Looking at spectra taken of the same sample at different scans and it is easy to see the difference. The 64 scan spectrum is much smoother.
There is an expected value for all occurrences. This happens when an infinite number of trials are run and then averaged together. The expected sum for rolling dice is seven, even though a 2-12 can be rolled.
The same thing is true of IR spectra. IR spectrometers will scan the sample through its range of light and average the absorbance values at each wavenumber scanned by the number of scans to produce a spectrum that has a value closer to the expected value than just one scan.
Sometimes less is more. If the functional group that you are looking for is a very strong defined peak in a bare region, 4 scans is enough because the peak will stand out no matter how much baseline noise there is.
However, if the functional group has a weak signal, more scans should be used so that the peak can be seen.
[animated gif of a molecule going from low number of scans to high number of scans] or [separate picture pages of individual scans (1,4,64,etc.) showing the smoothing effect]