here's the explanation, according to robert ritch, chief of the glaucoma service at the ny eye and ear infirmary:
"One account explained motion in the latter as arising from differential retinal processing times for bright and dim image regions (Faubert & Herbert, 1999). However, this account is not consistent with the observation that after fixation, apparent rotation slows to a stop over several seconds. Slowing to a stop implies a saturating adaptive process. We show that three known visual mechanisms can predict the illusion: (1) nonlinear retinal adaptation to luminance, in which the light response and dark response adapt most rapidly at the lightest and darkest parts of the image, respectively; followed (optionally) by (2) contrast normalization; and then by (3) a standard motion detector (e.g. Reichardt, 1961; Adelson & Bergen, 1985; Johnston, McOwan & Buxton, 1992) operating at an appropriate scale. This general framework predicts opposing motions from dark and light adaptation in the standard gradient illusion; this could underlie the finding by Fraser & Wilcox (1979 that different people see different directions of rotation. In the Rotating Snakes illusion, dark and light adaptation induce motion in the same direction, which makes the illusion more compelling. Inspired by "Rotating Snakes", we created a new gradient illusion, with alternating gradients from gray to white and gray to black, that is likewise more compelling than the standard gradient illusion. Corroboration for the role of luminance adaptation is that after a flashed exposure to the standard gradient illusion, a uniform white field appears to have a short-lived, fast illusory motion, whereas a black field has weak motion in the opposite direction."