User talk:Groovier

About judging 'human-ness' w/o gathering or recording data

If I'm not mistaken, this is one of the desired properties of your research project.

“ If the method performs well, the final software will be a service that judges user registrations in real time, without recording any data. ”

however isn't it plainly obvious that if you do not record any data then one trivial way to bypass this test is to generate a set of event with a real user (and pass the test), record and reuse this series of events endlessly? This is a problem for which I cannot figure out a way to solve without involving at least some levels of data gathering. What is your comment on this? Thanks in advance. 209.6.76.173 02:37, 13 February 2018 (UTC)

@209.6.76.173:: You are correct, I didn't really think things through when I wrote that. We are not really sure what fraction of the spambots were specifically written to interact with Wikipedia and how many are general-purpose ones - could well be most, but of course even then it would not be wise to rely on that in the long term. We would need some way to gather data about spambots at least - one approach could be to move data gathering to the edit page, where a significant fraction of the spambots are caught by the edit filters on submit so there would be no need to retain (non-spammer) data. (Also, seems a lot harder to use replayed human data for editing the page.) --Tgr (WMF) (talk) 08:12, 16 February 2018 (UTC)