The idea of 3 IDs for RG was discussed at length in
https://forum.inaturalist.org/t/agreeing-with-experts-and-research-grade/3718/43 and
https://forum.inaturalist.org/t/issue-with-users-automatically-agreeing-to-an-identification/2987/36
For groups where there are lots of identifiers (such as birds) it wouldn’t be a problem. But in some other groups, getting two identifications is already hard enough, and requiring three may mean more observations failing to get to RG.
Personally I really wish there was some way of preventing RG from being reached by just one reliable ID, whether that happens via the “one ID plus subsequent agreement from the observer” scenario, or the “one ID plus agree-bot”. But as is obvious from those other two threads, there seems to be no simple way to achieve that without some downside.
Just as a thought for discussion, how about a trial period (6 months??) of requiring 3 IDs for RG, and then look at whether it really does result in a significant fall in the number of obs reaching RG. If it does, we could go back to 2 IDs and think again. Or perhaps go to 3 IDs for unaffected groups and 2 for the harder ones. The experimental approach.