It is because there are two dimensions to the problem. Noise amplitude is only part of the picture. Frequency is the other part. If the noise covers a narrow band it can be eliminated with a filter. If it is wideband, then it can not be easily eliminated with a filter. Think of it as averaging the amplitude squared over a bandwidth to give Volts per root Hertz. The reason for using amplitude squared can be answered by thinking about the average value of a sine wave over a single period. Yup -- the answer is zero -- not a useful average when talking about AC signals.