Is that computer polls are “forward looking” (ie. predicting point spreads between teams in future games) and human polls are “backward looking”.
Computer polls have discovered that they are their most accurate when they use some weighting on the previous year’s results.
Then the question you have to sort through is whether they are able to accurately rate teams who make a huge improvement/decline year over year or if humans simply overreact to the results from a narrow span of time (ie. smaller sample size).