3 ms·
I think a pair of issues are being improperly conflated here with respect to the complexity of human systems. We see a similar problem contained within software
by jandrewrogers 9y ago
I think a pair of issues are being improperly conflated here with respect to the complexity of human systems. We see a similar problem contained within software engineering (and chemical engineering): most humans, including most engineers, are terrible at analysis and engineering of complex distributed systems. Human systems are just another example of this problem domain.
Systems level engineering breaks a problem-solving heuristic that people rely on to make analysis simple: that changing a variable does not feedback into the value of that variable, which allows you to break a big problem into a bunch of trivial small problems that can be solved independently. For systems level engineering problems, you have to consider the interaction of all components simultaneously, which is a very different type of analysis (you are designing for locally stable and hopefully approximately optimal systems level equilibria). It also has characteristics that go against the way humans want to design things: the general design pattern is that you can't design for "good" outcomes per se, you can only design systems that mitigate "bad" outcomes, with optimal "good" outcomes being the natural result of efficient pathology minimization. Most people tend to design for the outcome they want, rather than minimizing the outcomes they don't want, but only the latter works.
Bringing this back to human systems, in a naive analysis you can take any variable, label it "good", and dial it up to eleven with some direct trivial changes. I almost always see it approached this way at a policy level. In a systems level analysis, it never works that way and you have less control of the tradeoffs than you might like as a matter of policy. We suck at human systems because, thus far, we haven't come to terms with the reality that a systems level analysis doesn't let you dictate the absolute values of "good" and "bad" no matter how you define them and controlling the relative values is far from trivial in a systems level analysis. We have tools for approaching this, as some types of engineering require it, but we don't apply it because it doesn't lend itself to the kind of naive analysis people understand and the tools don't allow the kinds of outcomes in human systems we wish we could engineer.
- rumcajz 9y agoI like the way you think. I've wrote a little satire along those lines once: http://250bpm.com/blog:88 http://250bpm.com/blog:88
- paraschopra 9y ago>Human systems are just another example of this problem domain. It isn't. Of course, both are systems of some sort, but there are several differences: 1. You can simulate a systems engineering problem and if it doesn't give you an acceptable solution in the first go, change some parameters and re-run again. In human systems, you (usually) do not get such luxury. 2. Components in an engineered system behave predictably. Humans are free agents who often act in irrational, surprising ways. You cannot simulate and analyze that. This is why there are black swans. We cannot bundle these two systems as one and the same thing.
- icebraining 9y agoYour view of engineering seems too narrow. How would you classify a traffic engineering problem? Humans area a crucial component, and it certainly doesn't have an objectively correct solution (except possibly in narrow cases).
- pixl97 9y agoCurrent traffic engineering is almost a complete human problem. Swap it out to driverless cars and it becomes more like routing IP packets.
- greenyoda 9y agoNot so simple, since driverless cars will still have to interact with pedestrians, bicyclists, etc. Not to mention that humans will still ultimately determine where the cars go, and when - even if all cars are driverless, massive numbers of people will hit the road the day before Thanksgiving.
- pixl97 9y ago>Not to mention that humans will still ultimately determine where the cars go Incentives. Right now the road network has no information on where you are going. Because of this it is near impossible to route or plan traffic beforehand other than using historical estimates. It is everyone for themselves. In a completely driverless world people could submit their driving route the day before a major event like Thanksgiving so the network has more information on what traffic will look like then. The people that plan ahead could then be given a 'queue credit' or 'queue boost' that routes their car in to the faster moving lanes, where those that did not get stuck in the slower lanes for longer.