Neural net input layer representations

How should I represent my neural net input, if each input node has 3 possible values?

For example, a position on a Go board may be empty, have a white stone, or have a black stone.

Should I use empty = 0, white = -1, black = 1? Or should I use 2 nodes for each input position? Something like empty = (1,0), white = (-1, -1), black = (-1, 1)?