JDN 2456233 EDT 14:09.
A couple weeks ago a friend of mine said something that really annoyed me. I tried to argue with him, but he wouldn't listen.
No, it wasn't about politics (though he does have far-right libertarian politics, so it's not a bad guess). It was about reductionism. What he said was this:
"Someday we will reduce psychology and economics to neuroscience, and there will be no need for psychologists or economists."
First of all, keep in mind I'm a behavioral economist, so this is like saying "Your life's work is completely worthless"; that's bound to make anyone upset. But occasionally it's true; I mean, come on, postmodernist literary theory?
But more importantly, in this case it's just not true. There is no plausible universe in which psychology and economics will be so reduced to neuroscience that we don't need psychologists or economists anymore.
Does this mean that there is a mysterious "ontological emergence" within psychology, that our minds are somehow "irreducible"? That certainly sounds pretty bizarre, and I think it's what my friend thought I was saying.
But it's not. Psychology and economics will be in some sense "reduced" to neuroscience; I would actually say they will be unified with neuroscience. The gaps between them will close and they will all become part of a broader universal theory. And no, there's nothing that's mysteriously "irreducible". Everything that happens in an economy is made up of people making decisions, and everything in the human mind is made up of neurons and chemical interactions. In that sense, I am a reductionist.
The problem is, my friend wasn't just talking about that kind of reductionism. He was talking about a much stronger sense, one in which the reduction renders the higher level completely irrelevant. Once we understand neurons, we don't need to worry about minds! Once we understand minds, we don't need to worry about economic systems!
To see why this doesn't work, consider the most successful reduction in the history of science, the reduction of thermodynamics to statistical mechanics. We have now fully explained everything there is to be explained about heat and temperature and entropy in terms of the energy and interactions of quantum particles. Reductionism successful.
But at the same time, if you want to know how hot something is, you don't try to measure all 10^23 quantum particles bouncing around in it. You simply don't have to, and couldn't if you wanted to. And when engineers make stress-strain curves based on the temperature of a material, they don't try to account for the wavefunction of every particle in the substance--again, they don't have to, and couldn't if they wanted to.
Imagine someone saying, "Someday we will reduce computer science to electrical engineering, so there will be no need for software programmers." Would that make sense to you? Would that seem like a very logical, scientific thing to say?
I hope not, because we have reduced computer science to electrical engineering--in fact, computer science was basically built up from electrical engineering, so there really was no reduction to be done. And yet, we still need software programmers; in fact, it's a very challenging, rewarding, and lucrative career, one of the few that weathered the 2008-2012 depression virtually intact.
Why? Because computers are complicated, and they are made up of billions of tiny pieces interacting in very complex ways. You can't possibly keep track of every 1 or 0 on the entire CPU. Indeed, to do so would require an enormously larger CPU, and then you wouldn't be able to keep track of that, and so on ad infinitum. (The size requirements grow combinatorically, so for a processor with a billion components, there are 2^(10^9) or about 10^(3*10^8) interactions to track.) You need to understand high-level laws like "classes" and "loops" and "functions" in order to make any sense of what's going on inside a computer (and even then, it's hard).
Now consider a human brain. At least right now, a human brain is about 1000 times more complex than even the most advanced computer. Eventually that may not be true--probably sooner than most people think; that's only 10 more doublings, or 15 years by Moore's Law--but still, it's true right now. And since a computer is already so complex we could never understand it in terms of its constituent parts, the human brain is even worse. We can't even understand the human brain at the high level of modules and functions, though we are beginning to--and that's what cognitive science is all about.
Now consider the world economy, which contains billions of people--and hence, billions of human brains. We may never make a computer as complex as the world economy--Moore's Law has got to run up against physical limits eventually. Sure, if you do the exponents, it seems like we'll be there in 50 years; but this is like arguing that because your credit card interest rate is 24%, in 50 years your $1000 balance will become $47 million. You'll pay it off or be bankrupt long before that, and likewise we're already running up against some pretty fundamental limits of computing power. Quantum computers will help, understanding the structure of the brain will help; but a trillion times more powerful than we currently have? I think that's pretty unlikely.
And that would just be enough to simulate the world economy (basically, make a copy of it). That would be about as useful as... having the world economy in front of you. The only advantage I can see is the ability to tweak variables for experiments in ways that don't actually affect people's lives. This could certainly be useful. But basically, it's a black box that you poke until it does what you want.
To actually understand the world economy at this level of detail, you'd need to be able to process the 2^(10^15*10^9) combinations (10^15 synapses per person, 10^9 people). This would require a computer able to process on the order of 10^(10^23) calculations, and feel free to give it a million years to work, you haven't reduced that number in any noticeable way. 10^(10^23) per second and 10^(10^23) per million years are basically the same processing speed; there are only 3*10^13 seconds in a million years, so it's the difference between 10^(10^23) and 10^(10^23 - 3*10^13). Even if Moore's Law somehow miraculously transcended the laws of time and space to have components smaller than a Planck length processing at faster than a Planck time, it would still take about 10^24 years to make a computer that fast. There are in fact enough years left before all the black holes evaporate... we think. Yay?
Fortunately, we don't need to do that. It's completely pointless to try to understand the world economy in terms of individual neurons. That's ridiculous. Even at the level of individual people seems pretty silly. We can model much better by looking for high level principles--like supply and demand, information asymmetry, externalities, and so on. These simplifications allow us to make computationally tractable models, which can then be applied usefully (albeit not perfectly) to the real world.
Sometimes I do think our models are too simple. We try to solve exactly instead of using the computational power we already have. Most equilibrium models literally have 2 firms, 2 goods, and 2 consumers; this can be solved exactly, but come on! At least do 10 firms, 100 goods, 1000 consumers. You're missing out on a lot of important interactions if you only look at 2 of each. And yes, 2^1000 is still too big to deal with, but make some more simplifications and you can probably get it down to 2^30, which is only about 10 billion calculations, well within what a modern computer can achieve. That means that instead of just simulating the system we can actually take it apart to some extent, look at the pieces, understand how it all fits together. At least potentially.
Are economies made of people? Of course. Are minds made of neurons? Yes. But it's not only stupid and pointless, but outright impossible in most cases to study a complex system at the level of its constituent parts. And to say that a system "just is" those parts is also pretty silly; the parts by themselves would not be the system, you have to actually put them together, in, you know, a system. A pile of steel girders is not a bridge.
Anyone who thinks this sort of thing, that reductionism means economics will be replaced by neuroscience, clearly doesn't understand complexity. They are not really doing reductionism, they are doing what Daniel Dennett calls greedy reductionism. They're not trying to understand the system in terms of its parts; they're trying to replace the system with its parts. It's like having a pile of steel girders and shouting, "look at my beautiful bridge!" A system is made of parts and interactions; and yes, you need the parts to have the interactions. You can't take away all the parts and still have a system, that makes no sense. But the parts by themselves aren't the system either, and nor can you simply add them up in any straightforward way. It's not linear like that; you're not just adding up. It's more like multiplying, and as I've just shown, multiplying can make very big numbers very fast.
Recent Comments