The Uncanny Valley hypothesis was first noted in 1970 by Masahiro Mori, professor of engineering at the Tokyo Institute of Technology, in his book entitled Bukimi No Tani (translation).

It refers to the idea that when people view human-like artificial characters which are not quite 'perfect', it can generate strong feelings of unease.

The 'valley' refers to the a dip in the graph of a human observer's affinity for replicas. In other words, if the replica is very obviously not-human, then people don't generally find it unpleasant - and if it was possible to create a completely 'perfect' replica, they also wouldn't find it unpleasant (they may not even sense it was a replica).

At least eight different theories have been put forward as a way of explaining the phenomenon. For the current list, see Wikipedia

To date however, there's no agreement about which, if any, may be the correct explanation(s).

In addition, some researchers question if the effect is even real.

The uncanny valley hypothesis, proposed already in the 1970s, suggests that almost but not fully humanlike artificial characters will trigger a profound sense of unease. This hypothesis has become widely acknowledged both in the popular media and scientific research. Surprisingly, empirical evidence for the hypothesis has remained inconsistent.

Source : Frontiers in Psychology, 6: 390.

Further reading : A 2012 interview with Masahiro Mori, IEEE Spectrum.