Think of it this way, simple black and white chessboard (8x8), 32 black squares, 32 white. 'Compress' it by removing every other square/pixel. If perfect, you get a mini 4x4 chessboard. If not, you end up with all black, all white or stripes.
Now 'uncompress' it and interpolate to original size. Simple interpolation of the perfect mini board gives you 16 black, 16 white and 32 gray pixels/squares. Do that hundreds of times across a single frame and hundreds more through time. All sorts of artifacts are created. Most are too small to notice by the human visual system. However, they will show up under close examination.
'Adding' pixels will always distort the original image as the data must be made up. Multiple compress/decompress cycles simply make the problem worse.