Hacker News Reader

Comment by bdhe | Hacker News Reader

People study the Price of Anarchy [1] to design algorithms to minimize the deleterious effects of decentralizing decision making to rational and selfish agents. Traffic flow is a textbook example of selfish behaviour leading to highly inefficient solutions as compared to a central authority that decides on each agent's behalf.

It is amazing to see the time come that technology can actually allow us to centrally decide the action of agents (or at least suggest useful strategies) that are provably more efficient than any selfish strategy (which is predicated on the assumption that a large fraction, if not all agents, will go with the suggested strategy, and not act maliciously).

[1] http://en.wikipedia.org/wiki/Price_of_anarchy

View on HN

From the look of the article the google maps solution is using lagged traffic flow data as a basis of directing waves of traffic. I think in practice, if this had a large uptake rate, by directing cars to areas with low traffic rates, you would inadvertantly cause traffic jams even though you send "waves" to other areas in a subsequent time period.

I like the idea of a traffic utopia where cars are directed around in the most efficient way, However i think that this can only realistically work with open information sharing, (not competition) between the major players i.e. Garmin, tomtom etc to increase take up rates.

> From the look of the article the google maps solution is using lagged traffic flow data as a basis of directing waves of traffic.

I wonder if they have plans to use the destination address as another basis as this gets more traction. As the number of users grow this can help estimate where traffic will be after time passes. Google can then become that central authority of ultimately routing cars in selfless directions for maximum efficiency. It would alleviate traffic simply by showing red on the other routes, discouraging them. This red is then the predicted heavy traffic. Or rather: follow the red path and you will cause traffic.

As the number of users grow this can help estimate where traffic will be after time passes.

Fascinating, thanks for this insight. It reminds me somewhat of how Google can predict flu outbreaks better than the CDC based on users' queries.

This is a wonderful thought... but at the same time all I can think is "Gee I hope google holds true to their motto"

Most interesting. This concept is new to me; any idea if/how it has been applied to market behaviors? It would seem to be fundamentally at odds with the efficient markets hypothesis.

The efficient market hypothesis has at its basis the assumption that relevant information is readily available and disseminated to all parties.

Up until recently, information on traffic has not been readily available, nor has it been disseminated to drivers in real-time.

People can only make decisions based upon the data that they have.

I can imagine that centralized planning could be valuable in a situation like this, where it's not possible for individuals to react in real-time to changing circumstances; though I'm sure that will be possible in the not-too-distant future.

There are situations when even full information leads to failure. Tragedy of the commons (overgrazing) is the most known example.

Another less known one is as follows: assume two routes from A to B, where one is shorter (say, 1 hour when not congested) but prone to congestion and the other one longer - say 2 hours regardless of traffic.

The selfish decision is to take the shorter route as long as its congestion hasn't slowed it down to the level of the longer route. So the "anarchy" equilibrium leads to everybody spending 2 hours.

A benevolent dictator (or enlightened citizens able to come up with a social contract) could of course come with a better solution - every day choose a limited number(#) of people who can choose the shorter road without congesting it.

(#) The optimal number obviously depends on the exact characteristic of the dependency between the number of people choosing the road and the latency generated. This just a model after all.

Reply on news.ycombinator.com