Well, if the komi is set to 6.5 one will choose to play black, if its set to 7.5 one will choose white. I'm not sure if that even gets rid of anything unfair (not like it really is unfair anyway).
The idea of using the swap rule instead of komi pie was more something I was curious about than something I find necessary. It deforms the game more I guess, since the first move is now not about finding the strongest move anymore, but one that is inbetween in strength, which might then snowball to the following moves and create an entirely new opening theory.
I also really enjoy the swap rule for games like Gomoku or Polyomino, where you have a nice number of unorthodox openings available. I think swap would work well in Go, but for most players (let's say 4 dan and less) the 1-point difference in komi is almost irrelevant with regards to the end result.
The small sample of AlphaGo vs. AlphaGo games published showed white winning a disproportionate amount of the time. Which led some to speculate that komi was too high.
With access to a larger dataset, have you been able to make any interesting conclusions about the basic Go ruleset? (ie: Black or white have an intrinsic advantage, komi should be higher or lower, etc.)
There is a video where Michael Redmond looks at a bunch of AG self-play games and says he thinks that the komi is right, and that White wins more games simply because AG is a stronger player as White than as Black. He gives some reasons for that, i.e. there are strategic differences in how to play White vs as Black, which AG apparently didn't figure out. Looks like AG0 has caught up though :).
I agree with Tasmanian Devil's2017-10-19 post: "Strictly speaking, the correct komi should be an integer [probably 7 based on JulianSchrittwieser of DeepMind and David J Bush's 2017-10-19 post] and correct (perfect) play should give a draw (result in a tie game)... I don't understand how anybody can argue with that other than "we want to make sure there are no ties because we want to buy only one 1st place trophy for the tournament." - but I think that problem could be solved another way... and I think a finished game with a tie score would be a beautiful thing. Although I can also imagine that perfect play with a whole number komi might not actually result in a tie game (even if Tasmanian Devil and I feel it "should") for a reason something like (to oversimplify) the reason the game of Hex (or tic-tac-toe-3-in-a-row) cannot end in a tie.
Ties are a problem because at high level, many games would end as ties as for example in chess.
Then the other problem is how you value a tie compared with a win?
Even in chess it is not easy and depends on the type of tournament and how the competition is doing, for deciding whether you should aim for a more aggressive (and risky) strategy.
I think there is a parameter in Stockfish for that but there was no agreement on how to set it because it should depend on the strength of the opponent.
The problem with Chess is that draws are all too common. In the candidates tournament 2016, Giri played 14 remis in a row! In games where draws are possible but not that common, like Go with integer komi or Othello, I don't see any big problems. If two players have played equally well, why should the colour they happened to play decide the winner? Ties for first place are possible even if draws are not possible: three players may have beaten each other cyclically. And here on Little Golem there was an example of an Amazons tournament with five players tied for first place!
The solution, as lazyplayer said, is to play Button Go with komi 7. The button is a special token worth 0.5 points. At any point, instead of making a regular move, a player can take the button, provided that none of the players has taken it before. (Officially, taking the button lifts ko and superko bans, but I'm not sure it should.)
It's interesting to note that Button Go with komi 7 is either virtually or fully equivalent to Go with komi 6.5 under territory scoring, depending on how you define territory scoring and whether you consider that taking the button lifts ko and superko bans or not.
Just to be clear: button Go is meant to be played with area scoring, which is what AlphaGo uses. Under territory scoring, komi 6 plus button always gives the same result as komi 6.5 without button, and fractional komi is simpler.