I don't think every inventor makes a good researcher, so having those check for prior art might not be effective. However, patent submitters (or even the general public) could be used to check the state of the art: i.e. when a patent is submitted, people in the field should be given the decription of what the device does, and if they come up with the same solution as the patent submitter, it's a "state of the art" submission that is not patentable.

Especially in the field of software, where coming up with an algorithm typically doesn't require much of an investment, a simple mailing list that patent descriptions are posted to (the description would contain what the patent is supposed to accomplish, not how it does that, e.g. "an algorithm to sort a set of numbers") would suffice: grant those patents that noone replies to, reject the patents where someone else posts the same way to solve the same problem.

-mendel