Water has been thought to affect the dynamical processes in the Earth's interior to a great extent. In particular, experimental deformation results suggest that even only a few tens of parts per million of water by weight enhances the creep rates in olivine by orders of magnitude. However, those deformation studies have limitations, such as considering only a limited range of water concentrations and very high stresses, which might affect the results. Rock deformation can also be understood as an effect of silicon self-diffusion, because the creep rates of minerals at temperatures as high as those in the Earth's interior are limited by self-diffusion of the slowest species. Here we experimentally determine the silicon self-diffusion coefficient DSi in forsterite at 8 GPa and 1,600 K to 1,800 K as a function of water content CH2O from less than 1 to about 800 parts per million of water by weight, yielding the relationship, DSi ≈ (CH2O)(1/3). This exponent is strikingly lower than that obtained by deformation experiments (1.2; ref. 7). The high nominal creep rates in the deformation studies under wet conditions may be caused by excess grain boundary water. We conclude that the effect of water on upper-mantle rheology is very small. Hence, the smooth motion of the Earth's tectonic plates cannot be caused by mineral hydration in the asthenosphere. Also, water cannot cause the viscosity minimum zone in the upper mantle. And finally, the dominant mechanism responsible for hotspot immobility cannot be water content differences between their source and surrounding regions.