I've long been an advocate for more emphasis on the computer rankings, you can clearly tell from the computer rankings what kind of bias there is from human voters. Computers have no bias, one of the worst instances is how losing in the final weeks hurts much more than it does if you lose early.
If Oklahoma State loses in the first week of Big 12 play to Iowa State, they drop down to 6-7 spot like they did. But, they win out the rest of their games while others ahead of them lose later, including Alabama; where does Oklahoma State end up? Likely a #2 spot, as if Oklahoma State is #3 when Alabama loses then voters say Oklahoma State is the better team because the Alabama loss is fresher in their minds. Voters have a bias on when you lose, among other biases.
Computer rankings don't care when you lose, they are always looking at your entire resume, the entire season and that is what the National Championship is supposed too be; the two best teams from the entire season.
Do you still need a human input? Probably, trusting a complete computer formula will probably have a few years where it gets you into trouble due to circumstances the computers can't see (human element). So you keep a portion human element, but 2/3rds is just too much. Proposed solution:
1/3rd Computer Rankings -- same computer formula's as current system
1/3rd Computer Rankings -- computer algorithms that take into account margin of victory & other details that current ones do not
1/3rd Human Voters -- not sure if Coaches or Harris, or perhaps an average between the Harris, Coaches & AP would be best suited here to let everything average out a bit more (more voters less likely a coo that hurts a team more than it should)