black swans and the point of science (preserved rant)
Last summer I was attending a Modeling Physics workshop while trying to read Nassim Nicholas Taleb's The Black Swan, and getting righteously indignant on behalf of science as I went. I wrote a long, beautiful rant which DW's "restore previous draft" feature accidentally ate; for the sake of tying up that loose end, I am going to include it here. My colleagues in this physics workshop are smart and fun people who love what they do and care a lot about doing it well. We are trying to get a better handle on how people build meaning from the world, and use it to find better ways to teach our students to create meaning. We are trying to learn how to teach students to make their own observations and draw their own conclusions from their data, rather than handing them our own pre-formed conclusions and insisting that they make observations that confirm our own interpretations. It's a delightfully anti-authoritarian approach, emphasizing that they trust their experience over the received wisdom, and it requires students to develop the kind of critical thinking skills that will be useful to them no matter what field they go into after high school. Better yet, explicitly developing these skills has been shown to help students in later painfully traditional higher-level physics classes as well. It's also depressingly different from the way I've been teaching physics, but that's okay - I can work on that. That's why I signed up for the workshop.
The take-away message for us is that the real goal of a physics class should not be to teach students physics, or at least, not primarily to teach them physics, but to teach them to think like physicists, to make observations and try to figure out what rules a system might be following, and then check if the rules they've come up with work. The particular set of rules that gets discovered is secondary, and you can use the same kinds of observation and reflective metacognition to determine the rules of each new system you examine. As a physics teacher, it was easy to pick up material I'd never learned, such as all the fluids topics, because I already understood how physics worked, even though I'd never been exposed to the material. Outside of physics, it's easier to look for patterns and take things apart logically. I know how to check its implications against my own observations, and look for deeper explanations when I hit some kind of contradiction. Essentially, the goal of Modeling is to use physics as a vehicle to teach empirical reasoning. It's great if students also learn some awesome bits of physics, and reinforce their sense of wonder with pieces like "each snowflake tells a story" or "your red sunset is someone else's blue sky". Physics is everywhere because it is the study of the system that underlies everything - but the important thing is that implicit in physics is a way of looking at everything thoughtfully and interactively until it makes sense. Those methods are even more widely applicable than the science itself.
Why use physics for this? As one of my colleagues put it, physics is a good vehicle to use to teach empiricism, because "physics is easy." The systems we choose to look at in physics are ones where empirical methods apply smoothly - extremely straightforward models can give some extremely accurate predictions. Part of why people like me and my colleagues love the subject is because we can apply empirical methods to it so easily, and they work. You look at the data, you try to look for logical patterns, and they pop out like holograms. Your systems are concrete and physical; your models work; if they're not perfect, you can refine them. Maybe it's a symptom of my own insecurity, but I'm always relieved when other physics people talk about seeking refuge from messy lives in this system where logic works. If it doesn't make sense, gather more data until it does. Tweak your model until it makes the right predictions. Keep checking your model's predictions against reality until you stop hitting inconsistencies. If you need to, replace it with a different model, but you can be confident that somewhere there is some way of looking at the information you have so that it makes some kind of sense. The unknown parts can be figured out later; even definitive holes in our information can be quantified precisely as things like entropy or uncertainty or probability.
Compare this to the various human social interactions we deal with. Logical models fail utterly to predict the behavior of another human, no matter how well we know them. Intuition doesn't actually do much better. It's not just that psychology is a younger science than physics, although that doesn't help; it's also an inherently more difficult system to try to apply mathematical models to. The data will never be clean; the models will never be perfect. Life will always be inherently disorganized and hard to measure. It's not just true for physics, but for all the squishy sciences, everything that tries to interpret and predict the behavior of particles that are unquantifiably different from each other. Psychology. History. Biology. Economics. Linguistics. There is no such field as psychohistory, and there probably never will be one, because social sciences are messy. For all that physics is the "hardest" science, it's a lot harder to come up with models for the "easy" ones. Physics is easy; soft sciences are difficult.
Trying to teach empirical reasoning starting with social science would involve a lot more of those awkward teacher moments where you do the demo, the kids interpret it wrong, and you have to explain what the demo was supposed to show, if it had worked right. If the particles had behaved as they were supposed to. The effect is bad enough in physics where we know everything that's going on - the table wasn't straight, the pulley got stuck on something, the charge leaked off into the atmosphere, the rail gun projectile somehow managed to weld itself to the track. Too many people come out of high school physics with the exact opposite idea we're trying to convey - that certain things are "true in physics, but not true in real life". Because we use simplified models that work in simplified cases, and then you try to apply them to more complicated cases and they fail, when what you need is a more sophisticated model. The frictionless inertial vacuum is a great approximation and gives you high-precision answers, except when it doesn't. And when it doesn't, if you keep going, there are deeper models, whether or not anybody has found them yet. Our workshop leader quoted Einstein on the first day: Make things as simple as possible, but no simpler. Beautiful things that don't describe anything aren't useful here. Contrary to the humor on science teachers' doors everywhere, if it doesn't work, it's not physics, or it's not enough physics, or the physics isn't right, or the physics isn't right yet. Physics is the set of models that describe physical systems in the real world. That's the whole point. The equations you learn in high school are an incomplete picture, and any physics teacher worth their salt is honest about that.
Obviously, I have some pro-physics bias here, which I should perhaps be careful of. I respect chemistry, which seems to do basically the same thing, trying to fit models to higher-order systems made of the same stuff (i.e., stuff), even if chemistry language is incoherent and impressive babble to me. I respect the attempts in other fields to make sense out of the world by collecting data and applying logic and looking for patterns, making predictions and checking how they work out. I am a fan of the scientific method in the actual sense of "run and find out", as opposed to whatever horrible list of arbitrary steps we make kids memorize in the first chapter of their science textbooks, year after year. When I was in college, I had a lot of disdain for people who tried to look at systems that I thought were boring and classist and unexciting (e.g. economics), but I am slowly gaining some respect for people who attempt to analyze these more difficult systems. Or trying to. I am seeing why it is interesting and where there are patterns; having a stake in the economy has helped me see why it matters. I've read some popular economics stuff and found it mostly interesting, especially when it connects to stuff I already know, either on the science side or the lived-experience side. It turns out that my undergraduate scorn may have been unfair.
Most recently, I've been working my way through essayist, empiricist, and "epistemologist of randomness" Nassim Nicholas Taleb's book The Black Swan, which addresses our human misconceptions that lead us to mispredict random events, the behavior and relevance of non-Gaussian randomness, and what he thinks we should do about it. A lot of the book is really interesting, but I can't help a certain feeling that after years of experiencing scorn from people in the hard sciences for working with softer data, he's trying to even the score and get back at them. The animosity between physicists and economists is not news, but Taleb's attacks on what he calls physics surprised me, especially as he weaves them into a impassioned argument for empiricism in economics. More on that later.
It is probably only fair to talk about the book first. The central icon of Taleb's book is the idea of the "Black Swan", the random piece of data that you couldn't predict, whose emergence nonetheless dramatically shifts everything. He opens with the anecdote that before explorers made it to Australia, everyone in Europe was convinced that all swans were white. He extends this metaphor to all kinds of other predictions made with incomplete data, trying to shed light on the fallacies we're prone to in an attempt to help people avoid catastrophically stupid actions, and maximize their reward potential. Humans are terrible at anticipating and preparing for the inherently unpredictable, and he argues convincingly that the consequences of this become more and more profound as the world becomes less discrete and more tightly interconnected. Taleb argues for more careful examination of empirical data, reduced faith in the predictions of appealing narratives, and a realistic assessment of risk that doesn't just ignore the unlikely because it's never happened before. He sprinkles in helpful anecdotes and thought experiments to create an entertaining and accurate depiction of the world we live in.
An important distinction Taleb draws is the difference between properties that are distributed normally, clustering tightly around some average value (he refers to this Gaussian-distributed type of information as the domain of "Mediocristan"), and properties that follow a wider distribution, falling off much more slowly at the extremes (he calls this "longer-tailed" kind of information the domain of "Extremistan"). Most human-scale variables are distributed in a Gaussian way - height, weight, intelligence, and so on. People's weights cluster around the average weight, and it's extremely unlikely that e.g. someone you randomly encounter on the subway has 100 times your weight, as this would give them the weight of a small bus. It's also unlikely that you'll encounter someone whose weight is only 1/100th of yours, since they would then have the weight of a few sticks of butter. In Mediocristan, outliers can be safely ignored, because they're not actually common enough or far enough from the mean to have a large effect - adding someone in the top 1% of weight to a large group of people won't significantly affect the average weight, so you can make predictions safely. In Extremistan, outliers can have a profound effect, and plans that ignore them are extremely dangerous. It's a lot more likely that you'll run into someone who has 100 times your income (a few million dollars?) than 100 times your weight; likewise it's more likely that you'll encounter someone scraping by below the poverty line than someone who is outweighed by a smallish melon. A millionaire in a large crowd would still have a dramatic effect on the average income of the crowd.
Income, along with all kinds of other variables where success breeds success, isn't distributed normally. The Occupy movement has made income inequality newsworthy, blame has been thrown around, and numerous proposals have been made to try to reduce it. I agree that it's a problem, but Taleb succeeded in convincing me that to a large extent it's an unavoidable consequence of globalization, which accelerates all the processes where success breeds success. The more interconnected the world is, the more your connections matter. Apparently we are all supposed to find jobs through "networking"; it's obvious that the main advantage a high-prestige education gives you isn't actually information, but connections.
The central conceit of Taleb's book is the black swan, and the human tendency to ignore the improbable, despite the huge conceptual and practical difference between "improbable" and "impossible". When a newscaster says there's a 10% chance of rain, if they're being accurate, one out of ten times, it will actually rain - and more likely than not, people will be angry at the newscaster. We struggle with probability on this very basic level, so it's not surprising that more advanced kinds of probability are even more difficult for people. We like narratives - the story of one kid who got vaccinated and developed autistic symptoms that very same day is much more convincing than massive amounts of statistics indicating that there is no correlation between vaccination and autism. Our search for patterns - an evolved skillset that has helped us out over the millenia - leads us to have a lot of difficulty analyzing things rationally and logically, and accepting the odds for what they really are.
Taleb is a survivor of the Lebanese civil war, and as such, he doesn't believe in history. He remembers the news accounts of the war when he was living through it, the experience of the war when he was living through it, and it was not a narrative. It was a series of random incoherent events. No one had any idea what was going on, probably including the reporters and the people making the decisions. But if you look up the Lebanese civil war now, it's a story, there's a plot and an ending, protagonists and antagonists, things that happened. He recounts the revelation he had, while hiding in a basement from some kind of aerial attack, when he discovered a book written by a war correspondent in Europe during World War I. Not a revised and polished account of what happened, not a history or an autobiography, but a collection of observations and letters, showing what "history" looked like as people were living through it. And the history Taleb had studied in school made as little sense as the history he was living through - in the moment, it's just a set of random events and occurrences. There's no pattern visible. Only afterwards does the human urge to make sense of things weave all those disparate pieces together into a narrative, whether or not that's a faithful portrayal. Our brains remember stories, not disconnected facts.
So Taleb believes in empiricism - recording what really happens. He is skeptical about narratives, and he doesn't believe in history. This all sounds fine to me. He's also got a weird beef against what he calls "physics" and the attempt to analyze the world using math and logic. The world isn't fundamentally rational, he asserts, and moreover, the world doesn't fit to these nice normal distributions the physicists like to use. Not all collisions are elastic. Physicists' models don't actually model the world, and it's a problem. Better not to bother with this whole modeling thing at all, if you're going to do it wrong.
Neck-deep in a workshop on the beauty of using physics to model the world - I wrote in my course descriptions for the next fall, "Physicists understand the universe by discovering rules and mathematical patterns. We look for relationships we can analyze mathematically, and use these relationships to make predictions." Physics is the science that uses math to model the world; if the math is not a fair model of the world, it's not good physics. Or it's not done yet. Physicists will be the first to tell you that physics is not done yet, that our picture of the universe is still incomplete, and we're continuing to develop better tools and better models. This is the sacred quest of science. And more pragmatically, this is why there are still jobs in this field. If we were done, there would be no jobs. I kind of wish Taleb could approach his own field with the same amount of humility.
Part of Taleb's problem, I think, is that his idea of physics is all wrong. As someone working on Wall Street, the only physics he's been exposed to is completely bowdlerized physics. Several steps removed from science. Normal distributions are perfectly reasonable things to use, say when you have a system of perfectly hard spheres engaging in perfectly elastic collisions. We come up with models for how to treat gases if they were ideal, and then corrections to those models for if they aren't ideal. We define terms like enthalpy and entropy and other things that have actual meaning. The math of finance seems to be a creepy mirror of the math of physics - at some point last year I was going
holdthesky's back entries, and I encountered The Greeks (finance), which appear to be relatively arbitrarily defined calculations that give you properties of a market. It's not clear if these measurements have any actual validity, but you can calculate them in ways similar to how physicists calculate things, and people claim they mean things. Also, the people who came up with the names were either (a) really ignorant about the Greek alphabet, (b) making fun of us, or (c) on more and better drugs than the guys who named the quarks. Theta, Vega, Vomma. Sure.
In the same breath, Taleb speaks out in favor of empiricism but against physics. And it offends me. He doesn't see that physicists are also champions of empiricism, and that physicists are just as strongly opposed as he is to grafting a beautiful mathematical system onto a reality that doesn't match it. The ultimate test of a model is not whether it's beautiful, but whether it works. That's why we don't have a grand unified theory yet, because we have to actually check if stuff works. It's also why we're not rigidly tied to the Boltzmann distribution, and there are several different models for how values can be distributed. There's this whole awesome field called complexity theory that studies the different kinds of distributions - with the long tails that standard normal distributions don't account for - that arise in highly connected systems, like those we see in social networks and the global economy. On superbacana's recommendation, I read this awesome book called Critical Mass and wrote briefly about it. Long tails, punctuated equilibrium models, ideas that people have been talking about for years, going back to Kuhn's writing about the structure of scientific revolutions. Physics is not the problem: physics is on this. I can't help thinking that if the Wall Street guys would just let the actual physicists handle this for them, we'd do a pretty good job. Better than they seem to, anyway, and perhaps with more humility.
So in short, Taleb's observations about the nature of humanity seem on-point; his musings on history were cool; his ideas about what science is and what it strives for made me angry and frustrated. Not sure if I'm going to actually finish this book.
At some point I will probably figure out what happened to
ccommack's copy of The Black Swan, and I will be faced with the question of whether to finish reading it. But right now I am trying to read Nate Silver's book The Signal and the Noise, and it's touching on a lot of the same ideas - thus my need to finish out the previous book with a post, before starting the next one. Silver is also a strong proponent of Bayesianism as the solution to everything. I am still passionate about science but my enthusiasm about Modeling is a little more measured after a year of trying to actually implement things on real students. But I am sure I will continue to have thoughts, and you will almost certainly have the opportunity to read them, as they come up.
This entry has
comments. Leave a comment? (Click "Log in with OpenID?", top right corner.)
The take-away message for us is that the real goal of a physics class should not be to teach students physics, or at least, not primarily to teach them physics, but to teach them to think like physicists, to make observations and try to figure out what rules a system might be following, and then check if the rules they've come up with work. The particular set of rules that gets discovered is secondary, and you can use the same kinds of observation and reflective metacognition to determine the rules of each new system you examine. As a physics teacher, it was easy to pick up material I'd never learned, such as all the fluids topics, because I already understood how physics worked, even though I'd never been exposed to the material. Outside of physics, it's easier to look for patterns and take things apart logically. I know how to check its implications against my own observations, and look for deeper explanations when I hit some kind of contradiction. Essentially, the goal of Modeling is to use physics as a vehicle to teach empirical reasoning. It's great if students also learn some awesome bits of physics, and reinforce their sense of wonder with pieces like "each snowflake tells a story" or "your red sunset is someone else's blue sky". Physics is everywhere because it is the study of the system that underlies everything - but the important thing is that implicit in physics is a way of looking at everything thoughtfully and interactively until it makes sense. Those methods are even more widely applicable than the science itself.
Why use physics for this? As one of my colleagues put it, physics is a good vehicle to use to teach empiricism, because "physics is easy." The systems we choose to look at in physics are ones where empirical methods apply smoothly - extremely straightforward models can give some extremely accurate predictions. Part of why people like me and my colleagues love the subject is because we can apply empirical methods to it so easily, and they work. You look at the data, you try to look for logical patterns, and they pop out like holograms. Your systems are concrete and physical; your models work; if they're not perfect, you can refine them. Maybe it's a symptom of my own insecurity, but I'm always relieved when other physics people talk about seeking refuge from messy lives in this system where logic works. If it doesn't make sense, gather more data until it does. Tweak your model until it makes the right predictions. Keep checking your model's predictions against reality until you stop hitting inconsistencies. If you need to, replace it with a different model, but you can be confident that somewhere there is some way of looking at the information you have so that it makes some kind of sense. The unknown parts can be figured out later; even definitive holes in our information can be quantified precisely as things like entropy or uncertainty or probability.
Compare this to the various human social interactions we deal with. Logical models fail utterly to predict the behavior of another human, no matter how well we know them. Intuition doesn't actually do much better. It's not just that psychology is a younger science than physics, although that doesn't help; it's also an inherently more difficult system to try to apply mathematical models to. The data will never be clean; the models will never be perfect. Life will always be inherently disorganized and hard to measure. It's not just true for physics, but for all the squishy sciences, everything that tries to interpret and predict the behavior of particles that are unquantifiably different from each other. Psychology. History. Biology. Economics. Linguistics. There is no such field as psychohistory, and there probably never will be one, because social sciences are messy. For all that physics is the "hardest" science, it's a lot harder to come up with models for the "easy" ones. Physics is easy; soft sciences are difficult.
Trying to teach empirical reasoning starting with social science would involve a lot more of those awkward teacher moments where you do the demo, the kids interpret it wrong, and you have to explain what the demo was supposed to show, if it had worked right. If the particles had behaved as they were supposed to. The effect is bad enough in physics where we know everything that's going on - the table wasn't straight, the pulley got stuck on something, the charge leaked off into the atmosphere, the rail gun projectile somehow managed to weld itself to the track. Too many people come out of high school physics with the exact opposite idea we're trying to convey - that certain things are "true in physics, but not true in real life". Because we use simplified models that work in simplified cases, and then you try to apply them to more complicated cases and they fail, when what you need is a more sophisticated model. The frictionless inertial vacuum is a great approximation and gives you high-precision answers, except when it doesn't. And when it doesn't, if you keep going, there are deeper models, whether or not anybody has found them yet. Our workshop leader quoted Einstein on the first day: Make things as simple as possible, but no simpler. Beautiful things that don't describe anything aren't useful here. Contrary to the humor on science teachers' doors everywhere, if it doesn't work, it's not physics, or it's not enough physics, or the physics isn't right, or the physics isn't right yet. Physics is the set of models that describe physical systems in the real world. That's the whole point. The equations you learn in high school are an incomplete picture, and any physics teacher worth their salt is honest about that.
Obviously, I have some pro-physics bias here, which I should perhaps be careful of. I respect chemistry, which seems to do basically the same thing, trying to fit models to higher-order systems made of the same stuff (i.e., stuff), even if chemistry language is incoherent and impressive babble to me. I respect the attempts in other fields to make sense out of the world by collecting data and applying logic and looking for patterns, making predictions and checking how they work out. I am a fan of the scientific method in the actual sense of "run and find out", as opposed to whatever horrible list of arbitrary steps we make kids memorize in the first chapter of their science textbooks, year after year. When I was in college, I had a lot of disdain for people who tried to look at systems that I thought were boring and classist and unexciting (e.g. economics), but I am slowly gaining some respect for people who attempt to analyze these more difficult systems. Or trying to. I am seeing why it is interesting and where there are patterns; having a stake in the economy has helped me see why it matters. I've read some popular economics stuff and found it mostly interesting, especially when it connects to stuff I already know, either on the science side or the lived-experience side. It turns out that my undergraduate scorn may have been unfair.
Most recently, I've been working my way through essayist, empiricist, and "epistemologist of randomness" Nassim Nicholas Taleb's book The Black Swan, which addresses our human misconceptions that lead us to mispredict random events, the behavior and relevance of non-Gaussian randomness, and what he thinks we should do about it. A lot of the book is really interesting, but I can't help a certain feeling that after years of experiencing scorn from people in the hard sciences for working with softer data, he's trying to even the score and get back at them. The animosity between physicists and economists is not news, but Taleb's attacks on what he calls physics surprised me, especially as he weaves them into a impassioned argument for empiricism in economics. More on that later.
It is probably only fair to talk about the book first. The central icon of Taleb's book is the idea of the "Black Swan", the random piece of data that you couldn't predict, whose emergence nonetheless dramatically shifts everything. He opens with the anecdote that before explorers made it to Australia, everyone in Europe was convinced that all swans were white. He extends this metaphor to all kinds of other predictions made with incomplete data, trying to shed light on the fallacies we're prone to in an attempt to help people avoid catastrophically stupid actions, and maximize their reward potential. Humans are terrible at anticipating and preparing for the inherently unpredictable, and he argues convincingly that the consequences of this become more and more profound as the world becomes less discrete and more tightly interconnected. Taleb argues for more careful examination of empirical data, reduced faith in the predictions of appealing narratives, and a realistic assessment of risk that doesn't just ignore the unlikely because it's never happened before. He sprinkles in helpful anecdotes and thought experiments to create an entertaining and accurate depiction of the world we live in.
An important distinction Taleb draws is the difference between properties that are distributed normally, clustering tightly around some average value (he refers to this Gaussian-distributed type of information as the domain of "Mediocristan"), and properties that follow a wider distribution, falling off much more slowly at the extremes (he calls this "longer-tailed" kind of information the domain of "Extremistan"). Most human-scale variables are distributed in a Gaussian way - height, weight, intelligence, and so on. People's weights cluster around the average weight, and it's extremely unlikely that e.g. someone you randomly encounter on the subway has 100 times your weight, as this would give them the weight of a small bus. It's also unlikely that you'll encounter someone whose weight is only 1/100th of yours, since they would then have the weight of a few sticks of butter. In Mediocristan, outliers can be safely ignored, because they're not actually common enough or far enough from the mean to have a large effect - adding someone in the top 1% of weight to a large group of people won't significantly affect the average weight, so you can make predictions safely. In Extremistan, outliers can have a profound effect, and plans that ignore them are extremely dangerous. It's a lot more likely that you'll run into someone who has 100 times your income (a few million dollars?) than 100 times your weight; likewise it's more likely that you'll encounter someone scraping by below the poverty line than someone who is outweighed by a smallish melon. A millionaire in a large crowd would still have a dramatic effect on the average income of the crowd.
Income, along with all kinds of other variables where success breeds success, isn't distributed normally. The Occupy movement has made income inequality newsworthy, blame has been thrown around, and numerous proposals have been made to try to reduce it. I agree that it's a problem, but Taleb succeeded in convincing me that to a large extent it's an unavoidable consequence of globalization, which accelerates all the processes where success breeds success. The more interconnected the world is, the more your connections matter. Apparently we are all supposed to find jobs through "networking"; it's obvious that the main advantage a high-prestige education gives you isn't actually information, but connections.
The central conceit of Taleb's book is the black swan, and the human tendency to ignore the improbable, despite the huge conceptual and practical difference between "improbable" and "impossible". When a newscaster says there's a 10% chance of rain, if they're being accurate, one out of ten times, it will actually rain - and more likely than not, people will be angry at the newscaster. We struggle with probability on this very basic level, so it's not surprising that more advanced kinds of probability are even more difficult for people. We like narratives - the story of one kid who got vaccinated and developed autistic symptoms that very same day is much more convincing than massive amounts of statistics indicating that there is no correlation between vaccination and autism. Our search for patterns - an evolved skillset that has helped us out over the millenia - leads us to have a lot of difficulty analyzing things rationally and logically, and accepting the odds for what they really are.
Taleb is a survivor of the Lebanese civil war, and as such, he doesn't believe in history. He remembers the news accounts of the war when he was living through it, the experience of the war when he was living through it, and it was not a narrative. It was a series of random incoherent events. No one had any idea what was going on, probably including the reporters and the people making the decisions. But if you look up the Lebanese civil war now, it's a story, there's a plot and an ending, protagonists and antagonists, things that happened. He recounts the revelation he had, while hiding in a basement from some kind of aerial attack, when he discovered a book written by a war correspondent in Europe during World War I. Not a revised and polished account of what happened, not a history or an autobiography, but a collection of observations and letters, showing what "history" looked like as people were living through it. And the history Taleb had studied in school made as little sense as the history he was living through - in the moment, it's just a set of random events and occurrences. There's no pattern visible. Only afterwards does the human urge to make sense of things weave all those disparate pieces together into a narrative, whether or not that's a faithful portrayal. Our brains remember stories, not disconnected facts.
So Taleb believes in empiricism - recording what really happens. He is skeptical about narratives, and he doesn't believe in history. This all sounds fine to me. He's also got a weird beef against what he calls "physics" and the attempt to analyze the world using math and logic. The world isn't fundamentally rational, he asserts, and moreover, the world doesn't fit to these nice normal distributions the physicists like to use. Not all collisions are elastic. Physicists' models don't actually model the world, and it's a problem. Better not to bother with this whole modeling thing at all, if you're going to do it wrong.
Neck-deep in a workshop on the beauty of using physics to model the world - I wrote in my course descriptions for the next fall, "Physicists understand the universe by discovering rules and mathematical patterns. We look for relationships we can analyze mathematically, and use these relationships to make predictions." Physics is the science that uses math to model the world; if the math is not a fair model of the world, it's not good physics. Or it's not done yet. Physicists will be the first to tell you that physics is not done yet, that our picture of the universe is still incomplete, and we're continuing to develop better tools and better models. This is the sacred quest of science. And more pragmatically, this is why there are still jobs in this field. If we were done, there would be no jobs. I kind of wish Taleb could approach his own field with the same amount of humility.
Part of Taleb's problem, I think, is that his idea of physics is all wrong. As someone working on Wall Street, the only physics he's been exposed to is completely bowdlerized physics. Several steps removed from science. Normal distributions are perfectly reasonable things to use, say when you have a system of perfectly hard spheres engaging in perfectly elastic collisions. We come up with models for how to treat gases if they were ideal, and then corrections to those models for if they aren't ideal. We define terms like enthalpy and entropy and other things that have actual meaning. The math of finance seems to be a creepy mirror of the math of physics - at some point last year I was going
holdthesky's back entries, and I encountered The Greeks (finance), which appear to be relatively arbitrarily defined calculations that give you properties of a market. It's not clear if these measurements have any actual validity, but you can calculate them in ways similar to how physicists calculate things, and people claim they mean things. Also, the people who came up with the names were either (a) really ignorant about the Greek alphabet, (b) making fun of us, or (c) on more and better drugs than the guys who named the quarks. Theta, Vega, Vomma. Sure.
In the same breath, Taleb speaks out in favor of empiricism but against physics. And it offends me. He doesn't see that physicists are also champions of empiricism, and that physicists are just as strongly opposed as he is to grafting a beautiful mathematical system onto a reality that doesn't match it. The ultimate test of a model is not whether it's beautiful, but whether it works. That's why we don't have a grand unified theory yet, because we have to actually check if stuff works. It's also why we're not rigidly tied to the Boltzmann distribution, and there are several different models for how values can be distributed. There's this whole awesome field called complexity theory that studies the different kinds of distributions - with the long tails that standard normal distributions don't account for - that arise in highly connected systems, like those we see in social networks and the global economy. On superbacana's recommendation, I read this awesome book called Critical Mass and wrote briefly about it. Long tails, punctuated equilibrium models, ideas that people have been talking about for years, going back to Kuhn's writing about the structure of scientific revolutions. Physics is not the problem: physics is on this. I can't help thinking that if the Wall Street guys would just let the actual physicists handle this for them, we'd do a pretty good job. Better than they seem to, anyway, and perhaps with more humility.
So in short, Taleb's observations about the nature of humanity seem on-point; his musings on history were cool; his ideas about what science is and what it strives for made me angry and frustrated. Not sure if I'm going to actually finish this book.
At some point I will probably figure out what happened to
ccommack's copy of The Black Swan, and I will be faced with the question of whether to finish reading it. But right now I am trying to read Nate Silver's book The Signal and the Noise, and it's touching on a lot of the same ideas - thus my need to finish out the previous book with a post, before starting the next one. Silver is also a strong proponent of Bayesianism as the solution to everything. I am still passionate about science but my enthusiasm about Modeling is a little more measured after a year of trying to actually implement things on real students. But I am sure I will continue to have thoughts, and you will almost certainly have the opportunity to read them, as they come up.
This entry has
comments. Leave a comment? (Click "Log in with OpenID?", top right corner.)