I'm part of several communities that are relatively decentralized. For example, anyone can host a contra dance, rationality meetup, or effective altruism dinner. Some have central organizations (contra has CDSS, EA has CEA) but their influence is mostly informal. This structure has some benefits (lower overhead, more robust) but one drawback is in handling bad behavior. If several people reported very bad experiences with someone at my local dance we'd kick them out, but that wouldn't keep them from harming others at, say, any of the hundreds of other events run by other organizations.
I have seen cases, though, where someone was fully removed from a decentralized community. Looking at why these cases succeeded and others failed, I think it took:
Clear misbehavior, in a way that nearly everyone would agree if they looked into it.
Detailed public accusations, so people can look into it if they doubt the consensus.
The combination of these means that you can have an initial burst of 'drama' in which lots of people learn about the situation and agree that the person should be kicked out, and then this can be maintained whenever they show up again. For example:
2016: Gleb Tsipursky from the EA community, for a range of shady things and a pattern of apologizing and then continuing (details).
2017: Jordy Williams from contra dance, after accusations of grooming and rape (details).
2018: Brent Dill, from the rationality community after accusations of sexual abuse, gaslighting, and more (details).
Unfortunately this approach relies on people making public accusations, which is really hard. We should support people when they do and recognize their bravery, but people will often have valid reasons why they won't: fear of retaliation, unwilling to have that level of public scrutiny, risk of legal action. In those cases it's still possible to make some progress privately, and we definitely need to try, but you keep bumping into the limitations of decentralization and defamation law.
To clarify why I think (1) and (2) are key for community-wide exclusion, however, let's look at two cases where this has been tried but was only partially successful. Within EA I've seen people apply this approach to Jacy Reese, but because the details of what he was apologizing for are vague he's only mostly kicked out and people aren't sure how to view him.
The second case is Michael Vassar, in the rationality community (with some overlap into EA). He's been mostly expelled, but not as clearly as Tsipursky/Williams/Dill. His case had (1) and (2) but not entirely:
Some alleged misbehavior was clearly bad (sexual assault) but some was strange and hard to evaluate (inducing psychosis).
The sexual assault allegation was public, but it was in a hard-to-follow Twitter thread, included accusations against other people, and mixed accusations of many levels of severity (assault, distasteful, being a bad partner). I don't fault the accuser for any of this, but as a "here's a link that explains the problems" resource it didn't work as well as the three more successful cases above.
The psychosis allegations were even harder to link, scattered across multiple threads with people changing their minds.
Vassar was widely banned (REACH, SSC), but there were still holes. For example, he was invited to speak at an online SSC meetup. With the publication of the Bloomberg article on abuse in the rationalist community, however, which contained additional allegations and also provided something clear to link to I think we now have (1) and (2) and the expulsion will stick.
Disclosure: while I'm on the BIDA Safety Team I'm
speaking only for myself. My wife is on the Community
Health Team at CEA, but I haven't run this post by her and don't
know her views.
I spent some time trying to think about a solution, but all solutions I imagined were obviously wrong, and I am not sure there exists a good one.
Problem 1: Whatever system you design, someone needs to put information in it. That person could be a liar. You cannot fully solve it by a majority vote or whatever, because some information is by its nature only known to a few people, and that information may be critical in your evaluation of someone's character.
For example: Two people alone in a room. One claims to be raped by the other. The other either denies that it happened, or claims that it was consensual. Both enter their versions of the event into the database. What happens next? One possibility is to ignore the information for now, and wait for more data (and maybe one day conclude "N people independently claim that X did something to them privately, so it probably happened"). But in the meanwhile, you have a serious accusation, potentially libelous, in your database -- are you going to share it publicly?
Problem 2: People can punish others in real world for entering (true) data in the system. In the example above, the accused could sue for libel (even if the accusation is true, but unprovable in court). People providing unpleasant information about high-status people can be punished socially. People can punish those who report on their friends or on their political allies.
If you allow anonymous accusations, this again incentivizes false accusations against one's enemies. (Also, some accusations cannot in principle be made anonymously, because if you say what happened, when and where, the identity of the person can be figured out.)
A possible solution against libel is to provide an unspecific accusation, something like "I say that X is seriously a bad person and should be avoided, but I refuse to provide any more details; you have to either trust my judgment, or take the risk". But this would work only among sufficiently smart and honest people, because I would expect instant retaliation (if you flag me, I have nothing to lose by flagging you in turn, especially if the social norm is that I do not have to explain), the bad actor providing their own version of what "actually happened", and bad actors in general trying to convince gullible people to also flag their enemies. (Generally, if gullible people use the system, it is hopeless.) Flagging your boss would still be a dangerous move.
At the very minimum, a good prestige-tracking system would require some basic rationality training of all participants. Like to explain the difference between "I have observed a behavior X" and "my friend told me about X, and I absolutely trust my friend", between "X actually helped me" and "X said a lot of nice words, but that was all", between "dunno, X seems weird, but never did anything bad to me" and "X did many controversial things, but always had a good excuse", etc. If people do not use the same flags to express the same things, the entire system collapses into "a generic like" and "a generic dislike", with social consequences for voting differently from a majority.
So maybe it should not be individuals making entries in the database, but communities. Such as local LW meetups. "X is excommunicated from our group; no more details are publicly provided". This provides some level of deniability: X cannot sue the group; if the group informally provides information about X, X doesn't know which member did it. On the other hand, the list is maintained by a group, so an individual cannot simply add there their personal enemies. Just distinguish between "X is on our banlist" and "X is banned from our activities, because they are on a banlist of a group we trust", where each group makes an individual decision about which groups to trust.