A very interesting article. Although for me it didn't really explain the need for the bias in the exponent field of floating point representations.
Excellent question. The short answer is that the bias allows you to get both positive and negative exponent values. So, if you want to get 2^-3, the bias allows you to encode it in a very straightforward way, by using the 8-bits for 124.
The longer answer is that it has to do with 2's complements, which isn't discussed here at all. It turns out that using a bias just makes life easier for the hardware, then having to interpret a 2's complement number and figure out if it is positive or negative.