Because, to the popular perception, computers are infallible constructs and humans are fallible constructs. Imagine a human being counting thousands of paper ballots, what's the likelihood that the count will be off by one or two?

It's also a case of different concerns cropping off: with paper ballots, there's a risk of miscounting, or disagreement on what constitutes a mark (remember the 2000 election, where the outcome of the election depends on the standards of how much of a mark you need for it to be indicated as a recordable vote). With computer ballots, the concern is over people hacking the tabulating machines or other computerized machinery. You're comparing vastly different risk profiles, and I suspect that most people don't have sufficiently-grounded knowledge to adequately compare them.