Improving Wetware

Because technology is never the issue

Lecturing Birds on Flying

Posted by Pete McBreen 03 Sep 2009 at 19:40

Lecturing Birds on Flying is another take on Nassim Taleb’s Black Swan idea, and Pablo Triana does a good job of explaining that a lot of what was supposedly solid theory around financial modeling, is actually not applicable in the real world.

It turns out that most models make the assumption that the probabilities that apply in markets are normally distributed (where anything beyond a three sigma event should be exceedingly rare), is incorrect and that the probability distribution has very “fat tails”, where a twenty-five sigma event is not unheard of. This work complements Nassim’s black swan book by restating the ideas in a slightly different way and by being written after another event that demonstrates that the markets do not have normally distributed probabilities.

The applicability to software craftsmanship stands out for me in the way that the models try to ignore the effect that individuals have on the market. The assumption is made that the actions of individuals are all independent, when it is blindingly obvious that this is not the case. Yes, the assumptions might work if the market was made up of lots of small players, but that is not the case. The markets only have a few players of any significance, the trades of any of these make the markets move, the small players make the random day to day noise, but the big players are what cause the problems - the too big to fail kind of problems.

Software Engineering is too big to fail

To get a good software disaster, you need the big ideas from software engineering that can influence lots of people to make similar mistakes on a project. The average level of software development is abysmal, most projects only succeed to the extent that they have one or two talented or experienced developers who manage to overcome the overall lure of failure.

Software projects should be built around the outliers that happen to be talented or experienced enough to succeed, but even then the sponsors should only bet as much as they can afford to lose. Nobody knows how to make massive projects successful - other than throwing money at it until it is politically acceptable to call the result successful.

The issues in software are different than in finance, but in both cases the underlying model of how the field works is incorrect, and the consequences of this mismatch between model and reality are what causes things to crash and burn.