At the end of the paper they describe how the come up with ELO rankings, and that in order to avoid some bias due to self-play only they include the results of

At the end of the paper they describe how the come up with ELO rankings, and that in order to avoid some bias due to self-play only they include the results of the AlphaGo's versus Fan, Sedol, etc.