges out there is the traveling salesperson or salesman problem where you have a bunch of cities or landmarks that you have to visit and you have limited gas money or airfare money and you have to sort of visit all the places, not visit

any of them twice and you have to end up back where you started. It's particularly difficult for computers if they don't get either brute force and or good hints. So imagine you're looking at a map. right, and you've got to deliver a package each to 20 cities and get back to the town you started in. What's the shortest route? And finding a way for computers to quickly find the optimal solution has been a notoriously difficult problem in mathematics and computer science. And they're better than this now, but the most straightforward way for a computer to do it and guarantee 100% optimal results is a strategy called brute force. Brute force tries every possible combination of roots and then simply picks the shortest one. The problem is with 20

stops, there are about 60 quadrillion different routes. And that's going to take your computer very, very, very long time to do. So finding shortcuts and strategies has been a goal in computer science. And over time, mathematicians have come up with all sorts of creative ways to limit the amount of work a computer has to do to get there. Now, on the other hand, maybe surprisingly, maybe not, humans are very, very good at this problem, seemingly without doing much work at all. To me, it does seem surprising.

How do humans do it? Well, it's unknown exactly, but it doesn't seem like we're doing the same things a computer does, even when a computer has shortcuts and hints. So with around 20 stops, research says the average person can quickly draw a route, between all the stops and back again, and we can do it within 1% of optimal.

And Kelty Stephen's point is it doesn't seem like our brains are computing really quickly or calculating or using systemic, abstract, logical rules. We're just kind of good at thinking about space. If we're thinking about space with our brains, we're doing it better. And the funny thing is, like, people have the sense that our brains and our nervous systems are working really,

fast and they're actually not. So they're one of the slower systems that we have. They're leaky. They have synapses that sort of lose the momentum. There is the electrical transmission that's very fast and zippy, but then that gets punctuated multiple times by these gaps where there's just neurotransmitter floating between neurons. The computer goes much, much faster. And for all that speed, it doesn't optimize a spatial trajectory all that well. So what makes us better at space than computers?

I mean, we have a body. One label for the kind of work that I do is embodied cognition. And you can find, actually, that amoeba that don't have any neurons inside them because they're unicellular, they can solve mazes very efficiently, and they don't have a brain to speak of. I mean, what they have and what we have is sort of a body that's heavily invested, in the rest of the world around it. And it's also, it's a body full of internal texture. So it's not that sort of we are large, inflatable plastic dolls with the brain inside and we can

bounce around. There's a lot of stuff that we incorporate through a whole lifetime of exchanges with the world leading up to that moment when we have to solve that trajectory problem. Kelty Stephen's point isn't just that, hey, humans are particularly good at moving around. it's that thinking is often not what we imagine when we describe it in computer terms. Now, there is an argument. Maybe we just don't know enough about the brain yet, right? Maybe like the watchmaker, if we keep taking it apart and finding the cogs and wheels in

smaller and smaller detail, we'll eventually understand the programming. But we're not just a brain in a body like a computer in a case. And maybe the mind isn't bound by the brain at all and includes our feet. and our hands and our bodies and our relationship with three-dimensional space, that the way we think is more fluid, more milkshaky, or in Stephanie Dick's terminology, messier. And in every attempt to make abstract models of the mind, we eventually learn they don't entirely capture reality

because reality is messy. And yet here we are, Matthew, in a world where artificial intelligence is on everyone's phone, people are building relationships with it, people are losing their jobs to it. What about it makes it seem to succeed at replacing humans if humans aren't like computers? So that is the big historic shift going on right now. That's what makes modern AI so unique and revolutionary, because modern AI is trained on a representation of that messiness. So large language models, or LLMs, the kind of AI that's prominent today, aren't like those early attempts

where they constructed formal rules about human decision-making. Modern AI makes patterns out of an ocean of messy human language. So here's Dick again. The hope was, can we just bypass people altogether? Like, what if we don't have to tell the computer what rules to follow in order to be intelligent or to have expertise?

What if we just give a whole bunch of data, to the computer and let it come up with its own rules to follow for prediction and classification by identifying patterns and correlations in all of this messy data. So on the one hand, contemporary AI is a bit of a capitulation because we had aspired to capture our own intelligence in these abstract ways that could be automated and it didn't work. We were the bottleneck. We were so resistant to any kind of this modeling that originally. race as the messiness that might in fact constitute us. But it led instead to this very different paradigm of intelligence as really messy pattern

matching. So that's the twist. Early AI tried to find a logical way of thinking. LLMs are trained on the traces representations of messy human thinking. And they blew everyone away, me included. there's no way I would have predicted that a statistical model of language use would have the capacities that LLMs appear to have. Because LLMs didn't just come away with the ability to converse. They appeared to have like expertise in physics or expertise in medicine or ethical reasoning capability.

Like so much seemed to be there in our language use because we write about all of these things, right? But again, maybe we don't write everything down. We have to ask these questions about what is not reflected in our writing to be learned. And related to our theme, I say all the time, like, AI has no access to the world. None. All it has access to are the data that we give it. And that data clearly are powerful and generative and have been able to impart a lot of information. but our data are also always incomplete and partial,

and they reflect certain people's values and not others. And so again, LLMs, they represent a different approach to the cultivation of intelligence and machines, bypassing models of ourselves, but giving it all of our expression and saying, what can you learn about, what can you learn? So Nala, I want to help Stephanie Dick tie all of this back to horses steam and 33,000. So bear with me. So what we've done is essentially the same thing. We've come up

with an abstract idea of what makes a thing a thing, then used it to replace the thing. So we came up with an abstract idea of what a horse could do, pull weights, in other words, and used it to replace the horse. Which was overall a good thing for everyone. And now we've come up with an abstract idea, a recreation of messy. human intelligence and are using it as a stand-in for actual messy human intelligence. When we talk about the risks of AI, mostly we've been talking about who will it displace, who benefits economically, who loses out, how do people find work if robots are doing it all, right? So steam engines replace the horse and artificial intelligence will replace us,

what happens to us? Those are all good concerns, but Dick points to another risk, and that is that like we've done before so many times, we confuse a model of human intelligence with the reality. Well, I am very concerned about the redefinitions of intelligence that are underway at this moment. And I hope that anyone who looks at the history of AI can recognize right away that intelligence is and has been more than one. thing. We've defined so many different intelligences into being throughout history. Intelligence as reasoning. Intelligence as knowing, social intelligence, emotional intelligence. And I think the challenge we all face right now is being able to appreciate, recognize, cultivate, and then have relationships

between all these different forms of intelligence. But I worry that's not what will happen because it's so alluring to be spared some toil in this burnt-out modernity that I think people will offload a lot of work to the tool. If the idea is we constantly define ourselves by the machines we create, then intelligence becomes not a messy and fluid human capability, but simply the things an AI does. Think of the qualities that can make LLMs feel intelligent. they've got confidence. They don't question themselves.

They don't become passionate. They don't argue with you. In fact, they tell you how right you are, how good you are, how smart you are. And so what happens to our understanding of intelligence over time? Well, the risk is that we think intelligence should be frictionless, that it should answer quickly, that it should accommodate endlessly and rarely push back. I really fear the loss of struggle because, it's from that struggle that everything that I value in my life has come.

And to invite a suite of tools that are designed and explicitly marketed as saving us from struggle, that sounds to me like the design of a really meaningless future as well as a profoundly deskilled one. I learned something about myself, which is that I don't love to read polished writing from my students. I love to read writing from my students where I see them trying to say something. Like they're wrestling with language in trying to articulate an idea that is new to them. And that is the opposite of how LLM's right. The most exciting moments in the history of ideas, I think, are the moments where people are really disagreeing about something. They have a fundamental,

incompatible disagreement about how to go forward in an academic discipline, in a policy project, in something. And if everyone is deferring to the same kinds of LLMs to support their decision-making, their knowledge production, their learning, their work, I think a lot of that pluralism gets lost. We shift our attention and understanding towards the tools we use. And if everyone is using the same tools, we're going to lose a lot of the pluralism that I think makes democracy function, that makes knowledge generate and keeps us on our toes in terms of not settling into autopilot about our lives. Because it's so convenient to not be challenged by the messiness outside.

And that's the importance, I think, Nala, of thinking about the horse, the steam engine, and 33,000. The steam engine doesn't mimic a horse. It mimics an abstraction of what a horse can do. It doesn't model its nature. So the question with AI isn't whether it models something real about intelligence. It clearly does. It finds patterns in our strange human thinking enough to mimic the chaotic way that we communicate. The question is what happens when we use it, thinking it genuinely captures the messy things that make real human intelligence. Things like disagreement in pursuit of truth or the good faith push and pull of democracy. Right? Those are horse things, not steam engine things. And so we've come up with a model of

intelligence and we seem to have this trend of believing that we are the models instead of these fluid, living, messy, difficult things that don't fully survive being modeled and abstracted and made mechanical. Long live struggle. Long live the messy human struggle, Nala. Thank you so much, Matthew. That was Ideas producer Matthew Laysen Rider, with 33,000 and the machines we think we are. This episode is part of our on-com.

number series. Special thanks to Stephanie Dick, Associate Professor at Simon Fraser University. You can find the full series at cbc.ca.ca.com ideas and wherever you get your podcasts. Technical production by Emily Carvezio. Our web producer is Lisa Ayuso. Senior producer Nicola Luxchich. Greg Kelly is the executive producer of ideas, and I'm Nala Ayyad.

For more CBC podcasts, go to CBC. c.ca slash podcasts.