My issue with creating a system of selecting teams solely by efficiency metrics is that it leaves the system open to manipulation by teams (you can always try to perfect it, but there will always be holes, open to exploitation)
Fair enough. I agree that there's a non-zero chance of an occasional outlier when using computers alone, but that's most likely to be due to absences by star players. A committee could be in place just to handle these kinds of things and otherwise stay out of the way.
I also don't see much room for intentional manipulation- the MasseyComposite isn't noticeably biased towards teams with harder/easier schedules. This year, it would have selected St John's and Indiana St instead of Virginia and Texas A&M, and just barely Pittsburgh instead of Northwestern. St John's played a harder schedule than Virginia, Indiana St played a weaker schedule than Texas A&M, and Pittsburgh played a weaker schedule than Northwestern (but similar non-con). It would have awarded a slightly better seed to Auburn, New Mexico, Colorado and Nevada, and a slightly worse seed to Clemson, Wash St, and S Carolina. None of these are outliers in strength of schedule in either direction. In each of these cases, betting odds are closer to the MasseyComposite than the selection committee's rank, or in between the two.
To get in, you really just have to (on average) do a bit better than a ~#40 team would against your schedule. To get a certain seed, you have to (on average) perform however that seed should against your schedule. There's going to be lots of variance from game to game, and the computers do a pretty good job at averaging all that out, while humans need to oversimplify.