The software agent that consumers interact with would have to be responsible for determining the accuracy of ratings.

I suspect that the most important aspect of such a system would be trusted participants: eg, insurance companies, city licensing agencies, etc. If you get one hundred great, but fake, reviews from shell accounts, it would be rather meaningless if your city didn't issue you a license on the blockchain, or you claimed false insurance which was somehow disputed on the same blockchain by the legitimate insurer.