Ricochet is the best place on the internet to discuss the issues of the day, either through commenting on posts or writing your own for our active and dynamic community in a fully moderated environment. In addition, the Ricochet Audio Network offers over 50 original podcasts with new episodes released every day.
Rob Long’s Data-Driven Utopian Dream
In the first 15 minutes of the latest Ricochet podcast (Episode #483), Rob said a couple of things that caught my attention. At one point, when talking about our communication- and data-centric technical culture, he suggested that the answers to all our big problems were probably in the wealth of data we’ve collected.
What came to my mind when he said that was the movie WarGames (1983), in which a wayward defense computer is discouraged from initiating Armageddon when it crunches the numbers and concludes that there’s no way to win a nuclear war. Setting aside the question of whether or not that’s a correct conclusion (and I recently re-re-re-watched Dr. Strangelove, in which Buck Turgidson makes a compelling contrary argument, so I’m really not so sure), what the computer in WarGames did was reach a kind of meta-conclusion. A thorough examination of the available information suggested that no good answers could be found.
If our “biggest problems,” whatever those are, can be “solved” by the application of ever more refined policy driven by ever more detailed knowledge, then perhaps Rob is right. But if, as I think is more likely, our biggest problems are complex — complex in a formal, and not casual, sense — then all the vast trove of linked and correlated factoids buried in the deep dungeons of Google’s digital fortress will, upon examination, yield not solutions but misplaced hopes for the next generation of technocratic central planners.
On the other hand, maybe all that data mining will lead to the same sensible conclusion reached by WarGames’ WOPR: that, despite all our data and all our brains, there’s no way to plan our way out of complexity’s maze. I’d be down with that meta-conclusion.
A few minutes later, Rob does it again, when he imagines New York City’s congestion and gridlock optimally managed by a centralized oracular database. As Rob sees it, all that’s missing is a knowledge of the destination toward which each driver is bound: given that, it’s merely a matter of crunching the numbers to determine the optimum trajectory for every vehicle, and voilà, unruly chaos is once again brought to heel by superior math skills.
That sounds nice, but it isn’t without its problems.
First, the specific task of optimal routing falls into a very large class of problems that are believed to be non-polynomial, or NP, in their complexity. This means that, as the number of parts of the problem grow, the difficulty of solving the problem grows much faster. Practically speaking, many, perhaps most, NP problems are not considered strictly solvable computationally; at best, decent approximations can be made.
It’s easy to imagine that anything approaching an optimal solution to New York City traffic routing implemented at the level of individual commuters would be so computationally intensive that one would be better off simply walking while the computers worked on the problem.
But this is all begging the question, because Rob was mistaken in his original assumption that knowing the destinations of all those millions of individuals is sufficient. It isn’t. One also has to know the urgency, the priority, the value placed on punctuality, of each of those millions of individuals. Otherwise, the thousand women off to buy new hats will likely have their transportation needs met at the expense of the three surgeons off to perform emergency transplants. Etc.
What Rob seems to be imagining is that central planning for traffic, lacking as it does the knowledge of the market, will nonetheless be better than central planning for anything else.
What the analysis of big data reveals is that simple rules govern independent actors, and that complex behavior arises from the interplay of those independent actors following their simple, but personalized, rules. That’s useful knowledge — and most useful if it discourages technocratic interventions in complex systems.
Published in Technology
Best conclusion of the day.
If James Lileks @jlileks doesn’t pick this for Post of the Week…..
The problem is dynamic and complex and cannot be optimized. But it doesn’t mean it cannot be improved. Traffic light systems, rerouting and future road construction are all steps that can be taken as a result of data.
Absolutely. And that’s the difference between viewing problem-solving as evolutionary, versus revolutionary.
By the way, I don’t think Rob Long actually imagines that anything can be perfected, and that utopia lurks in the data. I think he’s smarter than that, and more traditional in his sensibilities. I’m responding to what he said, because a lot of people actually believe that sort of thing. I don’t think he does.
One of the problems with trying to game something out in terms of traffic patterns is you have to work in the reality that a lot of people want to do the same thing in a very narrow time frame, and then you have to decide if the cost-benefit is worth the expense. Hotels have to have water heating capacity large enough to deal with the reality that most everyone wants to take a shower during a two-hour period in the mornings. The other 22 hours are spare, unneeded capacity, but if your customers get cold water at 8 a.m. because all the 7:30 a.m. showerers used the hot water up, no one’s coming back to your hotel.
That’s why you see cities periodically trying to get the companies there to stagger their work hours, in order to lower the volume of traffic at the height of rush hour. The other option, more highways or more mass transit vehicles, either isn’t economically viable or would face NIMBY opposition. That’s leaves either trying to voluntarily lower the number of people working the same time shifts, or using the power of the state to coerce people into certain modes of transportation (Bloomberg as mayor was obsessed in his later years with creating congestion pricing, where cameras would photo vehicle license plates in traffic south of 96th Street in Manhattan, and then toll them if they’re there during certain hours of the day. Cuomo’s taken it up now, so it may eventually become law — it’s along the lines of states like Texas tolling HOV lanes for people without 1-2 passengers, but it’s still annoying in taking something that’s been free since the dawn of the automobile and now making you pay for it).
Outstanding post. You call attention to one of the pre-eminent superstitions of our day, although it’s been around since that willful idiot Descartes. A little knowledge can be dangerous, and most people have only a superficial knowledge of the physical world. Every scale of reality emerges from interactions at a smaller scale, and exhibits distinct characteristics that cannot be predicted except by watching the system run.
I think a modern characterization of that common superstition is that one may predict behavior at one scale by applying rules learned in another.
Misplaced charity. He said it, he made reasoning based on it, why would you think he doesn’t believe it?
As a kid I loved Isaac Asimov’s science fiction. He posited “psychohistory” which would predict the future of human events through statistical analysis. Much more complicated than that, but just another example of 20th century dreaming of the ability to master human behavior, which I would say has been close to completely discredited as an idea. Unless you are an unrepentant leftist, in which case we just haven’t done it right yet.
Besides, data can’t drive. It can’t even be the navigator, though it can certainly be useful.
I have a fundamental disagreement with the terms “data-driven” and “science-driven,” even though sometimes people just use those terms because they are trendy. I once did some technical volunteer work for a project that had “data-driven” in its name. I thought the project was a good one, even though I thought the name was wrong.
Excellent observation. And at the heart of the problem between dreamers and doers. Or as a friend likes to say, worker bees and butterflies.
No, I don’t think it’s misplaced. While I often disagree with Rob, he’s never given me the impression that he believes complex processes can be managed, nor that humans are particularly good at doing so when they try. I think he was just riffing on the remarkable potential of networks and our modern data-gathering capabilities, and not really trying to make a serious point.
Oh, and thanks for the nice comment. ;)
Yes. The Foundation series, which I loved when I was a kid (Asimov was my favorite author when I was growing up), is unreadable dreck now, both in terms of literary quality (okay, that’s just Asimov) and philosophically.
But then, Asimov was a self-declared socialist — a lot like Bernie, really. So, surprise.
If I was a Google exec I would hack that system to send everyone away from the route I would like to take. The political elites would want in on that action. In no time the serfs would be diverted off Manhattan unable to direct their own vehicles.
Or perhaps (and here is where things get interesting), if you were a big IT company that provided dynamic routing data to millions of drivers, and that also had access to social media posts, email, and other information about those drivers, then maybe some of your overzealous employees would get together and agree to make “unfortunate” routing decisions for a few tens of thousands of critical swing-state voters on election day as they leave work and head to their polling places.
Interesting comment, Henry. I think, however, that you have an underlying libertarian assumption. You appear to assume that all of our problems will be solved if we find an efficient way to allow everybody to do whatever they want to do.
I was listening to an audio podcast of an old C.S. Lewis essay (or perhaps broadcast), and he used an interesting analogy for morality. The analogy was to consider our society to be a fleet of ships. The proper functioning of the fleet requires effective operation at three different levels:
We have a great deal of trouble, at present, even achieving agreement as to the first level.
Asimov was also something of a luddite. Over his whole life, he only traveled twice by air, as far as I can find, both as a young man while serving in the military so it wasn’t his choice.
In the earlier decades of the past century, when basically nobody knew all that much about computers etc, he could get away with a lot of what even then was basically nonsense. But looking back on it now, it’s pretty bad. Even unforgivably so.
Absolutely not.
In fact, I think that complex systems are not amenable to solution. The best we can do is achieve reasonable compromise. I wasn’t suggesting that a free market will achieve perfect results, but rather that Rob’s assertion that knowing all of the destinations will somehow be sufficient to achieve some idealized outcome is mistaken.
Maybe he assumes some additional rating based on the destination? Women buying hats are not going to Mercy Hospital, and doctors needed for emergency surgery are not going to Macy’s.
But then the women figure out to select a destination of a hospital that is NEAR Macy’s…
If he were a good technocrat (which I don’t think he is), he would call for an ever increasing, ever more invasive data gathering process, in order to optimize the management of this chaotic system.
Ultimately, of course, we can reduce the degree of chaotic unpredictability by imposing restrictions on behavior. More than one way to skin a cat, after all.
You’d love Mises and Hayek. They actually proved scientifically what you are saying.
What, you think I figure this stuff out myself? I was an avid reader in my misspent youth.
Top down notion is insane. Of course it’s possible that at least some human problems for fairly large populations, (a few thousand perhaps) have a top down technical data management answer that might be functional for a few seconds. After that everything has changed.
Actually that’s what I was hoping. Thanks for not being offended.
Henry,
Your observations are so very important. From Musk’s self-driving car to Boeing’s commitment to AI our faith in big data big complexity systems is a fool’s game. Unfortunately, we may be seeing this play out in the most awful terms with China’s response to the Coronavirus. Their top-down closed access model may be up against forces of nature that won’t be ruled that way. They seem to be trying to force the round peg into the square hole and now they hope a bigger hammer will help.
China’s Xi urges more policing as virus toll rises
Destroying the free flow of information will only increase paranoia and rumors. This will disrupt efforts to deal with the virus rather than enhance them. The title to Hayek’s book may take on a new and terrible meaning.
The Fatal Conceit
Regards,
Jim
Yes, but do you think Eric Holder would give up his motorcade w/ police escort if his minions made it easy and hacked the system in his favor? The point is not to divert people away, but to force them to wait at blocked intersections and watch him go by while he reads a book behind dark glass. There are some serious chick-attraction points in a display of power like that.
Eldar Ryazanov’s film, Forgotten Melody for a Flute (Забытая мелодия для флейты) gets at part of this by using radio communication to have regular traffic, including an ambulance making an emergency run, held up while a near-politburo member’s chauffeured car has free passage through Moscow’s streets. But in that scene the big guy sits alone in the back seat of the car. There is no motorcade. You can hack all the data systems you want, but if there is no motorcade with police escort, it’s not real power.
More data doesnt equal more science. Look at climate “Science” its got all kinds of data to look at, but they end up selecting the data points that support their conclusions and ignoring anything else. The data is used to support the theory instead of the theory describing the data.
The inaccuracies with their predictions are not caused by the data selection, its caused by a lack of fundamental understanding in how the atmosphere works and how it distributes heat – until we understand it and model it correctly more data will not help.
I do think there is a lot broken in climate science right now. I think it’s an open question whether any amount of data and solid theory would allow long-term predictions to be made with significant accuracy. Chaotic systems defy reliable modeling.
In other words, it isn’t a matter of weak data or bad models: even if those problems are addressed, complexity frustrates prognostication.
This is why I like maps. I already know the alternate routes and don’t need no stinkin’ app to guide me. ;)
In an older episode of Mad Dogs and Englishmen, Kevin Williamson mentioned Soviet programs in what they called Cybernetics; this was the attempt to use computers to process data for more efficient central planning. Recently, people have bought into the same idea being possible because of increases in computing power. Williamson points out that people who believe that computing will lead to better central planning have it exactly backwards: while we have more computing power, that power has allowed for more and more transactions to occur faster and faster. The amount of new data greatly exceeds our ability to process it.
Well, Skynet took the same data as WOPR and came to a different conclusion, it also solved the traffic congestion problem.
And, No matter how much data might already be collected, is no guarantee that it is the CORRECT data, and SUFFICIENT to develop a solution to a given problem even if we were successful in coming up with some kind of formula to use it.
As I was reminded in another member post, the current study of cancer, for example, is almost entirely about people who are diagnosed with and perhaps likely die from cancer, especially if currently-available treatment is unsuccessful. But there is little to no research done on people who die from other causes, who also had cancer. So the total amount of information available on cancer is actually rather limited and focused in one or a few directions.
It may well be that highly successful treatments for cancer could never be developed based on the limited information currently available, even if someone came up with good methods to analyze it. A lot of the potential data is simply missing.
It also reminds me of the claims especially from the left, that voter fraud – one of the left’s favorite recreational activities – is not really a problem. But the left doesn’t LOOK FOR voter fraud, and doesn’t want anyone else to look for it either. So the lack of widespread evidence is not due to voter fraud not existing, it’s basically because it is rarely looked for. And when voter fraud is found, it usually is a result of checking things that are somewhat incidental, not because there are structures and measures in place to look for it.
For example, the number of mailed voter registration confirmations – including same-day registrations in places that allow it – which come back “no such address” or where the address is a factory or warehouse, or some building that has been vacant for years or even demolished, etc.
That doesn’t happen because a Department Of Voter Fraud is looking for them. They just happen because the Post Office will return undeliverable mail. But even if someone were checking, it’s too late: the votes were already counted.