54
$\begingroup$

This is in response to a request from David Roberts from this Meta question: Has MO been reduced to an AI training service, and if so what should we do about it?.

Should MathOverflow require users to register an account to post any content including questions, answers, and comments, as a possible way to combat the increasing amount of AI content that is posted here? (EDIT: Italics part added later for clarification.)

See also the previous question: Should MathOverflow require registration to ask a question? [2017]. But I think it is appropriate to ask again, because "the facts have changed," so to speak.

EDIT: I don't know how to ping a mod, but given the overwhelming response in favor of requiring registration, and given the fact that I continue to see tons of unregistered users who are very likely AI posting all kinds of weird stuff, is there any way we could implement this now?

$\endgroup$
14
  • 1
    $\begingroup$ Just to clarify - the suggestion is to require registration for both asking and answering - or only for asking questions, as in the previous post? $\endgroup$
    – Martin Sleziak
    Commented Sep 26 at 4:22
  • 1
    $\begingroup$ I will add a link to this post on Meta Stack Exchange: Why are unregistered users not allowed to post questions on some sites? The answer posted there lists the sites which require registration - some of them have this restriction only for asking questions, some of them for making any post. $\endgroup$
    – Martin Sleziak
    Commented Sep 26 at 5:55
  • 4
    $\begingroup$ @MartinSleziak I guess both to ask and answer. Just, get rid of unregistered accounts altogether as an option. $\endgroup$
    – Sam Hopkins
    Commented Sep 26 at 12:14
  • 2
    $\begingroup$ "require users to register" ... Someone who only reads, never responds ... that is stilll a user ?? $\endgroup$
    – Gerald Edgar
    Commented Sep 26 at 16:11
  • 8
    $\begingroup$ If I understand correctly, math.SE requires registration before asking questions. Does anyone have a good sense for whether or not they are getting a lot of poor quality AI-generated questions? $\endgroup$
    – Zach Teitler
    Commented Sep 27 at 3:19
  • 4
    $\begingroup$ @ZachTeitler Concerning Mathematics, I will just mention that they require registration for posting questions since October 2016 and for posting answers since December 2022. There was a discussion whether registration has made an impact - but it was back in 2017, long before the problems with AI: Registration to ask a question – has it made a difference? $\endgroup$
    – Martin Sleziak
    Commented Oct 2 at 5:19
  • 2
    $\begingroup$ "I don't know how to ping a mod...." Probably no need to do that – I would expect that at least some of the mods read meta regularly. $\endgroup$
    – Gerry Myerson
    Commented Oct 7 at 7:43
  • 1
    $\begingroup$ @MartinSleziak I don't know how to respond to you in chat, but I added a clarification to the post that the registration should be required for posting anything to MO. $\endgroup$
    – Sam Hopkins
    Commented Oct 7 at 12:30
  • 1
    $\begingroup$ @SamHopkins Thanks for editing. As already mentioned in the above comment, several sites opted for allowing only the registered users to post questions and some sites even to post answers. I am not sure whether a similar option exists for comments, but I found this feature request: Do not let unregistered users “comment everywhere” (at least on certain sites). (In case somebody is wondering what was said in chat, this was in the MathOverflow chatroom.) $\endgroup$
    – Martin Sleziak
    Commented Oct 7 at 13:26
  • 7
    $\begingroup$ We have noticed that. There is one small formality before we can pass this on to SE (technically speaking, this sort of decision needs to pass through the board first). I have raised the issue last night. $\endgroup$
    – Asaf Karagila Mod
    Commented Oct 7 at 20:44
  • 1
    $\begingroup$ @MartinSleziak, re, since "comment everywhere" requires 5 reputation, how do unregistered users have the reputation to do it? $\endgroup$
    – LSpice
    Commented Oct 9 at 1:08
  • 2
    $\begingroup$ @LSpice Clearly, at the moment they can get reputation from their posts - we'll see whether MO decides to change this. But even if a site requires registration from posts, I suppose reputation can be obtained from suggested edits and possibly from association bonus. (But I suppose that unregistered accounts with reputation obtained in this way are very rare.) $\endgroup$
    – Martin Sleziak
    Commented Oct 9 at 5:48
  • 2
    $\begingroup$ AI spam is also going to increase now that four Nobel Prizes were awarded for AI and machine learning, so this is a timely question. $\endgroup$
    – Hollis Williams
    Commented Oct 20 at 14:46
  • 1
    $\begingroup$ Somebody must have felt really angry about this question, because it has been attempted twice to close it (both attempts got invalidated by the reviewers). Anyway, it would be interesting to ask, in a few months, if this change has diminished the number of AI-composed posts. $\endgroup$
    – Alex M.
    Commented Nov 2 at 16:34

3 Answers 3

Reset to default
37
$\begingroup$

I can confirm that MathOverflow now requires a registered account in order to post on the website.

[Update: Originally SE had blocked questions from unregistered accounts, not answers. This allowed unregistered accounts to post "trivial" answers that turn into comments, or answers in general. This had been blocked as well now, and any type of posting on MathOverflow for now is requiring a registered account.]

$\endgroup$
12
  • 6
    $\begingroup$ This should go in the MO history meta post I think $\endgroup$
    – David Roberts Mod
    Commented Oct 12 at 11:27
  • 1
    $\begingroup$ I was just able to post a comment in main without logging in, from an anonymous browser window, so something still doesn't work as it should. If you write an answer without being registered, it gets converted to a comment. Sorry for "polluting" a thread on the main MO, but the same method does not work here in Meta. $\endgroup$
    – Federico Poloni
    Commented Oct 15 at 15:01
  • 1
    $\begingroup$ @FedericoPoloni: Your link leads to this page. Secondly, you should not be able to post any comments without registration, so I am very confused. Can you point out this comment? Can you leave me on this answer? $\endgroup$
    – Asaf Karagila Mod
    Commented Oct 15 at 16:05
  • 1
    $\begingroup$ @AsafKaragila Sorry for the wrong link, here is the correct one: mathoverflow.net/questions/480706/… . It's a comment on mathoverflow.net/questions/480706/… . I posted an answer to that question without registration, from an anonymous browser window, and then got a notice that the answer was automatically converted to a comment. $\endgroup$
    – Federico Poloni
    Commented Oct 15 at 17:11
  • 2
    $\begingroup$ @FedericoPoloni: Thanks. I have passed that on to the SE people. $\endgroup$
    – Asaf Karagila Mod
    Commented Oct 15 at 20:32
  • 2
    $\begingroup$ Help center also says that people can comment only on their own questions until they get 50 reputation points. Asaf's answer only mentions that posting is prohibited for unregistered users (which I assume means questions and answers, but the wording is slightly open to interpretation). But the comment posted by @FedericoPoloni also contradicts the 50-points-barrier. $\endgroup$
    – Martin Sleziak
    Commented Oct 16 at 5:53
  • 5
    $\begingroup$ I tested from a private window. It looks like unregistered users are only blocked from asking questions and not from answering questions. The ability for unregistered users to post comments is a separate feature, which converts any answers from any users that are "trivial" into comments - this bypasses all other limits on comments, including moderator-imposed comment locks. This has happened before to another site that requested it; it was resolved after a bug report. $\endgroup$
    – gparyani
    Commented Oct 16 at 8:25
  • 2
    $\begingroup$ @FedericoPoloni: Should be fine now. $\endgroup$
    – Asaf Karagila Mod
    Commented Oct 16 at 16:13
  • 4
    $\begingroup$ @AsafKaragila Thanks! Now it looks different, indeed I have to log in first to add an answer; the method I used doesn't work anymore. $\endgroup$
    – Federico Poloni
    Commented Oct 16 at 17:09
  • 7
    $\begingroup$ Has this measure succeeded in stemming the flow of AI content? $\endgroup$
    – Alec Rhea
    Commented Oct 17 at 19:32
  • 5
    $\begingroup$ @AlecRhea: It cannot and will never be "the solution". But it helps curb a little bit of the automating process that one can use, and it makes a slightly more work of creating an account and posting. $\endgroup$
    – Asaf Karagila Mod
    Commented Oct 19 at 8:13
  • 4
    $\begingroup$ My impression is that AI-generated posts almost disappeared since introduction of the new policy. Is this actually the case? $\endgroup$
    – Moishe Kohan
    Commented Nov 3 at 14:36
122
$\begingroup$

Yes, we should require registration (i.e. we should change things from how they are).

Upvote this answer if you agree with it.

Feel free to edit in reasons.

My (Sam Hopkins's) thoughts: for whatever reason, we are clearly seeing a big increase in the amount of AI junk here on MO, and it is making the site worse (see e.g. the other question). Many of the AI spam users are unregistered, and adding an extra layer might deter them.

$\endgroup$
7
  • 25
    $\begingroup$ I would go further, and prohibit AI training without permission from the people who generated it, although that ship sailed a long time ago. $\endgroup$
    – Andrej Bauer
    Commented Sep 26 at 8:38
  • 16
    $\begingroup$ I think likely we will need to go further and prohibit registration without confirmation the participant has a pulse and the intent to talk about mathematics rather than train computer algorithms. But registration is a productive step. $\endgroup$
    – Ryan Budney
    Commented Sep 26 at 17:57
  • 7
    $\begingroup$ For the record, there is currently one downvote. I believe the intention is to count the upvotes on the two alternatives, so downvotes should not count at all. Otherwise someone can vote twice (up for one, down for the other), which does not make any sense -- then everyone would have to vote twice to have an equal vote. $\endgroup$
    – Jukka Kohonen
    Commented Sep 29 at 22:30
  • 1
    $\begingroup$ @JukkaKohonen Stackoverflow already has a perfectly reasonable way of counting votes. There are evidently 9 possible ways of voting on this pair of answers (upvote, downvote, or novote for each), each with an evident effect on the vote tallies, nothing tricky about it. There's nothing to be gained by changing the rules. $\endgroup$
    – John Pardon
    Commented Oct 1 at 0:31
  • 5
    $\begingroup$ I am not proposing any changing of the rules. I am merely reminding what the OP already specified as the voting protocol for this question: Upvote if you agree. Nothing more; thus the number of upvotes is what is counted (how many agree with answer 1; how many agree with answer 2). You can downvote, flag, comment, whatever you like but that is not an upvote, and is not part of "the number of upvotes". $\endgroup$
    – Jukka Kohonen
    Commented Oct 1 at 13:33
  • 3
    $\begingroup$ @JukkaKohonen, re, except, which you obviously know, but just to say it explicitly for anyone unfamiliar with meta norms, that flagging is not an appropriate response to a post just because you disagree with it. (The only currently valid flag reasons are "Spam", "Rude or abusive", "Not an answer", or "In need of moderator intervention", and I believe that it is clear that none of those currently apply. ("Currently" as a weasel word, in case this CW post is later edited.)) $\endgroup$
    – LSpice
    Commented Oct 1 at 18:19
  • 3
    $\begingroup$ @LSpice, you are correct of course, thanks for the correction. "You can flag" was an unfortunate wording from me, what I tried to convey was "Whether you downvote, flag, comment, it does not matter for the counting of the upvotes". $\endgroup$
    – Jukka Kohonen
    Commented Oct 2 at 0:32
1
$\begingroup$

No, we should not require registration (i.e. we should keep things the same).

Upvote this answer if you agree with it.

Feel free to edit in reasons.

My (Sam Hopkins's) thoughts: some legitimate users prefer to post anonymously. While you can still register an account with a pseudonym, I think registration requires at least an email address (correct me if I'm wrong) and perhaps some legitimate users might not want to supply that.

$\endgroup$
5
  • 22
    $\begingroup$ It’s easy to get anonymous email addresses from gmail, proton mail, etc. Requiring email addresses does not force anyone to disclose their identity. $\endgroup$
    – Andy Putman
    Commented Sep 26 at 2:50
  • 5
    $\begingroup$ I recently got a good answer from an unregistered user, and in my area (mathematical logic) I have not seen AI generated posts yet, but I guess my experience is not typical. $\endgroup$
    – Dmytro Taranovsky
    Commented Sep 26 at 17:14
  • 8
    $\begingroup$ One possibility (requiring software changes) is to make posts by unregistered users limited visibility until a positive review. That is you can still see them if you want to, and you get notified if your question is answered by an unregistered user, but they will not clutter the front page and the like. Also note that a registration requirement may shift AI spam from unregistered to newly registered users. $\endgroup$
    – Dmytro Taranovsky
    Commented Sep 26 at 18:11
  • 1
    $\begingroup$ @AndyPutman Small correction - from ProtonMail, yes (and alsoTutaNota) but GMail requires a telephone number or second email address on sign-up. $\endgroup$
    – Robert Furber
    Commented Oct 3 at 15:23
  • 1
    $\begingroup$ @RobertFurber is it new? Last year I still got a new gmail address without supplying any info. $\endgroup$
    – Vladimir Dotsenko
    Commented Oct 31 at 1:25

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .