Wikipedia talk:Delegable proxy/Abd's message

Copy of mail sent to: English Wikipedia <wikien-l@lists.wikimedia.org>

At 12:23 PM 2/8/2008, WJhonson@aol.com wrote: >In a message dated 2/8/2008 9:21:00 A.M. Pacific Standard Time, >ian.woollard@gmail.com writes: > >I cannot be kind about this, these people are engaging in, or >recommending OR, and are trying to hide behind the cloak of consensus. > >We don't want or need consensus in the Wikipedia, we want *informed* >consensus.>> >--------------------------- >When experts are "called in" to give opinions in contentious issues, the >warriors cite Canvassing and Meatpuppetry. > >How do we address that?

Favorite question, actually. This happens to be my specialty, i.e., the theory of how very large organizations might be able to efficiently measure consensus, plus guarantee (at least in general) that the consensus is an informed one.

I apologize for the length of this, but I don't have time to edit it down, so ... following the principles implied herein, if anyone finds this of value, commenting on it and repeating what seems important about it could be very useful. It is not necessary that everyone read this, and, indeed, one person might be enough. So, if it serves you to read it, great. If not, I will not be offended if anyone passes it by.

We might notice that the problem described above is a generic organizational one, and it starts to bite organizations when they reach a level where, for example, if everyone shows up at a meeting, or even a major fraction do so, the meeting becomes too large to function, plus there is classic participation bias. Participation bias is a double-edged sword. On the other hand, it tends to favor decisions being made by the informed. However, there is also a higher incentive to participate on the part of those who are biased in some way, who have an axe to grind. So direct democracies, such as New England Town Meeting, tend to be replaced when the town size reaches a level that the number of fanatics and blowhards and tenacious debaters becomes large enough -- they are a certain percentage of any population that does not exclude them -- to raise the noise to a point that the meetings start to break down, the gatherings become tedious, expand the time consumed, increasing participation bias even further. Familiar?

Mailing lists and the like (including internet forums, which is what project or talk pages are on Wikipedia) increase the manageable scale, because, unlike the case with face-to-face meetings, readers can selectively read, they do not have to sit through a long and boring speech, and, if there is voting involved, they only need pay attention to actual and generally short questions, plus, to be informed, they may read a few posts from writers they trust, or skim over what has been posted. However, there comes a point where, again, the noise level is too great. That is not a specific size, because there are mailing lists with tens of thousands of subscribers that remain functional -- though mostly because these lists aren't really trying to make any kind of decisions, they are only discussing. When they try to make controversial decisions, there can be, if there is no process to prevent it, eruptions of highly contentious posts. At a certain point, though, *any* open mailing list is going to run into the problem of scale. I am only here considering unmoderated lists.

There are classic solutions or processes, each one unsatisfactory from various perspectives:

(1) Electoral democracy, where elected representatives reduce the number of participants. The big problems with elections is that security issues are raised: who can vote (sock puppets?), how do they vote, and who counts the vote. Further, whenever there is an election, there are losers, which may represent voters who end up unrepresented. There are voting systems that ameliorate this problem, so some kind of electoral democracy can be a solution, and this is generally how societies, when the exercise of sovereignty is concerned, have moved, and this and the second solution are the approach generally taken by nonprofits.

(2) Oligarchical control. There is single person or a small number of persons who have decision-making authority, and membership participation is limited to advising those people. Electoral democracy is, in one sense, oligarchical, but I distinguish this here by considering that the oligarchy is a fixed oligarchy, not subject to election but which elects its own continuing participants. This is actually the Wikipedia formal control structure, but it operationally, for the most part, defers to the third process.

(2) Anarchist approaches such as are (more or less) the status quo on Wikipedia, which works much better than some would expect but which turns out to be, as well, horrifically inefficient in ways that, in my opinion, will make it unsustainable. Like consensus process in voluntary communities, over the years, participation declines as people decide that they have better things to do that sit in tedious meetings, going through what it takes to find consensus. (Note that I am in favor of finding consensus, and am merely pointing out that traditional methods of finding it, when the group size gets large, become increasingly impractical). This approach does have legs, as Wikipedia shows; but participation bias can, again, cause the apparent consensus, as was being pointed out, to deviate from what would be a true consensus if one explained the situation to every member and asked them to vote.

(3) Proxy democracy, as practiced in the business world (where it is a democracy of property, i.e., own so many shares, one gets so many votes, and may assign those votes at will, to anyone, to cast. This allows, in theory, all shareholders to be represented at an annual meeting, whereas only a few are actually present. It is rigorously democratic, in theory, but in practice various conditions have developed that frustrate it, including large numbers of uninformed shareholders who assign proxies on request by the board of trustees, thus making proxy revolutions very difficult. As an example, the California State Automobile Association is a proxy democracy: members each get one vote. But this organization, like many in its class, was founded by people selling insurance, if I'm correct. And the board sends out proxy solicitations, which enough members helpfully sign and send back, that when an insurgent (perhaps pursuing his own agenda as a trial lawyer) tries to mount a candidacy to represent other interests than those of insurance companies (which only partially coincide with those of motorists), it's almost impossible, and there is no means of reaching the members directly, often, since the board controls access to the membership list and member addresses. But it is proxy democracy that contains the seeds of a solution, because it can create full representation without elections. Proxy democracy has seen little use in most nonprofits; however, as an example, some Green parties allow members to name proxies for voting at conventions; anarchists have commonly opposed this because it can create a situation where the anarchists have a majority at the convention, but are outvoted by those holding proxies. This opposition to proxy voting is actually quite generic, and most nonprofit membership organizations prohibit proxy voting, and discussing why this is the case is not necessary here, except to say that it is basically the same argument, made by the oligarchy that has effective control through participation bias, and believes that this control is essential for organizational success.

Wikpedia actually incorporates, already, many elements of what I would consider an advanced system. In particular, one critical aspect of such would be the separation of judgement from power. That is, there is advice and there is action, and the advice, properly, should be disconnected from action; this principle is more commonly known as the "independence of the judiciary," and represents avoidance of the direct exercise of power by judges. While it may appear that judges control, generally they do not; rather they issue decisions which may be considered binding, but the judges themselves typically cannot compel compliance themselves; rather, public servants ("the executive branch") act as authorized by the decision of a judge, and if the public servants refuse to act, the judges have no power. But, of course, that is not normally what happens.

With Wikipedia, ArbComm is technically only advisory. Jimbo Wales is not legally obligated to follow the decisions of ArbComm. Nor is an apparent community consensus binding on anyone. Rather, an ArbComm decision or an apparent consensus are enforced, if they are enforced, by individuals entrusted with enforcement powers, and none of them can be compelled to exercise that power, unless they choose, themselves, to be bound, and, if I'm correct, such choices are revocable. Ultimately, legal authority over Wikipedia is in the hands of the Foundation, i.e., the Board. The community consensus advises the Board, and the Board would disregard it at its own peril. Legally, however, the Board is obligated (and the members legally obligated, with possible teeth of they neglect the obligation) to insure that Wikipedia, for example, acts responsibly, for the protection of the public, so it is possible that a clear community consensus could appear, perhaps requiring certain expenditures, and the Board properly would refuse if it considered this fiscally unwise, risking bankruptcy and thus harm to the public.

So what is the function of the community consensus on Wikipedia? It is to generate and express advice, whereby the community advises itself and its servants and those who hold responsible positions as volunteers or employees. The consensus does not control directly, but it may effectively do so; in each case, however, each action of control is at the discretion of individuals entrusted with the power of action.

This kind of organizational/legal structure has been used before, with phenomenal success. It was used by the model for what I call Free Associations. Bill Wilson, co-founder of Alcoholics Anonymous, carefully studied what had caused similar temperance movements to fail in the past, and he consciously designed AA to avoid the pitfalls. He separated AA itself from the service board known as AA World Services, Inc. AA itself "ought never be organized," but then he went on, "but we may create service boards or committees directly responsible to those they serve." There is no central control in AA, not over the fellowship itself, nor over meetings or members. Individual meetings are fully independent, and may, if they choose, defy even a very strong majority opinion in AA (even what can be called a consensus), normally without any consequence at all. But it doesn't happen much! Each meeting is self-supporting; no action is taken centrally to start or help maintain local meetings. There is, however, a representative conference once a year, using delegates elected by a special process calculated to facilitate the development of consensus at the conference level by including some representation of significant minority opinion. I'm not presenting AA as perfect, it isn't, but it is very, very good at showing how anarchistic structures can function and be extraordinarily successful. Whatever one might think of the AA program, it became, in a few short years, almost the only show in town, and there is, in fact, at least one meeting in every town in North America and in many around the world.

However, AA was able to succeed largely because there are very few decisions to be made on the level of the overall organization. Every meeting is independent, and decisions are made by those who participate, and there is no enforced common standard. It would be as if, here, there were no enforceable policies and the editorial standards for every article were determined by the editors of that article. (To some extent, this is true, and it is both beneficial and harmful, depending. A consensus of editors can pretty much get away with anything, until and unless the consensus breaks, by an editor noticing it who thinks it wrong. It is not a consensus any more.) Universal membership is an important aspect of this. With a few exceptions -- which have been vigorously debated and are debated still -- every AA meeting is open to every AA member, and membership in AA is self-defined. I.e., say you are a member, you are a member unless it is proven otherwise. "The only requirement for membership is a desire to stop drinking." Further, "Any two alcoholics gathered together for the purpose of sobriety may call themselves an AA meeting, provided that, as a group, they have no other affiliation." This "other affiliation means, as I understand it, that there is no "Baptist AA," where you would have to be a Baptist to attend the group. (However, the exceptions: there are women's meetings and men's meetings and there are, I think, some gay and lesbian meetings; these are theoretically violating the AA traditions -- but they also have Rule Number One, IAR. As long as those who maintain local meetings lists don't exclude the meeting from lists, there has been no consequence; meeting lists will warn potential attendees of any such special requirement, and these special meetings don't exist in areas where alcoholics don't have lots of other options.)

But what would happen if the scale became very large, with common standards being expected for all meetings? In AA, meetings do tend to have strong resemblances, and there are strong traditions about, for example, no cross-talk, i.e,. avoiding commenting directly on what others have said, focusing on "principles, not personalities." But there is no overall rule requiring this, with sanctions against those who violated it. Comment on someone's sharing of their experience, and the most that is likely to happen is that people will shun you, or quietly suggest to you that it was inappropriate. (On the other hand, AA is composed of human beings, including some who are quite contentious, sometimes still fueled by their disease, and so whatever can happen probably does happen, occasionally, at AA meetings. I've only been to open meetings, I am not an alcoholic, but an observer of AA.)

Here, we have an encyclopedia to build and maintain. It is expected that certain common standards will be maintained, and maintaining them requires increasing precision of definition, plus application of the standards to real situations, which constantly vary in implications, there are exceptions to every rule except one.

So how to judge? This is where structure comes in. Wikipedia has a combination of the above described forms, but most decisions are made through the anarchist model, subject to participation bias. And, above, there is one method described that can balance out participation bias without excluding minority opinion, and that is proxy representation. However, creating a proxy structure could seem to be adding complexity and perhaps bureaucracy, and, when suggestions like this are made, most think of proxies as holding power, controlling. But if we look at proxies as elements in a system that only advises, but does not control, the picture changes. The security issues disappear. The elements we love about Wikipedia, such as "no voting," -- which means that decisions are made by servants based on arguments presented, not votes -- are retained. However, sometimes the process by which arguments are presented is horrifically inefficient, sucking in the expenditure of huge amounts of editor time, and often frustrating to all sides, plus, of course, a lot of it is, indeed, voting. I don't know what "Delete, per nom" means, otherwise. Obviously a lot of Wikipedia editors think we are voting in an AfD.

(In AA, the General Service Conference controls nothing, but its advice is respected. It nominates Board members, which the Board routinely elects. It approves of edits to publications, to insure that they reflect consensus. -- a "consensus" for these purposes is considered a 2/3 vote at the Conference level.)

Contentious decisions commonly involve interpretations of guidelines and policies where it is possible that there are quite legitimate alternate interpretations. While the theory of the process is that it is arguments that are being considered, not votes, in the end, in practice, an actual decision is made by one person, sometimes a person who has less experience and understanding of the arguments than those who commented. In the end, this is one person's decision. I would keep it exactly that way, but I would focus on what advice that person has received and to what extent it may be considered to be the collective understanding of the community. Suppose we had a proxy system -- never mind how we would get it for the moment, please -- so that there were a few commentors, each one of which presented a different comment, it becoming a tradition not to repeat comments, i.e., not to make comments of the form, "Me too." So there might be, with a controversial decision, two comments. Are they equal in power? Presently, in theory, yes. The decision maker has the right to choose, or even to make some totally different decision, having been advised. (in this case the most common decision would be no consensus, perhaps with renomination seeking additional comment.)

But, now suppose that one of these comments is from a high-level proxy, representing, directly or indirectly, 99% of Wikipedia registered users who have participated in such a system, and lets suppose there are, at least, thousands of them. This person has been massively trusted as being likely to have a good understanding of the issue and to comment fairly and neutrally (or, at least, representing in some sense the POV of those who chose him or her). And the other comment is from a new editor, who has been trusted by no one. *other things being equal,* which has more weight? Indeed, which decision is more likely to stick and not be reversed.

I can imagine some immediate objections, so I urge a careful reading of the above. "Other things being equal" is very important. This includes, for the deciding administrator, indecision as to which argument presented is the best. First of all, is this an inconclusive decision, "No consensus"? I'd suggest not! The fact is that admins do use the number of "votes" to determine the validity of a decision, and will consider the existence of opposing arguments as not relevant if one side is snowed. Decisions contrary to the majority are made, but I've never seen a snowed expression of opinion reversed by an administrator. But snowing takes up editor time. Rather, I'd suggest, weighting "votes," acknowledging them as votes, *but not making them binding*, is actually the Wikipedia way, it is what has been done *to a degree*, already. All that a proxy system would do is to measure participation (how many editors participated in the decision-making process, directly and indirectly) and the sense of the community (how many editors may we predict would vote a certain way if asked.)

(In this example, even though one proxy represented thousands of users, a careful decision maker might want to see at least one concurring opinion, and for the proxy or any user to contact another user to look at the issue could be quite appropriate. Here, I'm sure, if this were a matter of weight, we'd see more comments anyway, and, given the position of the proxy, if this is working, it's highly likely there would be some confirmation. But what I'd predict we would not see is an MfD with 600 votes and highly repetitious argument.)

Now, as to the latter, when proxy systems are proposed, one of the most common objections is "How can I be sure that the person I pick will vote the way I would want." The answer, of course, is that you can't. A proxy should be chosen as a person one you consider most likely to make a good decision if you don't participate. In the example above, with that high-level proxy, suppose you were a client of that proxy. You might watch what your proxy is doing, and if you see a vote you don't like, you simply vote yourself, effectively cancelling your vote. If this happens more than rarely, you would properly consider changing your proxy assignment. But, I suspect, most people would not watch their proxy unless someone called misbehavior to their attention. (If a proxy is blocked such that the proxy cannot continue to function, i.e, more than transiently, all the direct clients, I'd think, should be notified. But all this can be done without formal structure to do it.)

I've been thinking of Delegable Proxy, where proxies are considered to be delegable, so if A names B and B names C, then C, in the absence of A and B, represents both of them. DP is indefinitely scalable, the "hierarchy" it creates is bottom-up, a fractal (hence "fractal democracy" is one of the names for this), requiring no elections, only a list of proxy assignments. Wikipedia needs no special tools to do this, it is all in place, everything needed, except people assigning proxies. No policies or guidelines need be changed, no additional burden placed on anyone that is not voluntarily assumed, no power transferred from those who currently hold it. No bureaucracy is needed, beyond the thinnest.

Closing administrators would not be required to consider proxy expansion of votes. Any user could expand votes using the proxy list. (It's not hard to do it by hand, even with thousands of votes involved, but there are also tools being developed and I assume that there will be quite adequate ones available readily as open-source software.)

Now, why would I think that the vote of a proxy would estimate overall consensus "if the users were asked." First of all, of course, the proxy only estimates the consensus of those represented by the proxy, and a collection of proxies only would estimate the consensus of all those collectively represented by those proxies. However, proxies and clients will directly communicate. I would never name a proxy that I could not contact directly and reasonably expect a response. One person representing thousands of users, directly, is probably going to do a very poor job of communicating with them, and the vote of this person, I might tend to discount. (And proxy analyses, in the end, are the ultimate responsibility of the one doing the analysis, who is not bound as to how to do it, the analyst could, for example, deweight or even ignore the votes originating with new accounts, or could deweight or ignore votes coming from identified users with strong POVs being pushed, or according to any standard at all. People analyze proxy expansions in order to help themselves make decision, or to advise others who trust the accuracy of the expansion. It's personal.)

Now, I've actually only scratched the surface of what could become possible with a proxy network in place. Collectively, the users have disposable income and resources that dwarf those of WikiMedia Foundation. If the users are coherently connected as is hinted here -- don't think that this description, while it is long, is in any way complete -- and there were some financial emergency, extraordinary sums could be raised in very short order, with practically no effort. All you have to do is convince a *few* high-level proxies and they will quickly and efficiently do the rest, through personal contact. My proxy would have my phone number. It's an automatic telephone tree, the person calling me to ask for a donation would be the one I chose as trustworthy. Note that this is *not* some central decision being made to ask all members for a donation -- as happens with the current system (but not personally), and that I'm not proposing this as a change; I said "emergency." And those proxies who thought it was a bad idea simply wouldn't do it. But, before any action was actually taken, there would be a reasonable estimate of success, based on a poll and proxy expansion of it.

The same with decisions about guidelines and policies. If there is an RFC, and for any given editor, we may assume that, roughly, an individual editor is likely to accept and follow a guidelines that has received the assent of his or her proxy, plus, of course, the proxy may have explained it to the client. Every user is more likely to think of guidelines and policies as being created by "us," rather than from on high; I think there is currently some cynicism as to how much guidelines and policies currently reflect actual community consensus, and this, in turn, fosters disregard for them.

None of this would cause any *immediate* change. Am I calling for action? Not much! However, if these ideas are of interest, a reader might followup on what occurred to me as a joke, more or less, I registered a sock puppet, User:The Community. This came from a comment by User:Zenwhat on the Village Pump, asking sarcastically, after it had been noted that the community was ignoring source reliability guidelines, "How can we block The Community (TM) for disruption?" Well, it's possible now.

But that account isn't going to be used for any editing outside the User page and Talk page, and quite possibly none there as well, except as authorized by, at least, some actual community.

And, in fact, I'm not likely to do much at all there, in terms of implementing, for example, a proxy list and explaining and proposing other details, if there is not at least one editor who seconds the effective motion to create this account to serve as a "secretary" for the community. Right now, I'm user "The Community," but, as such, since I have not been advised by the community by any means which I consider truly representative, I must, in that function, remain silent, and I am merely a volunteer serving temporarily until replaced.

None of the actual actions that I would propose, if this second appears and there is some level of participation, will require overall community approval. They are all legitimate under current guidelines and procedures and would be highly unlikely to be disruptive in any way. On the other hand, it is possible that some will consider this threatening for various reasons. As an example, people who are very comfortable with the status quo, who believe that they understand the best interests of Wikipedia better than the average user (and they are probably right), may think that this could have an ultimate effect of reducing their influence. While they may be correct, I'd suggest that a system as I would envision it could enable them, if they are correct about their wisdom and understanding, to become more effective with less effort. Such people would naturally become high-level proxies if they do, in fact, represent the mature understanding of the community and wish to so serve. High-level proxies would be proxies who have been trusted by those who are themselves heavily trusted. These higher-level proxy assignments would be those made by experienced Wikipedians, who know very well the details of how this user operates, and the depth of understanding and integrity shown in his or her edits. Proxy democracy is a peer system, even though it tends to create a hierarchy, rooted in personal judgement and trust.

If it does not work, very little effort will have been expended. No software changes (unless it turns out to work well enough to justify that; there are already people working on public-source delegable proxy systems and I could see the possibility that MediaWiki might reasonably have a proxy field for each user and some tools for calculating proxy expansions, though I, myself, prefer to *not* make proxy expansion a central tool, for security reasons.)

I am calling a meeting for the purpose of considering the voluntary and decentralized implementation of proxy democracy on Wikipedia, for purposes as described above, and I have reserved a space for it, User Talk:The Community. I have, there, volunteered, as User:The Community, to serve as temporary chair of that meeting. Considering this as a motion to elect, if there is a second to that motion, discussion of it may begin there. (If it is seconded, amendments would be in order to change the designated secretary, or the duties, or whatever.) Currently, there has been a little discussion, from a user, one of the unsung Wikipedia heroes who patrol new page creation, wondering WTF this was about. Other than that, while the formation of the account was mentioned on the Village Pump, policy page, it appears to have been unnoticed.