Rollin', Rollin', Rollin'

Over at Twisted Sifter, we find a link to a article about a scientist who was studying the riding of bicycles. In an experiment related to his paper, he turned a bicycle loose without a rider 800 times and mapped the different routes it followed before falling over.

The interesting thing is that the number of repeititions led to a largely symmetrical collection of paths. Although it's unlikely the bike followed the exact same path more than once, it begins each time on a similar straight line before beginning to wobble. Bunches of the wobbles start out the same before diverging, and only in a very few does the bike travel the longest distance before it falls over.

The map shows a couple of things -- one, we can see how an algorithm works. The equation is tweaked with at one variable or another and produces a different result, and the bike is tweaked by hitting an imperfection in the road, feeling a gust of air, not receiving a push at its precise balance point, and losing energy the longer it travels. 

We also see how an algorithm doesn't always work and why sometimes, Amazon suggests really odd books for you given your previous taste. The "tweaks" in both the algorithm and the paths of the bike are random and can't be predicted with much precision. Until more data gets entered -- in the form of more tweaks in either the algorithm or the bike path, there's not very much that can be predicted about where the bike will end up on this attempt. When Amazon's algorithm acts before it has enough data, it might wind up recommending you read a book about and printed in ancient Sanskrit. You'll tell it no and it will make that another tweak in its path.

And by the way: 800 times! I hope that was done over a couple of days or so.