[tt] NS: Why complex systems do better without us

[Premise Checker]

Why complex systems do better without us
http://www.newscientist.com/article.ns?id=mg19926681.500&print=true
8.8.6

* Mark Buchanan

WE HUMANS prefer the tidy to the untidy, the ordered to thedisordered. We 
like pristine geometrical regularity, and eschew whatis erratic and 
irregular. We want predictability and, more thananything, we want control.
In these confusing times, it might seem as if we have little powerover 
anything. Instead of letting it get us down, though, perhaps weshould take 
comfort from the work of Dirk Helbing, a physicist atthe Swiss Federal 
Institute of Technology (ETH) in Zurich. Helbinghas been studying the 
movement of tens of thousands of cars on roadnetworks; the workings of 
vast webs of interacting machines onfactory floors; and other systems, 
where the complexity of whathappens and why routinely defeats the human 
mind.
What Helbing and others are finding is that our penchant forregularity and 
control is seriously misguided. In many situationsthey are discovering 
that it is better to give up some of ourcontrol and let systems find their 
own solutions. Often the answersturn out to be unlike anything our minds 
would imagine, yet theoutcomes are far more efficient.
The findings come as something of a relief to today's engineers, whoare 
increasingly dealing with problems too complicated for them tosolve. Take 
one of the earliest successes chalked up by machinesallowed to take 
control.
Back in 1992, General Motors were having trouble managing theautomated 
painting of trucks at an assembly plant in Fort Wayne,Indiana. Machines in 
10 different paint booths could paint trucks asthey came off the line, but 
because the trucks came off in anunpredictable order and the painting 
machines needed sporadicmaintenance and repair, finding an efficient 
assignment of trucks tobooths seemed impossible.
General Motors' visionary engineer Dick Morley suggested letting 
thepainting machines find a schedule themselves. He set out some 
simplerules by which the various machines would "bid" for newly 
availablepaint jobs, trying their best to stay busy while taking account 
ofthe need for maintenance and so on. The results were remarkable, ifa 
little weird. The system saved General Motors more than $1 millioneach 
year in paint alone. Yet the line ran to a schedule that no onecould 
predict, made up on the fly by the machines themselves as theyresponded to 
emerging needs.
Production processes generally depend on so many inputs, parametersand 
factors that even small changes in the set-up can lead to wildlydifferent 
and unpredictable consequences. That is why it is almostimpossible to 
predict what will happen in a new production linebased on previous 
experience. "Managers sometimes take performancein past set-ups and try to 
estimate what will happen in a newsetting by interpolation," says Helbing. 
"This often gives very badresults."
To cope, he says, engineers need a healthy respect for the 
complexunpredictability of these systems and how natural human 
inclinationsoften lead to undesirable outcomes. "You can't steer these 
thingslike you can a bus," says Helbing. "You have to learn to use 
thesystem's own self-organising tendencies to your advantage."
Helbing has come to this view by an unusual path. Though he trainedas a 
physicist, he became fascinated in the early 1990s by parallelsbetween 
physics and human movements. "I was inspired by thesimilarity between 
fluid flows and how people walk aroundobstacles," he recalls. For nearly 
two decades, he and colleagueshave been studying the mathematics of 
collective human motion, whichexplains why Helbing now holds a chair in 
sociology.
Social scientists usually focus on the variability of humanbehaviour, 
which is hard to predict. But Helbing argues that in manycases it isn't 
very important. That's because circumstances oftenconstrain peoples' 
options so much that humans respond almostautomatically to external 
forces, making their average behaviourpredictable. On the roads, for 
instance, people generally driveclose to or just over the speed limit, 
similar to the wayself-propelled particles repel one another when they get 
too close.
Although the behaviour of individuals is often simple, thecollective 
patterns to which it leads can be counter-intuitive,making common sense a 
faulty guide to what might happen. Forexample, it is generally true that 
traffic jams become more likelyas traffic density increases. It's not 
always the case, though, asHelbing's group has shown.
Consider a two-lane road carrying both cars and trucks, where thecars are 
moving faster on average. At low traffic densities, thecars have plenty of 
space to overtake and can easily pass thetrucks. As the traffic density 
increases, drivers find it moredifficult to overtake because other 
vehicles are in the way.However, evidence from simulations and real 
traffic flows shows thatat a critical density of traffic, the obstruction 
to lane-changingbegins to have a beneficial effect. Because drivers tend 
to stay inone lane, they disturb the flow of traffic less, leading to a 
highertotal throughput of vehicles.
Similar counter-intuitive results show up in crowds of people. 
Insimulations and experiments, Helbing's team has confirmed what theycall 
the "slower-is-faster" effect. When people try to escape from aroom 
through a doorway, more get out if everyone stops rushing asthis prevents 
obstructions. Surprisingly, it turns out that placingan obstacle in front 
of the door can actually enable people to getout faster, as it helps to 
regulate the flow of people and maintainits fluidity. "A suitable obstacle 
can improve the outflow by about30 to 40 per cent," says Helbing.
What makes it work is that crowds adjust to local conditions. Whentwo 
streams of people meet at either end of a narrow passage, youmight expect 
a jam to form as only a chaotic trickle of people passthrough. But in real 
life, people often do something completelydifferent: they organise so that 
a group goes through first in onedirection and then the other, as long as 
the density isn't too high.The crowd organises itself spontaneously to a 
better outcome.
Helbing has found that you can model crowds using ideas akin tothose from 
physics. As a queue grows on one side of the passage, itproduces something 
resembling the pressure of a fluid or a gas. Ahigh density of people 
pressing together ultimately acts to drivepeople through the opening, 
thereby relieving the pressure.
Further work has convinced him that systems involving pedestrians,traffic 
and products flowing through factories often work insurprisingly similar 
ways, hence lessons learned about one may alsoapply to another.
Last year, Helbing and Stefan Lämmer at the Technical University ofDresden 
in Germany began wondering if traffic lights could also beengineered to 
cut congestion. According to a report by David Shrankand Tim Lomax at the 
Texas A & M University in College Station,congestion in the US alone costs 
an estimated $78.2 billion, wastes4.2 billion hours in delays and 10.9 
billion litres of fuel. So thepotential impact of efficient traffic flow 
could be huge.
This would mean giving traffic lights a way to adapt theirbehaviour, which 
most of today's systems lack. At the moment,engineers force traffic into 
patterns that appear favourable. Lightson main roads stay green longer 
during peak hours, for example. Butit's the engineers who do this tuning 
based on average conditionsobserved in the past; most traffic lights don't 
have the flexibilityto respond to changing conditions on their own. 
Engineers also takesome things for granted, such as the notion that lights 
must bemanaged from a central control.
Lights can do a better job, Helbing and Lämmer have found, if theyare 
given some simple operating rules and left to organise their ownsolution. 
To demonstrate this, they developed a mathematical modelthat assumed 
traffic flowed like a fluid, a well-established trafficengineering 
technique. The model also describes what happens at roadintersections, 
where traffic entering from one road has to leave byanother, much like 
fluid moving through a network of pipes.
Of course, jams can arise if traffic entering a road overloads 
itscapacity. To avoid this, Helbing and Lämmer make the lights at 
eachintersection respond to growing traffic pressure, like the peoplegoing 
through the passage. Each set of lights carries sensors thatfeed 
information about the traffic conditions at a given moment intoa computer, 
which then calculates the flow of vehicles expected inthe near future. The 
computer also works out how long the lightsshould stay green in order to 
clear the road and relieve thepressure. In this way, each set of lights 
can estimate for itselfhow best to adapt to the conditions expected at the 
next moment.

Best left alone

This isn't enough, however, because the lights might adapt too much.If 
they are only adapting to conditions locally, they might causetrouble 
further away. To avoid this, Helbing and Lämmer have deviseda scheme 
whereby neighbouring lights share their information so thatwhat happens 
around one traffic light can affect how others respond.By doing so, the 
self-organised lights prevent long jams fromforming.
Despite the simplicity of these rules, they seem to work remarkablywell. 
Helbing and Lämmer have demonstrated in simulations thatlights operating 
this way should achieve a significant reduction inoverall travel times and 
keep no one waiting at a light too long(See diagram). Nonetheless, the 
behaviour of the lights doesn'tgenerally fit with human notions of what 
ought to be efficient. "Howlong lights stay green is unpredictable," says 
Lämmer. Yet theaverage journey times go down and become more predictable.
What's more, the scheme eliminates other irritating problems thatafflict 
traditional traffic control. At quiet times, driverstypically have to wait 
far longer than is really necessary atintersections because the lights' 
schedules are designed to serve alarge number of vehicles. And in the 
middle of the night, lightskeep stopping cars even when there is no need. 
The self-organisingtraffic scheme eliminates these problems because the 
lights remainresponsive to local demands, for instance sensing an 
approaching carand changing to green to let it through.
Town planners are beginning to look at self-organising lights as 
apractical solution to looming traffic congestion. Helbing and Lämmerare 
working with a local traffic agency in Dresden, Germany, firstto test and 
then hopefully to implement the idea. In earlysimulations based on 
Dresden's road layout, they have hadencouraging results. "We've found 
significant reductions in waitingtimes and fuel consumption, and we can 
also accelerate publictransport," says Lämmer. Authorities in Zurich, 
Switzerland, havealso been taken by the idea.
Yet Helbing and Lämmer suggest their scheme only begins toillustrate the 
potential for self-organised traffic flow. Thetechnology to make cars also 
sense and respond to local conditionsalready exists, and many of us may 
soon cede at least some controlof our car to on-board guidance systems. If 
cars can talk to oneanother, Helbing and his colleagues have shown, they 
could improvetraffic conditions even more - greatly reducing the severity 
of jamsat times and possibly even eliminating them altogether (see 
"CruiseControl").
The wider lesson is that we just can't trust our intuition when itcomes to 
the super-complex systems that we depend on today. We maynever learn 
exactly how to control these systems in the traditionalfashion and the 
best way to cope may be by learning new principlesfor letting them manage 
themselves. Engineering isn't just aboutsolving problems any more, but 
building systems that can solve theirown problems. Being in control, it 
seems, may increasingly demandbeing a little out of control.

Cruise Control

Some of today's cars already contain technology that lets a driverhand 
over some control to on-board devices. Unlike conventionalcruise control, 
which simply maintains a driver's chosen speed,"adaptive" cruise control 
(ACC) uses radar to sense the distance andspeed of the car in front. By 
updating that information severaltimes a second, the system maintains the 
car's speed and separationdistance and automatically brakes if the car in 
front slows down, oraccelerates when the leading vehicle does. Also, it 
responds fasterand more accurately than human reflexes.
In recent simulations, engineer Arne Kesting at the TechnicalUniversity of 
Dresden in Germany, working with Dirk Helbing at theSwiss Federal 
Institute for Technology in Zurich and others, havestudied how the 
technology might help traffic respond to emergingproblems. While it may be 
some time before most automobiles on thehighway are fitted with adaptive 
cruise control, the researchershave shown that even a small fraction of 
users could make a hugedifference.
At the moment, drivers can respond only to conditions they encounteror 
perhaps hear about on radio reports. ACC cars could easily befitted with 
sensors able to receive signals conveying local trafficconditions from 
roadside monitors or other cars. Kesting and hiscolleagues suggest that 
these signals could reduce congestion bymaking cars equipped with ACC 
drive more intelligently.
For instance, cars flowing out of a traffic jam could automaticallydrive 
closer together in order to clear the jam faster. Meanwhile,cars 
approaching the jam would slow more gradually, rather thanbrake abruptly 
when reaching it. This would maintain greaterfluidity in the traffic, 
improving road capacity and stability oftraffic flow. Kesting's 
simulations suggest that if 25 per cent ofthe cars were using ACC, the 
scheme could eliminate many trafficjams. Even if only 3 per cent of the 
cars were equipped, traveltimes could be significantly reduced.
Kesting and Helbing are currently testing these ideas withVolkswagen and 
hope to see the scheme on real roads in a few years.

Related Articles
* 'Crowd quakes' could be predicted by CCTV analysis
* http://www.newscientist.com/article.ns?id=dn12607
* 7 September 2007
* Fast track
* http://www.newscientist.com/article.ns?id=mg18224413.800
* 3 April 2004
* Could smart traffic lights stop motorists fuming?
* http://www.newscientist.com/article.ns?id=dn13306
* 12 February 2008
* It's flaming freezing
* http://www.newscientist.com/article.ns?id=mg16722474.200
* 15 July 2000

Weblinks

* Stefan Lämmer's homepage, Technical University of Dresden
* http://tu-dresden.de/Members/stefan.laemmer/
* Dirk Helbing's homepage, Swiss Federal Institute of Technology
* http://www.soms.ethz.ch/
* Mark Buchanan's homepage
* http://pagesperso-orange.fr/mark.buchanan/indexMB.html
* Crowd turbulence: the physics of crowd disasters by Dirk    Helbing, 
Anders Johansson, Habib Al-Abideen
* http://arxiv.org/abs/0708.3339