During the last few days, I get at least one computer-generated vandalism review (a.k.a. honeypot) per day, sometimes two.
I wouldn't object if this happened maybe once per week, but seeing how you can review at most 20 edits per day, getting 1-2 surveillance probes (i.e. up to 10%) within these seems a bit hilarious.
If review robots are really such a big problem, one needs to find a different solution. If community reviews really don't work, then maybe one should not have them in the first place (though I can hardly imagine that so many people run review bots, to what avail?).
In any case, the solution cannot be to harrass people who are actually trying to do review properly. I don't recall ever failing one of these, but still the system keeps bugging me. What for?
Doing a proper review involves not only skimming over the text and hitting "Reject!" on the first occasion. Or, not reading at all, and hitting "Accept!".
A proper review involves reading and trying to figure out what someone was trying to say, even if it looks wrong or is worded badly. Even it it comes over as gibberish -- not everyone is a native English speaker. Possibly an edit can be improved rather than rejected.
Reading the honeypot gibberish (which, to make it worse, sometimes looks almost like something a real person -- presumably a bit stoned -- could have written) takes time. It may come as a surprise, but time is valuable, even time that comes from a community "for free".
As it is now, I am getting more and more inclined to simply press "Reject!" on any edit that I see which doesn't look 100% right at the first glance. Because hey, it's probably another darn honeypot probe anyway. I'm probably not the first to feel like this, there are most likely people who already do just that.
Read as: Excessive honeypotting may drastically reduce the review quality.
At the very least, probe frequency could be reduced for people who have not failed any (or very few) probes in the past.
If someone has consistently passed a number of tests, either the tests cannot distinguish robots from humans, or the subject is really a human. In either case, there is no valid reason to persist doing them.
Since robo reviews are perceived as such a big problem, one might consider not giving out review rights automatically just because some reputation threshold is reached.
Instead, users could gain a "tentative review" privilegue. Meaning they can cast their approve/reject votes, but they do not necessarily approve or reject anything. That would mean more reviewers per edit are needed, of course (to to the real approves). As you have "correctly" approved so and so many edits (see Jack's "conformist tracking" suggestion below), you gain the real approve privilegue.