In his Foundation series of books, Isaac Asimov imagined a science, which he termed psycho-history, that combined elements of psychology, history, economics, and statistics to predict the behaviors of large population over time under a given set of socio-economic conditions. It’s an intriguing idea. And I have no doubt much, much more difficult to do than it sounds, and it doesn’t sound particularly easy to begin with.
Behavioral modeling is currently being used in many of the science and engineering disciplines. Finite element analysis (FEA), for example, is used to model electromagnetic effects, thermal effects and structural behaviors under varying conditions. The ‘elements’ in FEA are simply building blocks, maybe a tiny cube of aluminum, that are given properties like stiffness, coefficient of thermal expansion, thermal resistivity, electrical resistivity, flexural modulus, tensile strength, mass, etc. Then objects are constructed from these blocks and, under stimulus, they take on macro-scale behaviors as a function of their micro-scale properties. There are a couple of key ideas to keep in mind here, however. The first is that inanimate objects do not exercise free will. The second is that the equations used to derive effects are based on first principles, which is to say basic laws of physics, which are tested and well understood. A similar approach is used for computational fluid dynamics (CFD), which is used to model the atmosphere for weather prediction, the flow of water over a surface for dam design, or the flow of air over an aircraft model. The power of these models lies in the ability of the user to vary both the model and the input stimulus parameters and then observe the effects. That’s assuming you’ve built your model correctly. That’s the crux of it, isn’t it?
I was listening to a lecture on the work of a Swiss team of astrophysicists the other day called the Quantum Origins of Space and Time. They made an interesting prediction based on the modeling they’ve done of the structure of spacetime. In a result sure to disappoint science fiction fans everywhere, they predict that wormholes do not exist. The reason for the prediction is simply that when they allow them to exist at the quantum level, they cannot get a large scale universe to form over time. When they are disallowed, the same models create De Sitter universes like the one we have.
It occurred to me that it would be interesting to have the tools to run models with societies. Given the state of a society X, what is the economic effect of tax policy Y. More to the point, what is cumulative effect of birth rate A, distribution of education levels B, distribution of personal debt C, distribution of state tax rates D, federal debt D, total cost to small business types 1-100 in tax and regulations, etc. This would allow us to test the effects of our current structure of tax, regulation, education and other policies. Setting up the model would be a gargantuan task. You would need to dedicate the resources of an institute level organization with expertise across a wide range of disciplines. Were we to succeed in building even a basic functioning model, its usefulness would be beyond estimation to the larger society.
It’s axiomatic that anything powerful can and will be weaponized. It is also completely predictable that the politically powerful would see this as a tool for achieving their agenda. Simply imagine the software and data sets under the control of a partisan governing body. How might they bias the data to skew the output to a desired state? How might they bias the underlying code? Might an enemy state hack the system with the goal to have you adopt damaging policies, doing the work of social destruction at no expense or risk to them?
Is this achievable? I think yes. All or most of the building blocks exist: computational tools, data, statistical mathematics and economic models. We are in the state we were in with regard to computers in the 1960s, before microprocessors. All the building blocks existed as separate entities, but they had not been integrated in a single working unit at the chip level. What’s needed is the vision, funding and expertise to put it all together. This might be a good project for DARPA.
“It’s axiomatic that anything powerful can and will be weaponized.”
I can hardly wait to see their ham handed attempts to manipulate it. Hockey stick graphs for everyone!
psycho-history, that combined elements of psychology, history, economics, and statistics to predict the behaviors of large population over time under a given set of socio-economic conditions
Genetics is going turn this on its head. Stephen PInker started with linguistics and some studies of twins. His book, The Blank Slate, refutes evolutionary biologist Stephen Jay Gould, who postulated a “Blank Slate” in newborns that could be conditioned to increase intelligence or alter behavior. Pinker’s concept, now proven by genetics, is that genetics is fate.
Now we have race based screening for prostate cancer, and Hawaii attorneys general suing because some groups have different reactions to drugs based on genetics.
The truth is that all races are different in more than skin color.
Sub-Sahara African blacks are quite a bit different in things like salt tolerance, hypertension, malaria resistance and other factors, like G6PD deficiency.
Asians have whiter skin the farther north they come from but none have blue eyes.
It’s going to get very complicated.
Given the state of a society X, what is the economic effect of tax policy Y. More to the point, what is cumulative effect of birth rate A, distribution of education levels B, distribution of personal debt C, distribution of state tax rates D, federal debt D, total cost to small business types 1-100 in tax and regulations, etc.
Mike is stealing from my schtick.
Your model posits environmental factors of various sorts acting on a population. Populations differ. The most widely known difference in IQ, but there are also differences in the Big 5 personality traits. Now add religious and cultural influences and decisions will differ between populations reacting to identical stimuli.
Mike K. – The idea that all races/ethnic groups are basically the same behaviorally never made any sense whatsoever. Human beings are the result of a complex natural process of biological evolutiion. Complex natural processes do not produce simple patterns like equality in anything. Differences in average height between different human groups range over about four statndard deviations somewhat similar to a range of about 3-4 standard dviations in IQ (2-3 standard deviations for the greater part of the human species). Of course a significant amount of this reflects environmental influences but the nearly 1 standard deviation in average IQ between US gentiles of European descent and Ashkenazi Jews is almost totally genetic.
Econometric modeling has been around for quite some time. How accurate have predictions from these models been?
the tools to run models with societies
Like climate models? Maybe eventually, but this doesn’t seem likely to be useful for the foreseeable future.
“That’s assuming you’ve built your model correctly. That’s the crux of it, isn’t it?”
We used to have a saying when running analyses – ‘garbage in, garbage out’
The problem with modeling complex systems, unlike mechanical systems, is not only that the many many initial variables and conditions are unknown, but that the interaction of variables are unknowable because of extreme sensitivity to initial conditions. Small, seemingly insignificant interactions can have an outsized effect or completely affect the outcome one time, then have no effect another time.
Mike K. – Sub-Saharan African blacks are also quite a bit different from European whites in IQ.
One of our most basic problems is that our current batch of “Elites” who infest the power structures of most government branches, agencies and universities believe they innately have this ability already. And it really pisses them off when we don’t just do what we’re told.
Unfortunately for us, the knowledge most “Elites” have on most subjects they are speaking about or acting on is only “Power Point” deep. And someone else made the Power Point for them.
In the Foundation books, Asimov recognized the power of this predictive power and wisely locked it away from the potential abusers, releasing the results/predictions as if from a time locked vault. They are great and very interesting books, I would highly recommend them and his Robot series to anyone who enjoys Science Fiction.
Oh, and we know that they don’t know what they are talking about. Forgot that part.
“Econometric modeling has been around for quite some time. How accurate have predictions from these models been?” Not much worse than astrology.
“Econometric modeling has been around for quite some time. How accurate have predictions from these models been?” Not much worse than astrology.
People used to say that about weather forecasting. It’s actually gotten surprisingly accurate for as much as a week out as they’ve gotten better at modeling. I expect the social-economic modeling I describe would follow a similar trajectory. Terrible at first, but better over time.
Economic models of people’s behavior are less accurate than the CO2 models of Global warming alarmists.
The Chicagoboy in me thinks, “What would Hayek say?”
To the blog’s proprietor – Is there a way to Google search for a comment I made on a post from a few years back without including results from the “recent comment” sidebar? Google is returning almost every post written on this blog when I search for my name on this site and I know for a fact that I have not been that prolific a commenter.
The simple fact is that this is highly questionable as to its practical possibility at any point in the near-term future.
First off, it’s a HIGHLY chaotic system you posit. VERY SMALL inputs can have exceptionally serious effects on the direction it goes. We simply don’t understand enough about how chaotic systems work yet. We only first RECOGNIZED them in the 60s and 70s and really began to explore them in the 80s.
This is a part of the reason why AGW models fail so badly… And they are much much less complex than the model you posit implementing and studying.
Second, as with AGW, you can bet your sweet bippy that any government out there who attempt such a model will abuse it to sell their agenda LONG before it’s valid or even viable.
Michael Hiteshew – We can hope that some time in the future econometric models will be able to predict the state of the economy a week in advance.
“All or most of the building blocks exist: computational tools, data, statistical mathematics and economic models. We are in the state we were in with regard to computers in the 1960s, before microprocessors. All the building blocks existed as separate entities, but they had not been integrated in a single working unit at the chip level. What’s needed is the vision, funding and expertise to put it all together.”
Those are actually software. The computational tools, data, statistical mathematics and economic models are software of various kinds. They run on chips of various kinds but that’s not really what’s important.
In a *nix system the various important software tools are not integrated into a single tool. It’s been found systems of simple tools are much easier to manage and to work well on the problems you might have. For instance, it’s simple tools that run *nix subsystems, all kinds of network tools, and system tools are made of of small bits of code that are very good at what they do. These are chained together by several methods, and call each other as needed.
Windose, that pitiful excuse for an OS, likes to make huge integrated messes, and then try for years to get them to run well. It’s not anything I need and frankly it’s a bad way to arrange software.
“but they had not been integrated in a single working unit” That is just fine, and not the problem.
The Chicagoboy in me thinks, “What would Hayek say?”
I think you misunderstand my intention. I’m not advocating a centrally planned or controlled economy or society. I’m simply saying it would be useful to model the effects of tax(es) or regulations or policies and their effects on the economy or society. An an example, modeling the effect of raising the minimum wage.
Think of it like this. If I raise the toll on a tunnel, some number of people will simply take another route. If I raise state income tax beyond a certain point, some number of people will simply leave the state for another or find tax shelters. Andrew Mellon, as Secretary of the Treasury, proposed that 25% federal income tax was the most that people would pay before they simply started moving their money into tax sheltered bonds. By reducing the top rate to 25%, capital investment increased, the economy grew, and total revenue increased to the treasury, despite the much lower top rate. He applied his empirical knowledge of how investors behaved to the tax code. It turns out he was correct. It would be nice to capture that knowledge and other knowledge like it to model the effects of various policies.
The simple fact is that this is highly questionable as to its practical possibility at any point in the near-term future. First off, it’s a HIGHLY chaotic system you posit. VERY SMALL inputs can have exceptionally serious effects on the direction it goes.
I disagree. We do not live in a highly chaotic society. We live in a highly predictable society, despite that fact that individuals engage in chaotic behaviors here and there and people and government engage in destabilizing behaviors. The fact that society doesn’t immediately fall apart tells me the social system is highly robust. And an model we built would need to be equally robust, such that small changes in inputs did not create nonsense as an output.
While you may not be advocating a centrally planned or controlled economy or society, a quarter of the US population currently does, in spite of the demonstrated imfeasibility of the idea. If anything remotely similar to what you are proposing were possible, that percentage would increase substantially.
However, Hayek would say that there can never be a computer big enough or fast enough to make determinations faster than a free market with unimpeded prices. Elasticities of demand in the face of infinite opportunities for substitution and consumer preferences that change daily as conditions change cannot be calculated with sufficient accuracy to be of use. Human beings are just too complex and irrational to be accurately modeled.
Keep thinking this way, and you’ll be on the Road.
the nearly 1 standard deviation in average IQ between US gentiles of European descent and Ashkenazi Jews is almost totally genetic.
Yes and that is what also gave them Tay Sachs and other lipid storage diseases. There is an interesting theory that some of the genetic diseases in Ashkenazi Jews are related to the same mutations that give higher intelligence.
The standard deviation between blacks and white mean is about 1 SD. The tails of the distribution are where the difference comes. I think universities are running out of right tail black students. I suspect BLM is coming from that phenomenon. John Derbyshire referred to it in his famous article, that got him fired by Rich Lowry.
In a population of forty million, you will find almost any human type. Only at the far, far extremes of certain traits are there absences. There are, for example, no black Fields Medal winners. While this is civilizationally consequential, it will not likely ever be important to you personally. Most people live and die without ever meeting (or wishing to meet) a Fields Medal winner.
(6) As you go through life, however, you will experience an ever larger number of encounters with black Americans. Assuming your encounters are random—for example, not restricted only to black convicted murderers or to black investment bankers—the Law of Large Numbers will inevitably kick in. You will observe that the means—the averages—of many traits are very different for black and white Americans, as has been confirmed by methodical inquiries in the human sciences.
Explosive facts.
However, Hayek would say that there can never be a computer big enough or fast enough to make determinations faster than a free market with unimpeded prices. Elasticities of demand in the face of infinite opportunities for substitution and consumer preferences that change daily as conditions change cannot be calculated with sufficient accuracy to be of use. Human beings are just too complex and irrational to be accurately modeled.
Keep thinking this way, and you’ll be on the Road.
Those are very good points.
Maybe deterministic forecasting is beyond our reach, but there are still plausible methods to recognize when things may be getting ready to significantly change
http://www.early-warning-signals.org/theory/
And there are reasonable attempts to create models by practitioners of system dynamics. The famous (at least in their world) pioneer Donella Meadows had her ‘leverage points’
https://en.wikipedia.org/wiki/Twelve_leverage_points
Although then you enter the realm of unintended consequences. We think we’re just chipping away at the edges of a system, when we may actually be tossing a rock into the pond of an even bigger encompassing system causing ripples in every direction.
A recent example was the reintroduction of wolves to Yellowstone
http://www.popsci.com/article/science/have-wolves-really-saved-yellowstone
It was initially hailed as a success, but now there are questions about whether we understand everything about the forest. Wolves help some things and hinder others.
Grurray, if you want to know what to do about wolves in Montana, be sure to ask New York Times readers.
California is now over run with mountain lions thanks to similar decision makers.
A broad summary of the social simulation field:
https://en.wikipedia.org/wiki/Social_simulation
Also see my post ‘The 480’, which describes the work of a company called Simulmatics for the 1960 Kennedy presidential campaign:
https://chicagoboyz.net/archives/45418.html
Simulmatics also did work for DoD involving the prediction of insurgencies and ways to combat them, encompassing both Latin America and Vietnam. One of its projects, for which it received $6 million in funding, was ‘Project Camelot’ which proposed to produce: (a) the systematic identification of the symptoms of the breakdown of a society, and (b) the identification of actions that might forestall such a breakdown. It was terminated after a left-wing Chilean newspaper found out about it.
Peter Drucker said about the project: ‘Camelot never could have produced results. It was pure, unadultrated, six-million-dollar fraud. There are no research techniques to answer such questions as ‘what causes the breakdown of a society,’ let along ‘how do we prevent it?’
“Hayek would say that there can never be a computer big enough or fast enough to make determinations faster than a free market with unimpeded prices.”
What exactly do you think all those trades are made on? The market is run on computers and although it might seem massive to you, it’s not all that huge.
As well the market is distorted by the very computers it runs on. HFT is real folks.
What exactly do you think all those trades are made on? The market is run on computers and although it might seem massive to you, it’s not all that huge.
As well the market is distorted by the very computers it runs on. HFT is real folks.
From a computational perspective, high frequency trading is a bit like very short-term weather forecasting done in competition with other weather forecasters. This is much different from long-term forecasting. The longer the forecast term becomes, the more the system has to forecast major trends and events accurately, and the harder it becomes to do so. (Also, HFT is middleman behavior – much of the profit comes from capturing fractions of bid/offer spreads, for which there is no parallel in forecasting non-market events.)
With current knowledge, long-term forecasting in financial markets, weather and social behavior alike is highly inaccurate. Finally, it seems likely that any accurate system for making long-term predictions about social behavior would produce feedback effects that moderate the predicted effects, as market forecasts sometimes do.
No one has done it, but if you took the computational power available to the machines attached to the Large Hadron Collider, one might have power approaching what might be needed to make accurate predictions.
Having the power is not enough, a scheme to supply useful data is also needed. ;)
We need sufficiently powerful computers, accurate data and accurate models. We aren’t there yet. The poor predictive accuracy of the current anthropogenic global warming models should be cautionary.
I don’t think you can predict the behavior of social systems with any accuracy without an accurate model of individual human behavior. No matter how much computational capacity you have.
Will NYC people move to to somewhere cheaper if taxes in NYC rise by X% ? The answer will partly depend on the degree to which urban living becomes/remains trendy among people in various demographics. Also depends on relative crime rates, which might depend on something like whether or not a Bill de Blasio wins a mayoral election.
Increased computational capacity doesn’t solve these problems.
I think the main benefit of social modeling is getting people to *think* about the complexity of the interactions in social systems…but don’t take the results of the models to seriously.
Some interesting work on social simulations, and decision-making in general, by Prof Dietrich Doerner:
“The Moro simulation puts the subject in charge of a third-world country. His decision-making must include issues such as land use, water supply, medical care, etc. Time delays and multiple interactions make this simulation hard to handle effectively…a high proportion of subjects wound up making things worse rather than better for their “citizens.” Human beings, Doerner argues, have much more difficulty understanding patterns that extend over time than patterns that are spatial in nature.”
http://photonplaza.blogspot.com/2003_12_28_photonplaza_archive.html#107275628844222087
Michael Hiteshew Says:
June 3rd, 2016 at 11:27 am
And the reason weather models are only good for a week is….
Chaos. In the environment. In the models. And the two can never ever match.
So what might we expect in the way of improvement? Well the first thing to improve is not the models or the computers. The first thing to improve is the measurements.
So can population behavior be predicted? In the very short term – maybe. Longer term? I can help. But first you have to answer a question. What new products will be introduced in two years? In twenty? Once we know that the rest is relatively easy. And by “relatively easy” I mean incredibly difficult.
Looking at the Deep Mind vs the Korean Go Master again. Deep Mind learned how to play by analyzing thousands of master level games and ended up beating the world’s premier Go Master 4 out of 5 games.
Many thought this was impossible, and it is amazing, as Go is a much larger problem than Chess.
This technique has yielded stellar results on a very complex game. It, assuming Google’s programmers can adapt their learning programming, would be a very good candidate for solving problems in large scale prediction of major trends.
Pengun, you are talking about games, where the rules are clearly defined and the counter-move to any move can only come from a small and limited set of clearly defined alternatives. All you need in such cases is a sufficiently powerful computer to do a huge number of simulations to determine the best-odds counter-move to any specific move. That is a much, much easier type of problem than are predictions in markets or other unbounded social or natural systems, with huge sets of often poorly understood variables, where the relationships between variables are not well understood, and where data are limited in availability and variable in quality.
Go is far too large to brute force. The Deep Mind computer learned to play good moves through observing masters play. It’s fundamentally different than the chess playing computers that have gone before. The actual number of possible moves is many times the number of atoms in the universe.
The actual set of market variables and the stuff that’s well delineated is easily handled by a machine that deals with Go like games. Getting it to do that is much harder.
Consider the analysis of mortgages and mortgage bundles, and the prediction of which ones will default–a topic that proved to be some some importance circa 2008. Factors that need to be considered to do this meaningfully (for the case of a bundle) include:
1) The probabilities of people losing their jobs, or having to take ones that pay less $
2) The geographical pattern of such job losses—apparently that models that were actually in use did not adequately consider correlation of economic problems across geographical regions
3) The willingness of people to walk away from a mortgage and let the property be foreclosed, rather than very painfully continuing to make payments
4) The extent of misrepresentation of incomes by borrowers, and collusion in this by mortgage brokers
5) The possibility that the bank selling the mortgage bundle is getting rid of ones it knows are likely to have problems, and keeping the good ones for its own account
6) The future direction of interest rates, which will influence payment levels required for future time periods (for variable-rate mortgages)
…and probably at least 5 or so additional key factors.
The actual set of market variables and the stuff that’s well delineated is easily handled by a machine that deals with Go like games. Getting it to do that is much harder.
So is it easy or hard? It seems to me that it’s very hard, because if the kinds of simulations we’re discussing were doable then the kinds of people who are programming Big Blue and similar systems would be making significant progress with them. They are not. No one is close. There is much progress with short-term simulations and data mining, but model-driven long-term predictions are still mostly wildly inaccurate.
“if the kinds of simulations we’re discussing were doable then the kinds of people who are programming Big Blue and similar systems would be making significant progress with them.”
If you say so. It does not follow that people interested in AI and gaming would care about making something to do what you want. Au contraire.
Big Blue is not up to the task anyway, Deep Mind may be.
If you say so. It does not follow that people interested in AI and gaming would care about making something to do what you want.
I remember listening to a researcher at MIT talking about how difficult even basic AI really is. As an illustration, he talked about how hard it is to write the programming for where a robot should choose to walk.
Sidewalks are white/grey/buff concrete. Roads are mottled grey/black asphalt. Easy!
Except some sidewalks are asphalt.
And some roads are concrete.
And some roads and sidewalks are brick or stone of various hues.
And some roads and walkways are unpaved.
And some roads are wide and some are are narrow. And some walkways are wide and some are narrow.
Where should I walk? One solution is to look at where the vehicles are moving and define that as a road. Same for walkways. But HOW do the people know where to walk and HOW do the divers know which are the roads? Hard question.
Some people walk in the road. Some vehicles – tractors, mowers, forklifts – are on the sidewalks and some are not on either, they’re in fields or lawns or loading areas. It ends up becoming a complex relational and pattern analysis problem. It makes you realize how much complexity the human mind deals with on a continuous, casual, and nearly instantaneous basis.
David:
WRT Doerner’s simulations, perhaps a good heuristic in many situations is simply to slow down the speed of one’s decisionmaking. This gives more time for the decisionmaker to incorporate feedback, including the unintended consequences of past decisions, into future decisions. Other things equal, one might expect experienced decisionmakers to be slower and more deliberate in their decisionmaking than, say, college students would be. Of course, slow-and-deliberate conflicts with the ethos of “bold, persistent experimentation” that tends to characterize leftist (and, to a lesser degree, non leftist) idealists.
I’ve come to the conclusion that Deep Mind doesn’t mark an advance in artificial intelligence as much as a sad decline in human ingenuity. Go masters of yesteryear would’ve wiped the floor with its silicon innards. The guys we have now simply choked.
Simulations can also cause damage if they reinforce the wrong behavior patterns, or if their use limits the scope of training in dangerous ways. The (very expensive) high-end simulators used for airline pilot training apparently did not until very recently provide accurate modeling of airplane behavior in a full aerodynamic stall; this contributed to training protocols which emphasized avoiding stalls and recovering from them before they were fully developed, but *not* recovering from a fully-developed stall…and apparently the airlines did not do, or the FAA require, full-stall recovery practice in the actual airplanes.
Some interesting links on this:
http://aviationweek.com/aerospace-manufacturing/faa-nasa-and-industry-team-improve-stall-simulation
http://aviationweek.com/awin/faa-readies-stall-capable-simulators
http://www.faa.gov/pilots/training/media/Evaluation_of_Stall_Models_for_Training.pdf
Also, an interesting piece on the role training and simulation may have played in the Colgan Airways crash:
http://fl250.blogspot.com/2009/05/thirty-seconds-of-confusion.html
“I’ve come to the conclusion that Deep Mind doesn’t mark an advance in artificial intelligence as much as a sad decline in human ingenuity. Go masters of yesteryear would’ve wiped the floor with its silicon innards. The guys we have now simply choked.”
ROTFLMFAO
Oh wow, thanks. Umm …. no.
Ignorance at this level is best left to the experts.
https://www.youtube.com/channel/UCP7jMXSY2xbc3KCAE0MHQ-A
Michael:
You misinterpret my usage of the term “chaotic”… And do so in a manner suggesting limited grasp of the branch of mathematics called ‘Chaos Theory’.
The ramifications of CT towards your proposed system are enormous. We have learned somewhat about so-called “chaotic systems”, but it is still a study in its infancy. Its issues, as applicaple to your proposal, are many decades away from useful answers. We will have good climate modelling long before your proposed system is even vaguely possible to provide anything but rudimentary understanding of or answers about the effects of “policy x”.
And far and away THE greatest danger lies in some government asshole claiming it functional and reliable long before it is, and rejecting opposition because… SCIENCE!!!
Pretty much the whole AGW thing but FAR worse because it’s not some vague thing like “climate” being discussed, but PEOPLES’ LIVES.
While pilots will ideally never be in the position of having an AOA higher than the stick-shaker activation, upset conditions or improper inputs can send an aircraft into the stall realm or beyond, a state pilots have never experienced and do not know how to recover from.
The Air France Flight 447 stall was not dealt with properly. Airbus 330 must have a simulator. What do they run on it ?
Air France has denied that its pilots were incompetent, but has since improved training, concentrating on how to fly a plane manually when there is a stall.
Both Air France and Airbus are facing manslaughter charges, with a judicial investigation led by Paris judges under way.
It’s about time. Airbuses are all “fly by wire.” I wonder how that affects stall recovery.
We should also give credit to Frederick Lindemann, who learn how to recover from a stall by doing it before anyone else knew how.
Well, technically it was a spin but the idea was there.
Among them was Frederick Lindemann, a brilliant but highly opinionated physicist who would become Winston Churchill’s most trusted scientific adviser during World War II.
At Farnborough Lindemann wore a business suit and stiff winged collar even in the cockpit. Despite poor eyesight and only 50 hours’ piloting experience, he took on the challenge of figuring out the science of the spin. In the summer of 1917, he made more than a dozen flights in B.E.2 two-seaters, doggedly repeating spin after spin over a camera obscura (or pinhole camera), which helped track the airplane’s twisting path and the movements of its control surfaces. As he whirled around, Lindemann memorized readings from the B.E.2’s instruments, including a spring accelerometer fitted to his seat to measure G forces, then jotted the figures down each time he pulled out. The result was a landmark report issued in March 1918 that presented the first hard data on an airplane’s path and attitude during a spin and the stresses it imposed on the airframe. This was the forerunner of all subsequent research on spin aerodynamics, including work that NASA still continues today in its purpose-built spin tunnel at Langley Research Center.