A Dispatch From The Future
October 29, 2014 1:29 PM   Subscribe

Evgeny Morozov writes for the New Yorker: The Planning Machine, on Project Cybersyn (previously) and Big Data.
Greg Grandin in The Nation responds: The Anti-Socialist Origins of Big Data
posted by the man of twists and turns (46 comments total) 25 users marked this as a favorite
 


Morozov is a quasi-plagiarist. His not-a-review heavily borrows from and barely aknowledges Eden Medina's book.
posted by kandinski at 2:52 PM on October 29, 2014 [2 favorites]


His not-a-review heavily borrows from and barely aknowledges Eden Medina's book.

but he does acknowledge Medina's book and quotes from it in the article. This is a really silly "hot take" on an article which could have been otherwise interesting.

If you follow the dots: if "big data" really allows corporations to calculate consumer demand as finely as the proponents claim, than that provides the counter-argument to the classic argument against planned economies i.e. than you could never replace the price system as a way of computing all of the demand inputs into the economics system.

But the article itself seems to miss it's own point. I don't know how much of that was Morozov or the editors at the New Yorker.
posted by ennui.bz at 3:17 PM on October 29, 2014 [3 favorites]


But the article itself seems to miss its own point.

That's more or less the Morozov modus operandi, isn't it? Spew out a bunch of suggestive stuff, much of it drawn from/repeating other people's work, and fail to connect the dots or draw out the really interesting consequences?

If you follow the dots [...] that provides the counter-argument to the classic argument against planned economies

Yeah, what the computer revolution means for the old planning argument is sort of a hot conversation topic among lefty theory types at the moment. Some of the Jacobin folks are coincidentally working on a piece on Cybersyn and big data that I'll be looking forward to reading rather than dwelling on the continual sort of disappointing failing-upwards that is Morozov's work.
posted by RogerB at 3:43 PM on October 29, 2014 [4 favorites]


The "cybernetic socialism" argument is basically that you can solve the central planning / resource distribution problem if you just throw enough computer power at it, and feed in enough information. It's a really nice thought, but there are a bunch of reasons to be pretty skeptical.

First, because there's an obvious path towards a No True Scotsman: anytime a cybernetic-governance scheme fails, its proponents will doubtless claim that the technology just wasn't there. That... doesn't inspire a lot of confidence. Just because computers have gotten faster and cheaper since the 1970s doesn't mean they're fundamentally different, and it's not clear that experiments like Chile's failed for lack of computing power. I'm not sure we'd do much better today.

Second, and more seriously, computers are very good at collating and processing data, granted. But they can't tell you what data is important to measure in the first place, much less what to optimize for. The decisions of what KPIs to measure and optimize for are really important, and represent in many cases huge assumptions about what outcomes are desirable. Even in the relatively unplanned, market economy of the US, we can't reach agreement on whether we should be optimizing for low unemployment or low inflation.

Enterprise Resource Planning (ERP) systems—which are the admittedly unsexy, modern analog to the 60s/70s dream of cybernetics—have one of the highest implementation failure rates in the IT industry. Something like 50% of ERP implementation projects fail, in terms of not even meeting their owners' self-imposed goals. That's miserable. We are not nearly good enough at this stuff to consider running an economy that way. We are barely good enough to run private enterprises that way, and (as basically anyone who has worked in a company with a big tentacles-into-everything ERP system can tell you) often the system doesn't do much useful anyway.

I'm not sure that you can reliably point towards a trend of computer-assisted "central planning" taking over in the private sector, despite the popularity of "big data" as a buzzword in the last few years. If anything, big companies have instead turned less towards central planning and more towards price signals (e.g. internal rebilling, "cost centers", monopoly-money budgets for internal spend) in order to allow greater degrees of flexibility.

Where you see a lot of "big data" type stuff it tends to be externally focused to the company implementing it. Amazon has a ridiculous amount of order data not only, and probably not even principally, so it can optimize its own warehouses* and internal order flow, but so that it can know more about you, the customer, and drive more sales. It's less akin to a socialist central-planning system than an intelligence agency.

* They use the data to optimize their own warehouses, but if they just wanted to do that there are easier ways than low-level data gathering. The traditional way is to identify a KPI, tie it to compensation, and let your employees figure out how to optimize the fuck out of it. Best hope you picked a good KPI.
posted by Kadin2048 at 3:49 PM on October 29, 2014 [16 favorites]


Funny, they made extensive reference to Cybersyn in the story Degrees of Freedom in Hieroglyph: Stories and Visions for a Better Future. Neat stuff.

I'm still waiting for someone to make http://wegetit.com - powerful idea when coupled with something like cybersyn or its erstwhile modern offspring.
posted by redbeard at 3:54 PM on October 29, 2014


His not-a-review heavily borrows from and barely aknowledges Eden Medina's book.

While I can't know either way, I have yet to see any remotely convincing evidence that this is true. Meanwhile, Morozov's responses to his critics list numerous specific other sources on which he also relied – yet that Lee Vinsel post, which seems to be the main public charge-sheet against him, calls this "utterly beside the point". It might be true or false, but it's obviously not "beside the point"; it is the point.

Personally, I also didn't take Morozov's description of Medina's book as "entertaining" to be a put-down. I suppose I can see how you could interpret it that way – but being entertaining is a high goal of good writing and one that many academics signally fail to achieve.
posted by oliverburkeman at 3:54 PM on October 29, 2014 [2 favorites]


I read the article earlier this month and thought it was an interesting narrative. It seems unlikely the New Yorker would somehow publish a plagiarized article. I also didnt get that the author was denigrating anyone...
posted by Nevin at 4:17 PM on October 29, 2014


(Perhaps the Jacobin writers are being overly sensitive about something they consider *their* topic, much like hipsters are protective of an indie band that has been embraced by the mainstream... They got scooped!)
posted by Nevin at 4:19 PM on October 29, 2014


that provides the counter-argument to the classic argument against planned economies

Planned economies were about providing for people's needs, not bespoke-tailoring their exploitation.
posted by mhoye at 4:20 PM on October 29, 2014 [1 favorite]


I'm always amused by any argument that "now computers are powerful enough to organize human endeavor, no, now computers are powerful enough, no NOW..."

Centralized planning of anything involving masses of humans exercising their own thinking and decision-making and action is absurd. Even in one tiny aspect of human affairs, we can only get decent but far from infallible elections predictions these days.

If the dream of the 90's has settled in Portland, the dream of the 30's has settled at the Jacobin.
posted by twsf at 4:21 PM on October 29, 2014 [3 favorites]


So is there some theoretical argument that computers can never manage a (at least partially) planned economy?

It seems like most of the arguments I've seen are based on the same set of failed efforts from the 70s and 80s. But the main reason people are re-evaluating the question is having seen what's been done in machine learning in the last 5-10 years in particular, where a number of problems that were assumed insoluble or only micro-incrementally improvable have been abruptly advanced by new methods. Machine vision, translation, speech processing, unsupervised learning, Watson-type deep learning, robotics -- all of these have seen qualitative breakthroughs in recent years, and frankly many of them seem *harder* than merely directing the flow of a few thousand commodities. This isn't remotely to say it's an easy problem, but a 50% failure rate of ERPs tells us nothing -- most of these hard problems continue to have high failure rates, but the key thing is that their success rates are no longer 0%. Like many of the other rapidly unraveling verities of modern economics (eg, in macro), it seems like this one is well worth revisiting. Let's see some simulations!
posted by chortly at 4:50 PM on October 29, 2014 [3 favorites]


Something like 50% of ERP implementation projects fail, in terms of not even meeting their owners' self-imposed goals. That's miserable. We are not nearly good enough at this stuff to consider running an economy that way. We are barely good enough to run private enterprises that way, and (as basically anyone who has worked in a company with a big tentacles-into-everything ERP system can tell you) often the system doesn't do much useful anyway.

My experience with electronic medical record system implementations (which are a lot like ERP installs, except I don't know that anybody ever died from a botched Oracle implementation) is the same. The failure point is not that the technology needs more processing power or better data-mining algorithms. The failure point is that any attempt to model reality in software is a simplification, and the details that get left out because management doesn't think they're important always turn out to important in some way that nobody thought about. Also, as soon as you define a metric that you're trying to optimize for, the managers who are responsible for it start figuring out how to game it. Also, project management in a large organization with competing constituencies is incredibly difficult.

All of those things are problems that are going to be multiple orders of magnitude worse in a system the size of a country than in a hospital or a business, and they are not problems that are amenable to algorithmic solutions - they are inherently problems of politics and value judgments.
posted by strangely stunted trees at 5:01 PM on October 29, 2014 [11 favorites]


Also, as soon as you define a metric that you're trying to optimize for, the managers who are responsible for it start figuring out how to game it

Campbell's Law!
posted by the man of twists and turns at 5:05 PM on October 29, 2014 [5 favorites]


These would seem to be general arguments against any complex simulation ("any attempt to model reality in software is a simplification and the details that get left out because management doesn't think they're important always turn out to important in some way that nobody thought about") and/or any attempt to systematically influence human behavior ("as soon as you define a metric that you're trying to optimize for, the managers who are responsible for it start figuring out how to game it."). But we successfully simulate complex systems all the time (eg, in all sorts of physical processes) and we optimize human behavior all the time via a vast system of laws and social policies, which while imperfect, are not hopelessly unpredictable in their outcomes. Campbell's law is so general it would seem to imply that outlawing murder or distributing food stamps or any other thing a government does will necessarily fail because it will be hopelessly gamed. Which I presume is a straw man, but I don't see what the more sophisticated version of the argument is, at least as it pertains to (partially) directed economies.
posted by chortly at 5:18 PM on October 29, 2014 [4 favorites]


yeah, the happiness measure and the like seemed remarkably naive— are you going to be honest when if you say you aren't happy, you might be dragged off to prison by the secret police?

to the extent that any of this relies on people not trying to game it, it will fail. GIGO, also.
posted by Maias at 5:18 PM on October 29, 2014 [1 favorite]


While I can't know either way, I have yet to see any remotely convincing evidence that this is true. Meanwhile, Morozov's responses to his critics list numerous specific other sources on which he also relied...

Nothing in the stories of Beer and Flores Morozov tells in the New Yorker piece can't be found in Medina's book.
posted by The Bridge on the River Kai Ryssdal at 6:04 PM on October 29, 2014


It would have been one thing if he'd been "yeah it's mostly based on Medina's framework but also I brought in some different stuff, here is exactly what it is" or "yeah it's based on Medina's book sorry that wasn't more clear" but doubling down on "it's mine, here's photos of my research, she didn't even read this stuff" when–incidentally–the story you end up telling and your conclusions about its meaning duplicate a recent, award-winning book, like, that's weird.
posted by The Bridge on the River Kai Ryssdal at 6:15 PM on October 29, 2014 [1 favorite]


rovides the counter-argument to the classic argument against planned economies i.e. than you could never replace the price system as a way of computing all of the demand inputs into the economics system

A simpler counter-argument is that the growth of large businesses means that there is no available choice between a planned and unplanned economy. The only choice is whether the planners are going to work for the government, a small group of large firms, or both.
posted by ROU_Xenophobe at 6:24 PM on October 29, 2014 [5 favorites]


"Also, as soon as you define a metric that you're trying to optimize for, the managers who are responsible for it start figuring out how to game it

Campbell's Law!
"

Ha! Usually I'm the one shouting that from the rooftops.

"But we successfully simulate complex systems all the time (eg, in all sorts of physical processes) and we optimize human behavior all the time via a vast system of laws and social policies, which while imperfect, are not hopelessly unpredictable in their outcomes. Campbell's law is so general it would seem to imply that outlawing murder or distributing food stamps or any other thing a government does will necessarily fail because it will be hopelessly gamed. Which I presume is a straw man, but I don't see what the more sophisticated version of the argument is, at least as it pertains to (partially) directed economies."

No, it's saying that measurements, especially proxy measurements, tied to incentives produce an increase in that measurement, not the underlying goal. For example, standardized testing as a proxy for teacher performance. Standardized tests are terrible for actually measuring teacher performance (even as they're pretty decent at measuring student performance), and when you tie an incentive (teacher pay/jobs) to that metric, you see a distortion to meet that metric, i.e. teaching to the test. Or to use one of your examples: Hunger in America is pretty hard to measure overall, though there are some ways to do it. However, it's easy to measure how many EBTs were given out. So suppose we want to decrease hunger as measured by the number of EBTs. If we tie a reward to that, managers will have an incentive to cut people off EBT rather than making sure people are fed.

Campbell's law is a specific form of the general economic truth that people respond to incentives.

The fundamental flaw in a lot of technocratic solutions is that a lot of social science or economic systems are both very difficult to measure accurately and dynamically responsive to being measured.
posted by klangklangston at 6:31 PM on October 29, 2014 [4 favorites]


Nothing in the stories of Beer and Flores Morozov tells in the New Yorker piece can't be found in Medina's book.

I'm definitely a bit biased here speaking as a journalist sensitive to the ease with which people throw around grave ethical accusations against journalists, but… are you saying, then, that (for example) Beer's letter to Mugabe is mentioned in Medina's book, when Morozov says it isn't? If so, wow. If not, this strikes me as more of the slightly hand-wavey "trust us, we're academics and the whole thing's a rip-off" tone that seems to have characterized the dispute. I should just read the book and judge for myself of course.
posted by oliverburkeman at 6:48 PM on October 29, 2014


Hunger in America is pretty hard to measure overall, though there are some ways to do it. However, it's easy to measure how many EBTs were given out. So suppose we want to decrease hunger as measured by the number of EBTs. If we tie a reward to that, managers will have an incentive to cut people off EBT rather than making sure people are fed. Campbell's law is a specific form of the general economic truth that people respond to incentives. The fundamental flaw in a lot of technocratic solutions is that a lot of social science or economic systems are both very difficult to measure accurately and dynamically responsive to being measured.

I guess this turns on what is specifically meant by "a lot". Every law is, in a sense, a technocratic solution -- food stamps, road tolls, cigarettes taxes, bans on alcohol for children, etc, etc. Sure, you can do it wrong, such as measuring hunger via EBTs distributed or education via multiple-choice questions. But I don't see what bearing such measurement problems have on the feasibility of managing complex systems using computers -- which we already do in hundreds of complex domains already, many of which (such as the thousands of laws on the books, or the management of cities or firms) include strategically responsive humans. We usually oppose standardized tests because they are a specifically terrible way to measure and encourage education, not because there is no possible way to measure or encourage education due to some inherent drawback in any incentive or measurement system.
posted by chortly at 6:58 PM on October 29, 2014 [4 favorites]


Just because computers have gotten faster and cheaper since the 1970s doesn't mean they're fundamentally different, and it's not clear that experiments like Chile's failed for lack of computing power.

It is more than a little likely that Chile's experiment failed because of US interference. It's on the record that the US government did everything short of sending soldiers in to carry out the coup directly to destabilize the country.
posted by idiopath at 7:05 PM on October 29, 2014 [6 favorites]


Can I just point out that Salvador Allende was a bad ass?
posted by signal at 7:06 PM on October 29, 2014


Kadin2048: Enterprise Resource Planning (ERP) systems—which are the admittedly unsexy, modern analog to the 60s/70s dream of cybernetics—have one of the highest implementation failure rates in the IT industry. Something like 50% of ERP implementation projects fail, in terms of not even meeting their owners' self-imposed goals. That's miserable. We are not nearly good enough at this stuff to consider running an economy that way. We are barely good enough to run private enterprises that way, and (as basically anyone who has worked in a company with a big tentacles-into-everything ERP system can tell you) often the system doesn't do much useful anyway.
As someone who has done implementations for a living, this is nearly always because management is loath to do process re-engineering. They try to customize the cheapest software they can get away to fit the way they're already doing business. Often including replicating their existing, paper processes for HR, procurement, et cetera. Thereby negating any gains they might have made by implementing an ERP and bringing the business into alignment with the often well-designed procedures embodied therein. (With the caveat that there are plenty of ERPs and similar software systems born in the batch processing era that are now trapped in amber and don't embody good procedures for a modern enterprise.)

However, I think you're correct in your larger point that current trends in "big data" are more akin to intelligence gathering rather than central planning. Much as I wish there were, I'm not sure there's a foolproof way to utilize computer analysis of the production problem for resource planning on the enterprise or national economic level, cf. the recent discussion here of Tracy-Widom distributions.
posted by ob1quixote at 7:15 PM on October 29, 2014 [5 favorites]


We usually oppose standardized tests because they are a specifically terrible way to measure and encourage education, not because there is no possible way to measure or encourage education due to some inherent drawback in any incentive or measurement system.

The fact that there might exist some hypothetical incentive or measurement system that might work doesn't really matter; what matters is that with what we have in front of us today, the performance that we have available to us, and the incentives that we could tie to them, would produce perverse or at least sub-optimal outcomes, and are therefore a bad idea.

It is, perhaps, possible in theory that a sufficiently powerful computer with access to sufficient data—in the limiting case, perhaps running its own model of the universe at some quantum scale, or a no-shit strong AI—could be a perfect economic-planning engine. (I'd personally still question this, because of course there are huge, gaping assumptions that occur when you choose what to optimize for, to that point that in choosing the optimization criteria you would in essence be doing the task that you want the computer to do. But whatever, we'll let that slide for now.) It doesn't matter. The fact that it might be hypothetically possible with some sufficiently advanced computer doesn't mean that what we have available today, or anytime even in the foreseeable future, is going to work well.

The examples of advances in machine vision, etc., while certainly boons for those particular fields, aren't directly applicable to resource-planning systems. There is a tremendous amount of interest in resource-planning systems, just for allocation within private industry, and if you could come up with a better ERP system—something that was quantitatively better than what everyone else has—you could probably hire Bill Gates as your shoeshine boy. But the technology isn't there. It's not really even close; it's not especially clear (as in, it's still a thing people are researching, subject to a fair bit of pressure from software vendors) that current-day ERP implementations actually work especially well at what they're intended to do (i.e. actual resource-allocation decisions).

The biggest gains attributed to them are frequently actually not due to the system facilitating centralized decisionmaking, but in allowing better communication between parts of a large enterprise, or between separate companies, so that the people who make decisions do so with better information. The centralized-decisionmaking stuff does happen, but the gains are small, because traditional hierarchical management does that roll-up stuff fairly well already.

All this ought to make sense if you stop and think about it, because even the most advanced computers we have today aren't nearly as smart as a person, so the idea of taking decisionmaking control out of the hands of people and putting it into a machine is slightly absurd. Using computers to better-inform people so that they can make better decisions, according to their own priorities, seems like it will be the better bet for the foreseeable future.

tl;dr: Resource allocations for even a large corporation, much less a country, are likely "AI-hard" problems.
posted by Kadin2048 at 7:21 PM on October 29, 2014 [3 favorites]


Is the claim that ERP has never gotten better, and can never work above some system size threshold? The previous post seemed to be saying that nearly 50% of such systems don't fail, so clearly it can work. But I suppose I'm unclear on what the force of the "AI-hard" or "Tracy-Widom" claims are here. If centralized decision-making does happen and offers "small" gains on traditional hierarchical management, then that shows that it can work. Indeed, many decisions computers make in certain domains are much better than humans, regardless of whether the latter are in some general sense "smarter". But in any case, the far-left dream of a command economy is not to be smarter than humans, or to be "foolproof". The aim is to work sufficiently well to allocate goods more fairly than capitalism does, without suffering a soviet-style collapse. The fact that ERP succeeds some of the time, is sometimes marginally better than traditional management, and seems to be steadily improving -- all suggest that it can indeed be done, at least on a modest scale.
posted by chortly at 7:41 PM on October 29, 2014 [3 favorites]


"Every law is, in a sense, a technocratic solution -- food stamps, road tolls, cigarettes taxes, bans on alcohol for children, etc, etc."

That's not true in any useful sense. Most laws in America have a grounding in English common law, and the origins of civil (or code) law in ancient systems like Hammurabi and Solon were not in any meaningful way technocratic solutions.

"But I don't see what bearing such measurement problems have on the feasibility of managing complex systems using computers -- which we already do in hundreds of complex domains already, many of which (such as the thousands of laws on the books, or the management of cities or firms) include strategically responsive humans. "

There's a really simple programming adage that sums it up: Garbage in, garbage out. Campbell's law points out that by tying incentives to metrics, you encourage garbage in. And, like I said, there are a lot of things in social science that aren't observable by direct measurement in the way they are with physical sciences. While you get some noise in, say, temperature measurements, it's nothing compared to things like self-reported happiness.

Look, I'm a person who generally favors institutional, technocratic solutions. But I'm also someone who has sifted through U.N. data on relative development levels, and been a U.S. census data collector, and with many things there just isn't a way to get concrete answers out of the abstractions — at every level, information is lost and without having a clear picture of what that information is, creating a program to effectively maximize it is impossible. Thinking otherwise is naive in the extreme.

This is even more true when you think about the biases and blindspots of institutions themselves.

"We usually oppose standardized tests because they are a specifically terrible way to measure and encourage education, not because there is no possible way to measure or encourage education due to some inherent drawback in any incentive or measurement system."

That's incoherent, and there are plenty of terrible reasons trotted out by opponents of standardized tests. Standardized tests are, by themselves, a pretty great way to measure both the intelligence of children and what they know, especially compared to legislatively codified curricula. The reason to oppose linking things like teacher salaries to child performance is because using standardized tests to measure teaching is a poor proxy that is extremely vulnerable to gaming.

Again, it's pretty simple.

1) People respond to incentives.
2) Many things, especially social markers but also economic behaviors, are really hard to measure.
3) Without being able to measure them directly, and with an incentive to manipulate the proxy measurement, people will manipulate the proxy and the measurement will cease to be a reliable proxy.

This doesn't mean that no systems can be made or that technocratic solutions are always wrong or anything of the sort. It's a well-known bias in systems that deal with reward-seeking behavior. It's similar to a related problem of forecasting, which is that humans tend to see trends as likely to continue. Or the general problem in markets that people favor short term reward over long term reward. It's structural.
posted by klangklangston at 7:42 PM on October 29, 2014 [3 favorites]


Can't favorite on my stupid tablet for some reason, but there are some really important insights upthread. Great discussion!
posted by saulgoodman at 7:44 PM on October 29, 2014


To get this back to the original topic, the goal of a command economy was originally to allocate food more fairly. I don't see why measuring food needs is impossibly hard, when we do similar things for food stamps, the earned income tax credit, medicaid, capital gains, and a thousand other tricky-to-measure things. Heck, we even know how to measure education well: get a bunch of well-paid expert teachers, and ask them to individually evaluate students. Sure, there are lots of problems in doing this stuff well -- people are evasive, bad at forecasting, strategic liars, and have quite a few other biases. Yet none of those constitute a demonstration that allocating resources among people is any more impossible than many of the millions of other things we already do to shape people's behavior using laws, rules, institutions, social conventions, and all the rest. We know that people are difficult but not impossible to influence. It's not clear that there's anything fundamentally different about controlling a complex system of people than controlling a complex system of any other sets of variables. Machine learning has made huge strides in these sorts of complex systems recently, and operations research is actually not that far from vision -- in fact, one of my textbooks has chapters on vision and OR only a few dozen pages from each other, both using similar neural networks. We already have a million rules that make the world a better place for abstract, evasive, lying, strategic, reward-seeking, uncertain humans. I have yet to see the argument why exporting fairly established techniques that at least moderately work within firms or cities to larger groups is necessarily impossible.
posted by chortly at 8:08 PM on October 29, 2014 [2 favorites]


It's not that it's impossible, it's that it's very, very difficult, and those difficulties to tend to be massively underestimated when people plan for these types of systems. To improve the success rate, the main barriers are social, political and managerial, not technological; they are the same barriers to implementing improvements to an economic system that exist when you are not trying to do it cybernetically.

So, yes, again, I wouldn't say that it's impossible, just that success would be an outcome of solving a lot of very hard problems rather than a means of solving them.
posted by strangely stunted trees at 8:28 PM on October 29, 2014 [3 favorites]


It's not that it's impossible, it's that it's very, very difficult, and those difficulties to tend to be massively underestimated when people plan for these types of systems.

Well, given that conventional wisdom is that it is outright impossible to run a command economy, getting it down to a hard problem that better ML might conceivably help a little with is, I suppose, progress.

(Not that this is something I personally am in favor of. It just struck me as one of these agreed-upon truths of economics that, like inflation-phobia or the laffer curve, largely seemed to depend on results that had been little updated since the 70s. But let's not rekindle those arguments too!)
posted by chortly at 8:49 PM on October 29, 2014 [2 favorites]


If the dream of the 90's has settled in Portland, the dream of the 30's has settled at the Jacobin.

jacobin hasn't published their editorial, though if it's some kind of apologia for a south american 70's models of state capitalism i'd eat my shoe. jacobin is one of the most interesting and thought-provoking journals that presently exists, imo
'the jacobin' tho lol
posted by p3on at 10:35 PM on October 29, 2014 [3 favorites]


chortly, I'm in your corner here. If some well-respected scientists and technologists think the Singularity is possible, surely the much more modest goal of rationally allocating resources throughout a society is possible. I mean, hell, Wal-Mart and Amazon must do this logistical calculation on a smaller scale already. Now, I'm not a believer in technological rapture like the Singularity fans, but I don't see why our capacity to use computers and math to model just about everything would spectacularly fail when it comes to the economy. Like you point out, we model much more mundane things and some significantly complex things all the time and no economist is jumping up and down with his can't-happen protests. I suspect the resistance to these ideas -- as with most things quasi-or-actually-socialist -- is ideologically based.
posted by Noisy Pink Bubbles at 11:21 PM on October 29, 2014


It's worth considering the entire idea of a centrally-planned economy and what its intrinsic strengths and weaknesses are before you insert a bunch of hypothetical technology into the mix, which will tend to exaggerate both.

Consider a widget factory operating in a simplistic price-signaled market economy: the factory produces and sells widgets, and if the demand for widgets exceeds supply, the price increases, prompting whoever is responsible for the factory to increase production (or, in a longer term, perhaps for someone else to go into the widget business). Given stable enough inputs, eventually you approach a steady state where supply and demand are matched and the firm's profits are driven down to a level governed by the cost of capital.

In a simplistic, centrally planned economy, the central planner--man, machine, whatever--looks at demand information which it receives directly, makes decisions about production, and transmits those decisions to the factory or factories, which produce the specified number of widgets. If the central planner did everything correctly, you arrive at the same end-state of supply and demand balance faster. That's the theory, anyway.

This second scenario has always seemed attractive, at least to certain types of folks (engineers seem particularly susceptible), because it lets you bypass all that wasteful, oscillatory hunting for the optimal production quantity--which gets worse and worse as you go along the supply chain, to the point where it can actually drive suppliers out of business with droughts and gluts--and jump right to the endgame. And to be fair, it does look like a pretty good arrangement.

But what makes it advantageous isn't the centralization per se. It's the information; it's being able to see demand data for the entire economy, rather than just what someone will pay for one of your widgets, and make a production decision based on that. But there's no particular reason why that process needs to be centralized. In the 19th and even early 20th century the available communications technologies would have almost demanded centralization, to take in all the demand and supply data from distant points and render it into something useful. Plus, how else to coordinate (a word that turns up a lot in discussions of central planning, sometimes ominously) the actions of multiple factories?

Modern communications allows much better visibility into demand indicators for the factory manager. Instead of basing their production decisions on the spot-market price for widgets, they have access to a wide variety of real-time demand indicators. The information that in a centrally-planned case would go to the central planner, is instead available to the factory managers, who attempt to make their own most-optimal production decisions taking it into account. In the limiting case, all demand and production data is available to everyone, and you jump right to the optimal outcome; it's nearly indistinguishable (in theory) from the centrally-planned case (in theory).

A lot of the gains attributed to ERP systems--ironically, much in big, heavy industries that were also targets for central planning--are due to that sort of information sharing. It's not so much that the computer makes a better decision, it's that the computer gives you greater visibility into information that you previously might not have had access to when making that same decision. The vision--rarely realized in practice--is that everyone's systems communicate and share data (suppliers, vendors, etc.) about each firm's internal state, so actions can be coordinated. At the same time, you buy industry-wide demand data from an aggregation/research service, which constantly pulls it in from whatever sources they can find and probably even does the modeling/forecasting for you, and that gets plugged into your forward plans. What you end up with is probably not very far off in terms of outcomes from what you'd get if the same thing were centrally-planned (again, except for the political question, which is outside the scope), and it's more resilient in some important ways (dealing with bad input data).

Overall, the evolution of technology over the past half-century has made central planning much less likely to ever take off, at least for economic efficiency reasons. If there was ever a sweet spot in history for central planning, it was probably in the era of the telex and pneumatic tube. Now you can get much the same results by encouraging information-sharing but letting the actual decisions happen on a fairly low level. That's where I see things going right now, and where they'll probably continue to head for the foreseeable future (as in, assuming no Singularity, Strong AI, etc.). A return to cybernetics or expert systems as originally conceived and attempted in the 60s and 70s seems very unlikely.
posted by Kadin2048 at 11:48 PM on October 29, 2014 [7 favorites]


Also,
conventional wisdom is that it is outright impossible to run a command economy
is certainly not true. At least, I've never heard anyone say it in those terms.

Command / centrally-planned economies seem to work rather inefficiently. They certainly do work. And even economies that consider themselves to be free markets (e.g. the US) are in some senses hybrids, with the "central planner" retaining some levers over the economy.

Plus there are obvious examples of command economies working, like the US economy during World War Two, but there I think even the staunchest defenders of that particular plan would not have argued that it was necessarily the most efficient way to produce goods, but rather the most expeditious method to arrive at a particular outcome (war production). That is certainly something that could happen again in the future, for a variety of reasons.
posted by Kadin2048 at 12:07 AM on October 30, 2014 [2 favorites]


One of the reasons the Soviet equivalent of Cybersyn never happened was a fear that it would work. The realisation was that the only way central planning would work would be with good data, that would have to be gathered remotely, and good communications to good central analysis and decision making. But for efficient local operation, the same arguments worked - you need to know about the other side of your inputs and outputs, as well as what's going on internal to your factory/farm/People's Ballet Company.

At which point you have well-informed local people with the ability - in fact, responsibility - to make their own decisions. Central control becomes looser. Unsurprisingly, there was a lot of opposition to this - to the extent that the Kremlin resistance to the idea leaked 'SOVIET COMPUTER TO MONITOR ALL CITIZENS' scare stories to the Guardian.

The ironies of this won't be lost on anyone involved in current (and passim) arguments about group autonomy within the organisation (a monster most recently spotted swimming in Loch Devops), nor to anyone behind the curtains in the NHS (described as the world's largest extant functioning Marxist organisation).

Upthread, someone commented that most failed ERP installations are due to companies not being bold enough in switching away from their manual systems to the best-practice inherent in modern software. In my experience, the problems often come from one-size-fits-all ERP parachuted into companies where the manual system has co-evolved with the corporate structure and method. It's like transplating a pig's heart into a crocodile on the grounds that mammals are far more functional than reptiles. I think the same arguments on both sides - you won't get the benefits unless you adapt to them/you won't get the benefits unless the system is adapted to task - would apply to any state-sized economic management system, but with the usual factorial increase in complexity viz component count.

What will happen, in ERP as in country/global economic planning/control, is that evolution will happen haphazardly and unpredictably. Cybergrokking large economic systems may be hard AI - it may be harder than that - but new technologies such as the decentralised anonymous ledger behind Bitcoiin have the ability to fundamentally restructure a lot of the assumptions behind how business is transacted. When that happens, especially when you have a lot of new tools sitting around, the potential for real innovation is high but its shape is particularly unpredictable.

The result of the Soviet fear of decentralisation (the argument against wasn't that it was escapable, but that it was undesirable) was that a lot of internal technology development was abandoned in favour of copying Western technology where it was deemed useful. The dangers of this approach are, I think, available for examination.

(Oh, and the passing comment that Cybersyn would at least bring full employment to graphic designers is truer than perhaps is generally realised - visualisation is the overlooked god that needs to be prayed to in Big Data, and if you have a talent for that prepare for the Malibu condo. Or shoebox in SF.)
posted by Devonian at 4:15 AM on October 30, 2014 [5 favorites]


Overall, the evolution of technology over the past half-century has made central planning much less likely to ever take off, at least for economic efficiency reasons.

Maybe central planning by governments, but central planning by Wal-Mart etc doesn't show many signs of running out of steam.

conventional wisdom is that it is outright impossible to run a command economy

I have to agree with Kadin2048 that command economies work okay, for limited purposes and with obvious problems. The US military is a command economy about as large as the economy of South Africa, Colombia, or the Phillippines.
posted by ROU_Xenophobe at 5:58 AM on October 30, 2014


are you saying, then, that (for example) Beer's letter to Mugabe is mentioned in Medina's book, when Morozov says it isn't?

Beer's letter to Mugabe doesn't appear in the book, it's true. The actual argumentative content of that paragraph in Morozov's piece--that Beer shopped Cybersyn derivatives around to other revolutionary-leaning/"third world" countries (and Canada!) post-Allende, but because of already existing bureaucratic institutions nobody was into it--appears in Medina's book with two different examples from programs (Uruguay and Mexico) that got further along than an unsolicited letter.
posted by The Bridge on the River Kai Ryssdal at 6:21 AM on October 30, 2014


conventional wisdom is that it is outright impossible to run a command economy
is certainly not true. At least, I've never heard anyone say it in those terms.


Well, the usual argument goes something like: the USSR tried central planning, the USSR collapsed, therefore central planning is bad, you dumb crypto-commie. Usually this argument comes from people of a more rightward bent (economists included).

I've never found this argument particularly convincing for a number of reasons:

1) Suppose we accept the assumption that the USSR's system was central planning based. If that is the case, the USSR was successful at growing economically -- sometimes at rates faster than the West -- for 70 or so years.
2) Of course, the USSR went through a variety of economic models -- war communism, NEP, five-year-plan/collectivization, etc. -- throughout its existence so the lack of nuance in this argument is unattractive.
3) We see a variety of successes (or at least existing systems) of what could be called central planning -- the military, Wal Mart, whatever in present societies, and they're not falling apart.
4) Economics was only one of the factors leading to the collapse of the USSR. And if economics played such an outsized role, how come the USSR didn't collapse earlier, under similar economic conditions?
5) Even if central planning led to a lower level of economic output, that might be a tradeoff that society might be willing to make, if other benefits were present (guaranteed employment, more vacation time, better working conditions, etc.).

So, yeah, I think the argument is more an appendage of right-wing triumphalism about the Cold War rather than a rigorous analysis.
posted by Noisy Pink Bubbles at 7:59 AM on October 30, 2014 [1 favorite]


I think the most straightforward response to that is the USSR's growth rates outpaced the West at some points in its history because the USSR started off less industrialized than the West.

When you start off with a pre-industrial landscape populated by peasants doing subsistence farming, you can get some pretty insane growth rates when you suddenly ram an industrial revolution through at gunpoint. (See also: China.) It comes at a disturbingly high human cost in pretty much every case where it has been tried, though.

Saying that the Soviet Union collapsed due to central planning is simplistic, and I don't think anyone is actually advancing that argument here so we don't really need to take it up. As the conclusion of one of the two great geopolitical struggles of the 20th century I don't think it's amenable to any sort of easy explanation.

central planning by Wal-Mart etc doesn't show many signs of running out of steam.

True. Although I'd like to take a second to appreciate the irony that the best working example of central economic planning that we can come up with today is Walmart.

But yeah, Walmart does do a good job of capturing certain economies of scale—largely, the ability to beat up on their suppliers; secondly, transportation—in a way that is difficult to do without that centralization. Where such economies of scale exist and offer a competitive advantage, you will certainly continue to see organizations built around them. But the centralization comes as a result of the economies of scale, not the other way around. And in a postindustrial economy, the number of places where there are significant economies of scale enough to justify the overhead of big central-planning organizations like Walmart's are probably the exception rather than the rule. In a lot of other situations, flexibility would be more important than tight coordination, and that leads to a very different and less centralized structure.

The real economy is a land of contrasts, &c.
posted by Kadin2048 at 10:59 AM on October 30, 2014 [1 favorite]


But what makes it advantageous isn't the centralization per se. It's the information; it's being able to see demand data for the entire economy, rather than just what someone will pay for one of your widgets, and make a production decision based on that. But there's no particular reason why that process needs to be centralized. ... Modern communications allows much better visibility into demand indicators for the factory manager. Instead of basing their production decisions on the spot-market price for widgets, they have access to a wide variety of real-time demand indicators. The information that in a centrally-planned case would go to the central planner, is instead available to the factory managers, who attempt to make their own most-optimal production decisions taking it into account. In the limiting case, all demand and production data is available to everyone, and you jump right to the optimal outcome; it's nearly indistinguishable (in theory) from the centrally-planned case (in theory).

So, in theory, a decentralized system of production based on cooperation (transparently sharing demand indicators and other data, instead of shielding it from competitors, so that individual production sites can make their own decisions) would combine the advantages of central planning and market economies while shedding many of their disadvantages?
posted by twirlip at 12:36 PM on October 30, 2014 [1 favorite]


"To get this back to the original topic, the goal of a command economy was originally to allocate food more fairly. I don't see why measuring food needs is impossibly hard, when we do similar things for food stamps, the earned income tax credit, medicaid, capital gains, and a thousand other tricky-to-measure things."

o_0

It takes an extreme combination of naivety and techno-optimism to think that capital gains taxes are an effective argument for a command economy. And the thing is that with food needs vs. EBT, poverty vs. the EITC, medical needs vs. medicaid, etc., we are both not actually doing that great a job in addressing the underlying problems and all of them are poor measurements of the underlying problems.

How would you actually go about measuring the food needs of a community? I've had to work through this on the board of a large, predominantly Section 8, housing co-op that provided food distribution for the community.

"Yet none of those constitute a demonstration that allocating resources among people is any more impossible than many of the millions of other things we already do to shape people's behavior using laws, rules, institutions, social conventions, and all the rest."

I still don't think you understand why this analogy doesn't support your point. It's substituting "Computers" for "A wizard did it." Laws, institutions, social conventions and "all the rest" are neither designed to work as command economies, nor have they been particularly effective, generally, in creating one.

"We know that people are difficult but not impossible to influence."

Yes, and we know that if we give them incentives to hit a measurement as a proxy for a policy objective, that they will focus on hitting the measure. This is seriously a well studied problem, with the fundamental formulation in 1977 still useful today.

"Machine learning has made huge strides in these sorts of complex systems recently, and operations research is actually not that far from vision -- in fact, one of my textbooks has chapters on vision and OR only a few dozen pages from each other, both using similar neural networks. We already have a million rules that make the world a better place for abstract, evasive, lying, strategic, reward-seeking, uncertain humans. I have yet to see the argument why exporting fairly established techniques that at least moderately work within firms or cities to larger groups is necessarily impossible."

No, you've yet to understand an argument on why quantification tied to reward encourages corruption. You have yet to address where the data comes from and how these systems can be built without being subject to the same corruption pressures as every other large-scale bureaucracy.

"chortly, I'm in your corner here. If some well-respected scientists and technologists think the Singularity is possible, surely the much more modest goal of rationally allocating resources throughout a society is possible."

1) The singularity is, right now, science fiction. That well respected scientists believe in it as a possibility (or inevitability) does not mean that it will come to pass.

2) "Rationally allocating resources" is not necessarily a coherent proposition. What does "rationally allocating resources" mean?

3) There are epistemological limits to utilitarianism that are being ignored here.

"I mean, hell, Wal-Mart and Amazon must do this logistical calculation on a smaller scale already. Now, I'm not a believer in technological rapture like the Singularity fans, but I don't see why our capacity to use computers and math to model just about everything would spectacularly fail when it comes to the economy. Like you point out, we model much more mundane things and some significantly complex things all the time and no economist is jumping up and down with his can't-happen protests. I suspect the resistance to these ideas -- as with most things quasi-or-actually-socialist -- is ideologically based."

Wal-Mart and Amazon are not command economies. Is the problem here a lack of understanding of terms? I mean, quick, what's the difference in goals between a privately-held company and the state? Can you see how that might influence the feasibility of "command economies"?

Wal-Mart has comparatively simple inputs next to a nation state, and even then the "centralized planning" is being overstated.

" Economics was only one of the factors leading to the collapse of the USSR. And if economics played such an outsized role, how come the USSR didn't collapse earlier, under similar economic conditions?"

Because it was a brutal dictatorship?

The USSR collapsed in large part due to the problems with ideological central planning. The very reason why Gorbachev got involved in higher government is because as an agricultural apparatchik he saw the mismatch between central demands and the local agricultural communities he was organizing, as well as the mismatch between the central reports and the local reports. Part of the Peristroika reforms was empowering people lower on the chain to make local decisions. There were a ton of other reasons of course, but it wasn't unrelated.

"But yeah, Walmart does do a good job of capturing certain economies of scale—largely, the ability to beat up on their suppliers; secondly, transportation—in a way that is difficult to do without that centralization."

This is one of those things that is getting lost here: Economies of scale are great, and are serious arguments for socialist models of public goods. Especially public goods that are hard to quantify, like health care. Single-payer is the way that the U.S. will eventually have to go to stave off health care collapse. And yeah, with that, computerized econometrics and efficiencies will be really helpful.

I'm annoyed at having the significant criticisms of central planning dismissed as right-wing fantods since I actually do support a socialist economy. However, there are fundamental, epistemological and philosophical problems with implementing control economies that have not been solved, and folks just waving the computers wand is not addressing that.
posted by klangklangston at 1:14 PM on October 30, 2014 [5 favorites]


There are also fundamental epistemological and philosophical problems implementing effective distributed decision-making systems. I.e. those that don't flap wildly, incubate systemic imbalances and suboptimal outcomes, randomly stall-out or disintegrate due to locally practical and globally horrible choices, local misinformation, bad time-horizons, missing feedback edges, collusion and rackets, etc.

People game markets just as hard as they game command economies. Markets fail just as spectacularly as command economies. Distributed systems have just as many misbehaviours as centralized ones. The spectrum between centralization and decentralization is not a spectrum between correctness and failure, in either direction. They are just different sorts of engineering, with different failure modes.
posted by ead at 9:14 PM on November 1, 2014 [1 favorite]


The USSR collapsed in large part due to the problems with ideological central planning.

I was reading this PDF ("A Reassessment of the Soviet Industrial Revolution") and the author would agree:
These judgements should not be read as an unqualified endorsement of the Soviet system. Dictatorship was and is a political model to be avoided. Collectivization and political repression were human catastrophes that brought at most meagre economic returns. The strength of central planning also contained the seeds of its own undoing, for it brought with it the need for someone to plan centrally. When plan objectives became misguided, as in the Brezhnev period, the system stagnated.
...
The interpretation of the Soviet decline offered here is the reverse of the analyses which emphasize incentive problems and the resulting failure of managers to act in accord with the plans. On the contrary, the plans were implemented; the problem was that they did not make sense. The strength of Soviet socialism was that great changes could be wrought by directives from the top. The expansion of heavy industry and the use of output targets and soft-budgets to direct firms were appropriate to the conditions of the 1930s, they were adopted quickly, and they led to rapid growth of investment and consumption. By the 1970s, the ratio of good decisions to bad was falling. Perhaps the greatest virtue of the market system is that no single individual is in charge of the economy, so no one has to contrive solutions to the challenges that continually arise. The early strength of the Soviet system became its great weakness as the economy stopped growing because of the failure of imagination at the top.
posted by Noisy Pink Bubbles at 9:50 AM on November 4, 2014 [1 favorite]


i think you can usually just sidestep the planned/market economy debate by reframing the (distractingly ideological) issue in terms of (more practical, but boring) monetary/fiscal/regulatory/tax policies.

brad delong's corporation as a command economy was pretty influential for me as is the notion of coase's penguin and what happens to cooperative/competitive organizational behavior dynamics when transaction costs approach zero. of course, not everything is free (yet) but we should be thinking about institutional adaptation -- both 'private' and 'public' (they were always 'mixed' but i'd say the separation is becoming more 'artificial' and blurred now) -- in terms of new realities for the spheres of production, coercion and cognition and not just try to preserve the status quo by unreflective fiat artificial scarcities.

in that regard 'the managerial revolution' is kind of fascinating to me -- cross reference with thomas piketty's 'super managers' -- and then see what's happening with technology, finance and central banking in terms of credit/resource allocation and reputation/trust incentives. that is, basically, how do you continue expanding the production possibility frontier (including environmental sustainability! and leisure ;) and how do you determine its fair division thru moral philosophy?
posted by kliuless at 11:26 AM on November 9, 2014 [3 favorites]


« Older The new terror   |   Indie Beef Newer »


This thread has been archived and is closed to new comments