We see this same type of thinking when dealing with all sorts of other complex, multi-dimensional systems. Economics, for instance. There are a huge number of people who think economies work in this same linear fashion. Or managing large groups of developers.
In some systems that are becoming more complex, such as avionics systems for large airplanes, it's getting to the point where the old root cause analysis methodology is still being used although it's getting less and less applicable (I'm speaking specifically of the crash over the ocean of the Air France flight, but there are other examples)
Our minds desperately want to live in a world with clear causality. Do X and Y will happen. When the world doesn't live up to our expectations, many times we just get out a bigger hammer.
Looking at this problem solely from a philosophical standpoint, it looks like there is a powerful argument in place for tiered systems to have some sort of distributed goal-seeking self-programming (machine learning), especially when dealing with large numbers of identically programmed/configured computers. That way the same combination of obscure causes wouldn't have such a disastrous multiplicative effect. Would be cool to chase that down further sometime.
The real world still has clear causality. Just because the links of causes and effects (and the involved feedback processes) are often too complex for most people to follow does not mean that they are not there. You might as well say that because most people can't do calculus, calculus isn't really useful.
Yeah this was not a keeper for me. "Causality is complicated..." yeah and if a butterfly flaps it's wings... blah... blah... blah...
This offers no insight as to how to improve process when failure occurs. Sure dogmatic application of root cause analysis is foolish, but the same obvious conclusion could be reached about dogmatically applying any type of management principle or analytic process. Failure to think outside of the box is a failure too.
What is annoying is that the author suggests that people should be skeptical of root cause analysis and 5 whys, without offering anything concrete as an alternative approach.
Every management technique is simply a practice towards further learning, but failing to practice anything simply because you can find flaws in everything accomplishes exactly nothing.
It's not that causality doesn't exist, or you shouldn't use the 5-whys, it's that we have a desire to focus on single causes instead of multiple ones, and understand systems and simple cases of cause-and-effect.
Systems are coded and tested for common paths. Extremely rare circumstances take systems down pathways engineers may have not planned for. When you layer systems, combinations of rare situations can cause "storms" that take everything down. There was a great article on AWS on HN a few months ago that made a point that bears repeating: they are doing something with PaaS at that scale that is quantitatively different than what's been done before.
But nobody is saying to give up causal analysis, only to recognize the limits of RCA when your causation tree can span into hundreds of nodes. The interesting thing here is that the human application of the analysis tool has it's own features that's just becoming apparent. Analysis tools have human environments they live in. Once you acknowledge that a a separate issue, there's no reason you can't continue to use those tools (and others) to work through the problems.
We see this same type of thinking when dealing with all sorts of other complex, multi-dimensional systems. Economics, for instance. There are a huge number of people who think economies work in this same linear fashion. Or managing large groups of developers.
In some systems that are becoming more complex, such as avionics systems for large airplanes, it's getting to the point where the old root cause analysis methodology is still being used although it's getting less and less applicable (I'm speaking specifically of the crash over the ocean of the Air France flight, but there are other examples)
Our minds desperately want to live in a world with clear causality. Do X and Y will happen. When the world doesn't live up to our expectations, many times we just get out a bigger hammer.
Looking at this problem solely from a philosophical standpoint, it looks like there is a powerful argument in place for tiered systems to have some sort of distributed goal-seeking self-programming (machine learning), especially when dealing with large numbers of identically programmed/configured computers. That way the same combination of obscure causes wouldn't have such a disastrous multiplicative effect. Would be cool to chase that down further sometime.