The Dutch radiobroadcasting VPRO has a wonderful science program called ‘noorderlicht‘ (site only available in Dutch, photograph taken from the site). Today one of the guests was Chris van Hinsbergen (sorry for those lacking knowledge of the Dutch language, this interesting curriculum from Chris is available in Dutch only). A young scientist who is working on a smart system to guide us through traffic jams. The current Dutch road has signs with time forecasts (23 minutes driving towards The Hague). These are notorious for wrong time prediction. Chris explained why: the methods used for estimating driving time are extremely simple. Every 500 meter the speed of cars driving by is measured. The system assumes stability in speed and expects this speed not to change. Chris is using a more complex way to estimate the driving time. A neural network will be looking for patterns in all of the speed data, that takes into account the time the speed is measured. No assumptions on traffic jam causes, just a learning system. No assumptions on stability. The neural network will ‘learn’ about patterns that forecast a change in velocity. I see a really interesting comparison to market research and marketing. I mean: how often do we assume stability? How many of our models are based on stability?