Wikipedia talk:Delegable proxy

Alternates / proxy ranking

Should we allow users to pick alternate proxies? E.g., A selects B as 1st choice and C as 2nd choice. If neither A nor B vote, then C casts votes for A and C. If neither A, B, nor C votes, then B's proxy casts votes for A and B. I think this increases the robustness of the system and allows greater choice on the part of the user. Sarsaparilla (talk) 21:59, 9 February 2008 (UTC)

Issue independence

I removed this text from the essay:

In one possible implementation, ranked proxies can be established by a user selecting a primary proxy, a secondary proxy, and so on. If the primary proxy does not participate in a discussion, then the user's vote can be cast by the secondary proxy. If the secondary proxy does not participate either, then the user's vote can be cast by a tertiary proxy, if there is one; and if not, then by someone in the primary proxy's proxy chain, or his secondary proxy's proxy chain, etc. This system, if combined with single-purpose accounts, could be used to implement issue independence in which different proxies cast votes on different issues. For instance, Ms. Gargamella might create two accounts, Ms. Gargamella the Geologist and Ms. Gargamella the Inclusionist. If Mr. User trusts Ms. Gargamella's views on geology-related articles but doesn't like her overall inclusionist philosophy, he might select Ms. Gargamella the Geologist as his primary proxy and Mr. Deletionist as his secondary proxy. Ms. Gargamella the Geologist would only cast votes on geology-related articles; on all other issues, Mr. User's vote would be cast by Mr. Deletionist.

I think this type of system would be über-cool, but maybe it's too complicated to try to introduce all at once? What do you think? Sarsaparilla (talk) 23:49, 9 February 2008 (UTC)

I'd say, keep it simple. Further, a ranked system like that obscures the return path. Would I accept a secondary proxy? Probably not. In a Free Association, I don't really gain anything by accepting a proxy; my power as "representative" is subject to revocation at any time, it only is actually effective to the extent that I can lead my clients. I don't want weak clients, who are going to argue with me on everything, who are going to complain about this or that action (which is quite different from asking why I did something). Indeed, I don't need clients at all. I present arguments, my arguments are the same no matter how many clients I have. And, remember, it's arguments that count, isn't it? However, my guess is that, if we had a system, I'd have a few clients and they would have a few cliens and there is no telling how deep it would go. Not everybody recognizes what I'm doing; indeed, few do, nor do I expect them to. What I want as a number of direct clients is only the number that I can effectively communicate with. If I had a thousand clients, each trying to influence me to vote their way, I'd go nuts. (Actually, not, I'd just withdraw my acceptance from more than a few, and I might recommend someone else to them, who might be a client of mine -- or not. DP (delegable proxy) is designed to distribute communications load.)

The single-proxy idea is tremendously powerful. I'm not at all convinced that adding the enormous complication of ranked proxies adds anything, and it dilutes the expression of trust. I trust the person to represent me, but not to make a good choice in delegating that? Note that when we elect public representatives, the fact is that they delegate most of their work to staff, whom they have chosen as trustworthy. Delegation is already a major part of existing system; DP is just a way of setting up the delegation in an extremely simple fashion.

There is, however, a very simple way to have special issue proxies: a special proxy list. Anyone can set one up, presumably for use in some narrow context. If a user does not use it, the general proxy stands. But, say, with Voting systems articles, I might name someone I specially trust in that area. The point is to not add any complication that a user *must* handle. One proxy assignment covers it all; but then, if the user wants to go to the trouble, a special proxy list can be used. —Preceding unsigned comment added by Abd (talk • contribs) 04:15, 11 February 2008 (UTC)

Green-Armytage's proposal

Of course, Green-Armytage proposed that proxy ranking be used for cases in which a proxy loop goes all the way back around to the original user. See http://fc.antioch.edu/~james_green-armytage/vm/proxy.htm#loops .

Ranked proxy lists can be used to resolve loops. One possible rule is as follows: "A vote shouldn't travel the same proxy path twice." Given the above case, A's vote has traveled the path A-->B, then the path B-->C, and then the path C-->A. Therefore, according to this rule, once A's vote returns to A, it should not once again travel the path from A to B. Instead, it should travel to the next proxy as ranked on A's proxy list. The proxy path rule is not very important, since such loops are not an especially daunting problem. Other rules are possible, for example "a vote shouldn't be assigned to the same person twice," in which case A's vote would be transferred to C's second proxy rather than being assigned to A once again.

Sarsaparilla (talk) 00:14, 10 February 2008 (UTC)

For reasons I've stated elsewhere, I'd prefer to leave possible complications up to those facing them. The basic proxy system should probably handle most situations. It's fail-safe in a Free Association context, where votes aren't moving power, they are merely advising. So if a loop causes some set of users not to be represented, first of all, it's possible to set up systems to notify those users (but this is way ahead of where we are, in a Wikipedia system), and, secondly, the overall statistics show unrepresented users. It is *not* essential that all users be represented, it is merely desirable. Notice that Green-Armytage mentions that there are many ways of doing it. What I say is, one step at a time. It is possible that ranked proxies could be of some use, but I don't like the diluted responsibility. Part of a proxy's job is to notify the client if there is something that the proxy thinks should require the client's attention. Which proxy is responsible for notifying the client?

When we get down to details, and see it operating, we will find many quirks that will need to be addressed. But getting to that position is quite hard enough without adding lots of bells and whistles in advance. What happens if we simply set up a proxy table? We need to find out!

But as to Green's idea, I would see that, if there is a loop absence, first all the proxies would be assigned according to the next rank. If there is no next rank, however, the existing proxy remains effective, so it is enough if one member of the loop has used another rank. This could solve certain problems that I have previously considered. The way I would technically handle it is with a different alternate proxy table, and this, again, leaves analysts free to use it or not, and they can actually do both. (We got this result using the primary proxy table, and it shifted this much using a secondary table -- which would only represent broader participation, it would change no votes cast in the first "round.") But -- we don't need to figure all of this out in advance.

It's very important to repeat that analysis isn't a central tool, in essence. It is something that anyone who wants to be advised does. (Sure, people can do it as part of *their* personal advice, but the point is that analysis is not centrally *controlled*, which allows it to be totally flexible and not particularly vulnerable to, say, manipulation by sock puppets, as would any fixed system with automated analysis and the raw data not visible. Minguo.info, I think, has such a system, I'm not sure. (We have recommended that the proxy table always be a public document, but some really want this to be secret. I wouldn't trust it, in the long run. Open access is a tremendous protection: anyone can verify that the proxy list is true for themselves and anyone they care to check it regarding, and anyone can use it for analysis in whatever way they choose.) —Preceding unsigned comment added by Abd (talk • contribs) 04:31, 11 February 2008 (UTC)

Expiration

Should proxies expire after a certain time? What about when users go inactive for long periods (e.g. several years), but someone is continuing to vote on their behalf? Sarsaparilla (talk) 22:23, 9 February 2008 (UTC)

It is up to whoever is analyzing the results. A past placement of trust is still a placement of trust. Please, stop referring to this as "voting on their behalf." It creates a host of impressions that aren't accurate. Proxy expansions are a means of estimating a broader consensus, balancing out participation bias.

When we have a proxy table with old proxies on it, we can consider how they might be handled, but, frankly, I'm not sure we should centrally interfere. One more piece of bureaucracy, something centrally imposed. However ... an account which has not edited.... we do need to be able to compile such statistics; I would, in fact, want to discount such (but it's simple: ask proxies to ping their client to verify: ask the client to confirm the proxy on the table.)

The table the way I would do it is not quite what Sarsaparilla set up. If a sig is put in the field by the editor, it would have with it the date of the action. So anyone taking that table and analyzing it would have date fields to use if they want. I'd certainly discount inactive accounts, but the period is up to me, if I'm the one who needs guidance. --Abd (talk) 04:36, 11 February 2008 (UTC)

Templates

I'm working on some templates. After you designate me as proxy, you can slap this on your user page: User:Sarsaparilla/Proxy Of course, the downside to this template is that it looks a lot like the sockpuppet template. Hmm, does anyone have a better graphic to use? Sarsaparilla (talk) 23:23, 9 February 2008 (UTC)

might work better. MBisanz ^talk 05:37, 10 February 2008 (UTC) or

Done! Sarsaparilla (talk) 06:06, 10 February 2008 (UTC)

Look, this will simply irritate people opposed to the idea. A simple table, which can be copied elsewhere for security, will do the job. Sure, if the numbers get really large, having it as a template would be one way of doing it, but .... the information actually flows in another direction, to use the proxy table, you start with a list of those who actually voted, and strike those records from the proxy table. Then the proxies in the proxy table are examined, and, if any matches are found with the list of those who voted, those records are added to the list of those who voted. And this is then done recursively until a pass finds no more votes.

Want to use secondary proxies? Run the procedure again, looking at the secondary proxies....

Anyway the point is that we need a list organized by member, not by proxy. Knowing how many people a proxy represents, if nobody else votes, can be useful in some contexts, but I don't see it immediately on Wikipedia. It's also not predictable until all the votes are in.... What if all my clients vote? Would I be upset? (Not!).

(Proxy rank has been proposed as a means of determining participation rights for a deliberative body, to deal with the noise problem. What happens on Wikipedia when some question really fires up the base? It gets pretty chaotic! It can become really difficult to vote, with edit conflict after edit conflict. Proxy systems, if significant numbers of members start to use them, could make all this much more orderly. If I have a proxy, I look at contribs for my proxy, and I'll immediately see if a vote was cast. I also can know, with a single proxy list, the whole chain of proxies. From the top, a DP hierarchy is a fractal, inordinately complex. But from the point of view of the user at the bottom, it is a single chain proceeding linearly, and, generally, looping back to itself at one point.

(Add secondary proxies and the whole picture becomes so complex that I can't visualize it. Do I have to check secondary proxies as well?)

What proxies will do is to make "me too" votes unnecessary. That's really the main effect as far as voting is concerned. Have a new argument to present that isn't already there. Vote and give the argument. Also vote if your opinion is different from that of your proxy or the one representing you indirectly. But most people aren't going to pay attention to most decisions. That is why it is far more efficient than polling everyone; polls generally produce decreasing returns as voters get tired of all the questions. DP polls, however, allow flexible representation by persons interested in the topic, in theory configuring the representation to fit the immediate problem. And there are many details to be worked out, but I don't see any of them as requiring centralized systems beyond a shared proxy table. —Preceding unsigned comment added by Abd (talk • contribs) 04:58, 11 February 2008 (UTC)

Userbox

There is also this userbox:

This user supports delegable proxy.

Sarsaparilla (talk) 06:07, 10 February 2008 (UTC)

Very Bad Idea

This is a Very Bad Idea. Whenever we have a !vote, we rely on knowing who is voting, and their reasoning, but also that the sample of people discussing is representative of the community (since we could never have every single user comment). By allowing someone to !vote via proxy you skew the sample in a non-random way. Additionally, verifying proxy !voting would create an enormous burden in cases where it matters. A thoroughly terrible idea, I'm sorry to say. – Mike.lifeguard | ^@en.wb 17:31, 10 February 2008 (UTC)

To make sure we're talking about the same thing, let me state my understanding of a couple terms. Skew refers to a non-representative distribution or distortion in a positive or negative direction. Distortion means an alteration for the worse.

Two problems that delegable proxy is intended to address are random fluctuations in participation which make debate outcomes unpredictable; and participation bias (resulting partly from opportunity cost) which skews discussion outcomes in a non-random way. Most debates include both of these distorting influences. The small sample size of some debates would make even a hypothetical random distribution unreliable. Deletion debates, for instance, are like a town meeting discussion in which less than 1% of the electorate shows up. Most jurisdictions have scrapped the town meeting system for the very reason that most citizens do not have time to adequately delve into the issues, rendering it vulnerable to slick campaigns and get out the vote efforts which affect the outcome. I favor initiative and referendum in the political realm, but it's undeniable that the number and complexity of measures that appears on the ballot confuses some voters, who may be inclined to abstain or to cast an ill-informed vote. Similarly, there are many articles on Wikipedia which involve arcane issues not easily comprehended without substantial investment of time, and when non-experts get involved, the decisions are often faulty. We have adopted Wikipedia:Canvassing#Forum_shopping in an effort to curb the worst abuses under the current system. The corporate world's solution to the low-turnout problem is proxy voting.

Depending on the implementation, technology might simplify the verification task to the point where the time saved by using proxies exceeds the time expended in verification. However, saving time is only one of the purposes of this system; perhaps more important is improving representational faithfulness and the integrity of the system.

If there is substantial participation bias in who participates in the proxy system, then yes, it could skew the results in an undesirable way. I say, let's try the experiment and if it proves a failure, then we can scrap it. But I think the literature on liquid democracy/delegable proxy suggests that it could work quite well for Wikipedia. It is used by the election methods interest group. Abd, can you provide us with some information on how DP has worked there and how it could be applied to Wikipedia? I'm going to post your message from the Wikipedia mailing list here, for readers' reference. Sarsaparilla (talk) 18:20, 10 February 2008 (UTC)

I certainly welcome the opportunity to clear up some misunderstandings. User:Sarsaparilla started this page to begin discussing the concepts, and they are not maturely presented, i.e., it is very, very easy to misunderstand what I would propose for Wikipedia, from the suggested implementations found here. User:Mike.lifeguard raises objections that are, indeed, quite obvious, if common assumptions are made. Let me address them, one at a time:

This is a Very Bad Idea. What's "this"? What "power" would it have, what would a "proxy vote" mean? "We don't vote," is very commonly said. In order to understand the proposal, all of these must be considered. Indeed, I do not propose that there be any formal effect from proxy "votes" at all, which disposes, just by itself, of the thrust of this objection, while raising another one, "If there is no formal effect, WTF in the way of good could this accomplish?" Separate question. One question at a time!

Whenever we have a !vote, we rely on knowing who is voting, and their reasoning, but also that the sample of people discussing is representative of the community (since we could never have every single user comment).

Yes; however, we have no means, currently, of verifying that those voting are a fair sample of the community. How could we know? We could poll the community, choosing users at random. But that would violate a number of traditions here, and it would create a new security problem, the old "who watches the watchers" problem. If someone could manipulate the supposedly "random" choice process, they could skew the outcome in a desired manner. Nevertheless, Warren Smith claims that the technical problem could be solved, with his proposal for DDJ: Direct Democracy by Jury. Citizens would be randomly chosen to serve on a Jury which would make legislative decisions. But this isn't what I have in mind, which is far, far simpler.

Currently, we know who is voting, and we, often, do some kind of analysis on the votes. For example, signs of sock puppetry are considered, and other characteristics of those voting. In fact, a closing administrator may use any standard he or she chooses. And this is the key: Votes on Wikipedia don't control bleep. They advise a servant, who uses his or her judgement. I would not change this at all, it's perfect. (And, on the other hand, the community, if it doesn't like what a closing administrator did, can easily undo it. So this seemingly arbitrary decision-making process that depends on the judgment of a single person is far safer and more efficient than it might seem to outsiders. Someone who knows that and who sees, in its current state of proposal -- which is very, very primitively expressed -- can easily think that we'd be dumping what they know, in fact, works well. Usually. It is when it does not work that there can arise some serious problems, and the process can waste a huge amount of editor time.

By allowing someone to !vote via proxy you skew the sample in a non-random way. Additionally, verifying proxy !voting would create an enormous burden in cases where it matters. A thoroughly terrible idea, I'm sorry to say.

Don't be sorry! Indeed, most people won't get so far as to criticize the idea. By taking the time to do that, this user is far ahead of the crowd. Now as to what he said:

First of all, there are no "proxy votes." There are only votes, cast by identified users, as there are now. But something else exists: a proxy table. No closing administrator is obligated to use it, at least not until and unless some community consensus appears to that effect, which is not a proposal I am making. Rather, any user may use the proxy table to "expand" a vote. The user may use the proxy table for standard, single-level proxy, but it will be far more useful to consider proxies as delegable. To understand how this would function to more accurately estimate overall consensus will take some extended discussion, I expect, very few will get it on sight. (A few do, they are quite rare, and some very, very bright people don't get it at first. Quite simply, this whole concept flies in the face of what we expect from organizational hierarchies.)

Some try to make the system complex, and, indeed, people will work out refinements that do that, such as single-issue proxies (on Wikipedia, these could be "article proxies," set up by any user adding a proxy table to an article, i.e., ArticleName/Proxy List). Again, the use of a proxy list is totally voluntary. Who uses it? Anyone who wants to better estimate the consensus that will appear if push comes to shove.

Here is how it might work: there is some dispute among the editors of an article, and, from, say, an RFC, there are three editors on one side and one on the other. What is the "consensus"? In strict terms, as often found in certain organizations, there is no consensus unless everyone signs on, and, indeed, the goal for articles is that this happen. But, as we all know, it does not necessarily happen immediately, if ever. There are some editors who simply won't compromise or accept what everyone else needs in order to be satisfied.

Now, I'm actually in such a situation, with an article that I work on all the time. There was an RFC put up basically challenging the position I had been taking on the article. Against me, almost alone except for occasional edits by occasional editors, were regular participants: a paid consultant for the organization promoting the topic of the article, a long-time Wikipedia user who lists himself as "associated" with that same organization -- but is not paid or an officer, as far as I know, and ... a user who was just revealed to be one more sock puppet of James Salsman, who has been an activist for that same organization. And then there is me, regularly attacked as an "anti" activist, though I have no formal affiliations. What's the consensus?

I'd claim there isn't any. Now, suppose there is a proxy table, and suppose that, from it, we can determine that I, directly or indirectly, represent a thousand Wikipedia editors, and they, collectively, represent a handful. The picture changes.

But this binds nobody. It is just a method of estimating consensus, by weighting the comments of participants. This can be used by someone making a decision. I can tell you that I'd use it myself to decide when to give up, and when not to give up, when to stand "my" ground -- which would be the ground of the community, quite likely, if I were indeed so widely trusted -- and when to accept that the community would not support me.

This kind of system actually creates a strong incentive to find consensus. Suppose that a conflict arises between two factions that, when we do proxy analysis, turn out to be more or less matched. If these factions actively dispute with each other, a great deal of heat can be generated, and not much light. In politics, you'd have two parties each spending great effort in order to produce an unpredictable outcome. There is economic incentive to find compromise, and the greater the degree to which consensus can be found, the more efficient the process. If a heavy consensus can be found, the result is a pushover, and this could be known and predicted in advance.

(This doesn't really work with "political parties," because they are generally set up and structured to see "victory" rather than, specifically, improvements to public policy. In fact, if a candidate from an opposing party changes his or her position to make a compromise, they will call this candidate a "flip-flopper." How dare they try to steal our supporters? Rather, I'd expect it to work with nonpartisan caucuses that want some *result* rather than an identified factional victory giving them power over others. I've often used Pro-Life and Right to Choose caucuses as examples. They may argue until the cows come home on where life starts, but there *are* issues on which they can come together, in theory, except that the political system associates issues together, so Pro-Life people, sometimes, end up having, in order to get what they want on that, voting for someone who is Pro-War. Or vice-versa. Delegable proxy systems, for various reasons, I expect, allow issue independence, because they do not depend on "officers." The power remains decentralized.)

But none of this is binding on anyone. I use the term "caucus" to refer to a group of people connected in some way by opinion, and seeking to act collectively. We call a proxy plus the proxy's "constituents" -- including indirect constituents -- a "natural caucus," which is defined by the single proxy "leading" it. (All the concepts of leadership are turned on their heads by delegable proxy, which is why I put the term in quotes. A proxy has no control over the clients, and does not bind the clients in any way. Indeed, the only power that the proxy has is to advise, and this will come out in more detail in specific proposals.)

Proxy expansions are used, generally, by two kinds of people (plus there are other applications I won't immediately address). First would be a person who needs to make a decision. In, say, an AfD, the closing administrator, in theory, is supposed to disregard the vote count, it is supposed to be a matter of the cogency of the arguments. However, if this were purely the case, and I were a closing administrator, I would delete from the page all votes of the form "per nom," or "per "username." These are pure votes, they add no argument at all. Would this be legitimate?

I'd think not. We *do* consider votes. An isolated opinion is not the same as one expressed or confirmed by many people. With proxy table in place, we set up a different method of considering "votes," which is a weighting of participant votes by a measure of how widely they are trusted.

Ah, what about sock puppets? Well, the closing adminstrator can use whatever tools he or she chooses to consider the vote weighting. For starters, an adminstrator who thinks this whole idea is bunk is completely free to disregard it. The risk? Well, if a proxy expansion shows something different than the basic vote count, and the administrator disregards it, and it is accurate, the decision is quite likely to be reversed. Is it canvassing to widely discuss the outcome of an AfD after it has closed?

The details of how all this will play out are not specified, and they don't need to be. Users and administrators and others who might be able to save themselves a lot of time and effort by being able to estimate consensus more accurately will figure out how to do it. For example, when we start dealing with large expanded vote counts, anyone can set up a tool to do an expansion that disregards all accounts registered after a certain date. Or that excludes a specific list of users considered by the analyst to be unreliable editors. In fact, all this already happens when administrators look at who has voted and the implications, only without the formal support of a proxy table.

A proxy table costs nothing. It take extremely little effort to set one up, and little effort to use it. (Tools are being developed that can take a user list and a proxy table and determine the expanded vote from a list of users and votes, but this can be done by hand, quite easily.) What a nomination of proxy means is "I generally trust this user to make a decent decision where I cannot myself participate," and, if it is delegable proxy, the decision of what proxy to name is include in that which is generally trusted. There is no assumption of specific agreement. However, in a proxy expansion, if the "client" has voted, and that client's proxy also votes, the client and all those who are represented by the client are not included in the proxy's vote total. In other words, if a client votes, it effectively, for that vote, cancels the proxy. Further, a client may revoke the proxy at any time. Here, with what we have now, it would simply be a matter of editing the proxy table to change the proxy assignment.

We also suggest that proxies be accepted before they are considered effective. Generally, I would place on a proxy table the assumption that, if you name someone as your proxy, you are giving that person permission to communicate with you, on your talk page, by email, or by other means (which can be removed as an assumption by some specific limitation -- but if there is a limitation, and I were using the table, I might disregard the proxy. I want proxies to represent actual relationships of trust and rapport.) Conversely, and this is very important when the scale gets very large, accepting a proxy is likewise permission to communicate directly.

Delegable Proxy will work best when direct client/proxy relationships are relatively few, i.e., we don't have single proxies directly representing many thousands of users. Some systems proposed limit the number of direct proxies, though I wouldn't do that. Nevertheless, I might discount the votes of such proxies in my own analysis of votes, because I would not trust that there is sufficient rapport between the proxy and all those clients that he or she would be able to bring them along if needed.

Proxy trees create automatic phone trees that can be used for rapid and broad communication in emergencies. It is not just that the proxy, in some sense, represents the clients, it is that the clients are quite likely influenceable by the proxy. If this were being used for actual direct decision-making, there would be lots of hazards that would have to be addressed. But it is not. It is being used as an advisory network, self-created, bottom-up, requiring no bureaucracy, elections, or rigamarole. Use it or don't. Nobody loses anything by not naming a proxy, except perhaps efficiency. --Abd (talk) 04:03, 11 February 2008 (UTC)

Nomenclature ("voting")

Should we call it something else besides "voting"? I.e., rather than saying, "x has the authority to cast votes on behalf of this user in Wikipedia discussions," should we say that x has the authority to "represent" or "speak on behalf of" this user? I can see that the word "vote" is raising red flags with some users. Sarsaparilla (talk) 19:01, 10 February 2008 (UTC)

No authority to cast votes, period. This is a way of running an analysis of votes weighted by proxy assignments. The voters vote for themselves, not for their clients. However, an analyst, with a need or desire to know, may use the proxy assignments as a way of estimating how broadly the user is trusted, and thus to, to some extent, deskew poll results.

As soon as we say that the proxy is voting for the user, we then create a picture of decisions being made by "vote." Wrong picture, at least here. No, the system is the same as now, formally. --Abd (talk) 05:03, 11 February 2008 (UTC)

What is an accurate but concise way to describe it? I think the project page is in need of a rewrite to replace all the references to votes with some other terminology. Sarsaparilla (talk) 17:30, 11 February 2008 (UTC)

User talk:The Community

There is also a discussion going on at User talk:The Community. Sarsaparilla (talk) 19:07, 10 February 2008 (UTC)

That account was set up to be a servant of "the community," specifically of those who participate in any rcommendations it makes through this process. That is, an open poll might be held, and the servant using the account The Community, serving as the secretary of this "meeting," would report the results. And might also unlock the doors before the meeting and make the coffee and put out the chairs, then stay to make sure it is all put away after the meeting, the lights turned out and the doors locked. It's a highly-sought after position in 12-step programs because people know they will benefit through the service, but not so highly-sought after that I have ever seen any contention over it. Somebody volunteers and everyone else says, "Fine." And if he or she fails to perform, someone else does it. "Our leaders are but trusted servants, they do not govern." --Abd (talk) 05:10, 11 February 2008 (UTC)

It could be implemented now

Another interesting aspect of using delegable proxy is that there could be the equivalent of one-time proxies taking place on an impromptu basis. For instance, suppose that I am the proxy of 5 users in a deletion discussion with two people who each are proxies for about 500 users. I make a really insightful argument for keeping the article, and they both vote "Keep, per Sarsaparilla." The votes of the users who appointed those proxies have basically been redirected toward my position, even though I was not their official proxy or even their proxy's proxy, according to the proxy table. In this sense, the proxies are not just delegable on a permanent basis, formalized through listings in the proxy table, but the delegation can also take place spontaneously on an ad hoc basis. Wikipedia's flexibility makes it well-suited for this system, in my opinion; the only hindrances I can see are technological, and of course inertia and cautiousness of the community toward adopting major changes in processes and structures.

A proxy system could be set up now, in which one programmed a script to check another user's contributions and automatically vote "Keep per x" or "Delete per x" or even "Agree with x," as the context and the user's particular action may warrant. It probably wouldn't be too hard to program, and one could leave it running when they went on vacation or whatnot. The bot policy states that such processes require approval before they can be run. But is there any reliable way to detect and enforce it in this situation, without running the risk of false accusations? Given that implementing proxies through scripts or even manually checking another user's contributions and adding "me toos" to discussions is already possible, why not streamline the process by using proxy tables and automated methods of tabulating the proxies based on those tables? Sarsaparilla (talk) 21:23, 10 February 2008 (UTC)

WP:NOT a democracy

So delegating votes doesn't have any meaning.

I'm also worried about party-formation in that kind of situation. It could be extremely disenfranchising. --Kim Bruning (talk) 15:35, 11 February 2008 (UTC)

One person cannot carry out the examination and deliberation of a hundred

...and it's silly for us to pretend that one can.

• The proponent of this proposal has said it will be useful in dealing with "gray area[s]", where "subjective" decisions about notability are called for. The contentious cases are the ones where the most eyes are called for, not the fewest.

• Discussions about deletion (for example) would no longer be about making sound arguments, but about trying to persuade a small number of elite proxy holders.

• The effect of interpersonal conflicts between large proxy holders would be magnified. Outcomes of AfDs would be distorted in cases where a large proxy holder has an unrelated personal anomosity towards a particular voter or nominator.

I offer a hypothetical case to illustrate the effect of fallible proxies.

A single editor can from time to time make errors about the notability or appropriateness of an article (again using an AfD as an example). For the sake of argument, assume that 95% of the time an editor will make the 'right' call. The remaining 5% of the time, they screw up a Google search, misunderstand the nominator's argument, misread policy, think the article is about a different topic, etc.

Consider a proxy holder that represents twenty 'voters'. Under the current system, about nineteen of those twenty will vote 'correctly' on any given AfD. Since one of the purposes of this proposal is to increase effective participation, it could be argued that all twenty wouldn't normally participate. If we instead assume that only five of those twenty actually vote on any given AfD, there is a better than 97% chance that four or more will vote 'correctly', and a roughly 0.1% chance a majority will arrive at the 'wrong' conclusion. In this same hypothetical case, the single proxy holder screws up 5% of the time.

This neglects 'network' effects caused by interactions between voters. More eyes on the problem means more editors who are going to point out the glaring errors (typos, confusion with similarly-titled topics, etc.). The more participants actually present at a discussion, the less likely such errors are to be left standing. Even if an early editor makes an error and fails to withdraw his vote, both later voters and the closing admin will be able to take the newer, corrected information into account. TenOfAllTrades(talk) 16:47, 11 February 2008 (UTC)

One could make the same argument about corporations. Suppose each shareholder has an equal chance – say, 95% – of making a good decision on any given vote. In this case, executing a proxy delegating voting rights for 20 shares to a single shareholder would create a greater chance of a bad decision because smaller sample sizes are more susceptible to random deviations. Mathematically, in this situation, as the number of voters approaches infinity, the percentage of shares being voted for a "good" decision will approach 95% in accordance with the law of large numbers.

However, the idea behind a proxy system is that people choose proxies whose chance of making a good decision is sufficiently high that it overcomes the disadvantages of not voting one's own shares oneself. A relationship of trust exists with that proxy, and the proxy, knowing the consequences if he makes a bad decision, has an incentive to be careful. It is more efficient, and results in better decisions, for a proxy to invest an hour in studying a proposal than for 20 shareholders to invest 3 minutes apiece. Many people under the current Wikipedia decision-making system take only a cursory look at the content before voting.

Let us look at another example of delegated decision-making on Wikipedia. We elected twelve ArbCom members who we believe are sufficiently trustworthy and capable of making decisions on complex cases. If all Wikipedians had equal ability and time to analyze these cases and render good judgments, then it would be a mistake to entrust these decisions to a group of twelve, because statistics would indicate they would have a greater-than-average chance of messing up than a larger group. Indeed, statisticians typically consider a sample size of at least 30 to be necessary to produce reliable results. See [1] You say above, "The contentious cases are the ones where the most eyes are called for, not the fewest." Could not the same be said about ArbCom cases? Yet, although any editor is allowed to provide input (as editors would continue to do under delegable proxy) we do not subject ArbCom decisions to community consensus. The reason we don't have many problems with people focusing on persuading ArbCom members rather than on the merits of the cases, or with interpersonal issues between ArbCom members, is that the members of this group have been chosen for integrity and other characteristics that would reduce the likelihood of such things happening. The same would occur with proxies. Sarsaparilla (talk) 18:15, 11 February 2008 (UTC)

The problem with many eyes is if we have more than Dunbar's number*2 number of eyes.

On the other tentacle, wouldn't delegates be elected based on popularity? --Kim Bruning (talk) 23:42, 11 February 2008 (UTC)

Can you please explain what you mean by those two sentences? It's not altogether clear to me. 71.63.91.68 (talk) 01:21, 12 February 2008 (UTC)

Past a certain number of people, discussion of actions becomes fairly random, and you can't actually reach a decision.

So from that point of view, it might be a good idea to have delegates.

On the other hand, the problem with having delegates is that eventually people might decide to "elect" delegates based purely on popularity, as opposed to how sane their ideas are.

Does that make a bit more sense? Else I can expand more if you like.--Kim Bruning (talk) 02:05, 12 February 2008 (UTC)

That already happens in representative systems. People vote for candidates based on their hair, clothes, mannerisms, etc. See http://www.jibjab.com/view/221737 . However, it was recognized long ago that despite this drawback, it was still better to elect decisionmakers rather than try to make all the decisions ourselves. Delegable proxy, in general, could be regarded a middle ground between the two systems, although on Wikipedia it would probably just be an enhancement of our current system, providing a bit more information to closing admins as opposed to being binding. Sarsaparilla (talk) 17:16, 12 February 2008 (UTC)

It has been stated that with modern information systems, electing decisionmakers need no longer be a necessity. --Kim Bruning (talk) 17:24, 12 February 2008 (UTC)

Who stated that? Given time restraints, it's inevitable that once the decisionmaking workload reaches a certain point, we will be delegating some of it, or else abstaining or basing decisions on cursory examination. However, we need not pool our votes with others to choose someone to represent both the majority and the dissidents. Each person can choose their own representative in a proxy system. In that sense, elections can be dispensed with. I would consider voting based on some sort of advisory system (e.g. a proxy firm) to be, in some cases, almost equivalent to delegating the decisionmaking. It's like how supreme court justices delegate a lot of decisions to their clerks. Sarsaparilla (talk) 17:49, 12 February 2008 (UTC)

Fair enough. You're redistributing workload. That makes sense.

Originally, the wiki (serendipitously) distributed workload, and kept it fairly low for each participant, as only a small number of editors participate on each page. Some of the systems developed early on (such as articles for deletion) accidentally subverted the wiki-model, and thus do not scale so well. Would you be able to consider wiki-model based distribution of workload? --Kim Bruning (talk) 22:05, 12 February 2008 (UTC)

Ways to game the system

Off the top of my head, I can think of several ways for this system to be gamed.

Consider:

1. Alice starts editing Wikipedia in January.
2. Alice delegates her votes to Bob, a sockpuppet Alice creates in February.
3. Alice stops editing as Alice, and continues to edit as Bob (who now votes with twice the weight).
4. Lather, rinse, repeat.

In a year, Alice/Bob/Carol/Dave/etc. have a dozen votes via proxy.

Consider:

1. Alice, Bob, Carol, and Dave join Wikipedia because they've heard about the YugiMonCruftWars deletion debate on CruftWarsForum.
2. Instead of showing up as obvious single-purpose accounts at AfD, A, B, C, and D delegate their proxies to long-time editor Xerxes. A, B, C, and D may make a few edits to bolster their credentials.
3. Xerxes votes as their proxy.
4. A, B, C, and D leave Wikipedia shortly thereafter.
5. Xerxes continues to vote as their proxy.

A savvy editor could easily accumulate proxy votes in this way over multiple AfDs. A particularly savvy editor could deliberately post calls to arms on external forums.

Consider:

1. Wally decides that proxies are a silly notion, and offers to cast votes to counter any proxy votes on an issue.
2. Alice, Bob, Carol, and Dave agree, and give Wally their proxies.
3. Proxies cancel out at whatever 'vote' is taking place.
4. Admin closing the discussion is back to square one—making a close based on the weight of arguments, rather than weight of numbers.

Much more seriously pathological outcomes are possible if you start having proxy wars between a group of (say) absolute inclusionists and absolute deletionists.

No doubt a moderately determined individual could find numerous other ways to subtly (or not-so-subtly) manipulate and undermine proxies. God help us all if anyone even things of taking them seriously for the purposes of RfA. TenOfAllTrades(talk) 16:47, 11 February 2008 (UTC)

Basic principle of Robert's Rules: don't debate a thing unless it has been specifically proposed and seconded in that exact form. At this point, what normally happens is that people imagine what the proposal is, and argue about it, so they are arguing against their own imaginations. Brainstorming turns this around. Suppose what has been said before has a great idea behind it, but details are wrong? Is there a way to imagine this so that it will work?

I say there is. But I can't write now, gotta go.

So, for the moment, I suggest, instead of thinking about what obviously won't work, think about what might work. That's what I did, twenty years ago. You just might independently invent it! Hint: take the distributed decision-making process of Wikipedia, and add a proxy table that users can edit to show their proxy, and the proxy can edit to show acceptance. *Nothing* said about how this is used. Now, how could it be used? There are obvious problems, but even by thinking about those problems, and being explicit about them, and especially if you think about how those problems could be solved -- or even bypassed to make them moot -- you are way ahead of most people. Most people are simply convinced that nothing can solve the problem of scale in democracy, so there is no use even thinking about it.

And, of course, this is why the problem is not solved! So how to deal with this situation. What if we do figure out how this might work, how do we get from here to there? I'll tell you my answer. We are already there. Nothing is stopping us from doing it except our lack of understanding of it, plus, for those few who do understand, a persistent belief that it is impossible to change the way people think. It's quite possible, one just has to know how to do it -- and the changes must be fueled by human nature, for we are naturally resistant to most change, for very good reasons. Essentially, the change agent is merely a catalyst, the energy comes from the people themselves.

So, what in the world am I talking about? Believe me, I do know what I'm talking about, but my experience is that if I try to explain it all at once, and in particular to people who have not considered the underlying problems and who have not struggled a bit with solutions, I'll be wasting a lot of keystrokes. Been there, done that, over and over again. I accomplished what needed to be done at that stage; this is now the second stage, where it must get easier or it won't work. I'll be watching and writing as I have time. I'm really not trying to be mysterious; search for "abdlomax delegable proxy" on google and you'll find a lot. "FA/DP" might return some good hits as well. The ideas have other names, but ... enough for now. --Abd (talk) 18:37, 12 February 2008 (UTC)

Why do we need this?

We already use delegable proxy. The users who actually participate in any discussion are already fully empowered to represent the community and every individual in the community. They are already delegated the community's proxy. Similarly, every admin has the community's proxy. Not participating in a discussion acts as a grant of proxy power to the participants, giving them authority to represent one, as full as any formal document. Why are any further formalities needed? Best, --Shirahadasha (talk) 07:09, 13 February 2008 (UTC)

Indeed! So we could change the contents under this title to explain that? :-) --Kim Bruning (talk) 16:52, 13 February 2008 (UTC)

Well, let me try an example. Suppose there are 1,000 users, consisting of 500 inclusionists and 500 deletionists. The inclusionists spend most of their time writing articles, while the deletionists spend most of their time doing maintenance work such as checking Recent Changes, perusing the AfDs, etc. By the nature of the kind of work they gravitate toward, there will be some participation bias in the deletion debates. Application of deletion policy will tend to lean in the deletionist direction.

Suppose we have delegable proxy. Inclusionists could organize to make their voices heard without having to spend too much time in the AfDs. Three of them could form a proxy loop, with A appointing B, B appointing C, and C appointing A. In the event that only one of them showed up at a deletion debate, that person could speak on behalf of them all. Or to be more precise, the closing admin could see that the member of that group is at the head of a proxy chain of three people. There are ways to automate this, similarly to how we automate RfA analysis. It would just be an additional tool for figuring out, what is the real level of consensus here?

We can say all day that deletion debates are not a vote. Does anyone really think that a debate with 20 deletes and 2 keeps is going to be closed as keep on the basis of the keep arguments being particularly cogent? It might happen in the event of a blatant factual error on the part of the 20, but it's unlikely to happen in borderline cases dealing with notability or other semi-subjective deletion criteria. And notability is a big issue of contention on Wikipedia, with the borderline cases constantly redefining the standards of what we can and can't keep here.

Delegable proxy helps level the playing field between those who invest large amounts of time in certain debates and those who don't. Arguably, it is the fault of those who don't participate directly in those debates that they don't influence the result. But it would be undesirable for everyone to spend a lot of time at AfD, MfD, policy debates, etc. Some people have different strengths and could be more productively involved elsewhere. Nonetheless, their views should still be represented, in a way that accurately reflects the number of people favoring a certain viewpoint (in this case, by editors choosing proxies who generally share their wikiphilosophies).

Of course, if someone sees their proxy totally blowing a lot of decisions, they can pick a different proxy at any time, and that serves as a safeguard. By contrast, if I see the deletionists totally blowing a lot of decisions, I can intervene by getting involved in more deletion debates, but do I really want to spend a lot of time on that if my strength is crystalline geometry article creation, or bot coding, or some other skill shared by a smaller subset of users? That would be like a nuclear physicist who owns 10 shares of IBM stock being told, "If you want your interests to be represented, come to the shareholder meetings." That would be an effective way to make his voice heard, but is that the best use of his time? It would make more sense for him to select a proxy (or have his investment manager select one), and go on his merry way, unlocking the secrets of the atom.

As my economics teacher used to say, "We do what we do best and exchange for the rest." 71.63.91.68 (talk) 02:46, 14 February 2008 (UTC)

"Rejected" tag

Isn't it a bit early to place the "rejected" tag on this article? This is a complex subject, discussion is still in early stages, and in my opinion we shouldn't discourage it by saying the decision has been made. As Abd remarked in reference to the disadvantages of basing decisions on early votes: "Some very good ideas seem to be flawed at first glance. So they could get a lot of negative votes. And then they become relatively invisible. The few who read them closely enough to understand them and see the value can’t overcome the inertia of the rest." 71.63.91.68 (talk) 02:46, 14 February 2008 (UTC)

Per Wikipedia:Policies_and_guidelines, "A rejected page is any proposal for which consensus for acceptance is not present after a reasonable time period." I still think the tag is premature, but the solution seems to be a rewrite of the proposal.

Jimbo's comments

They are available here: See User_talk:Jimbo_Wales/Archive_32#all_sides_can_agree_in_principle_to_an_orderly_process_of_making_a_determination_of_what_to_do 71.63.91.68 (talk) 04:16, 14 February 2008 (UTC)

excellent idea

This is an excellent idea. I have been pushing for a an English language wikipedia house of representatives for some time now, but those with the power seem to prefer not to promote democracy . We can form groups, associating, etc. Our right to associate can not be "rejected". However, the valuation of a !vote as standing for more than that !voter is something we must let develop naturally and not push for automatic acceptance of it. There are real problems of people having multiple accounts, so no system of automatically adding up accounts will work. This needs to be allowed to grow and find acceptance through being useful and believable; not being forced down anyone's throat. WAS 4.250 (talk) 04:21, 14 February 2008 (UTC)

Yes, I think the proposal is in need of a rewrite to reflect the clarifications Abd made above; and also some examples of how it would be helpful might be in order. What do you think of the deletionist/inclusionist example above? Could an example be come up with that wouldn't be biased against a particular factions of Wikipedians? In reference to not forcing it down people's throats – maybe the article should avoid, then, the language about giving people the "authority" to speak on one's behalf, and simply describe it in terms of providing information that can be useful to closing admins. If the idea catches on, modifications can evolve later. 71.63.91.68 (talk) 04:36, 14 February 2008 (UTC)

What I was, as far as I know, the first to propose was a combination of delegable proxy with what I call Free Association traditions, taking from the operating traditions of Alcoholics Anonymous, which were, in my view, the most important factor in the massive success of AA. Many people in AA have thought, over the more than sixty years since it was founded, that there was something in those traditions for the world, but, as far as I know, nobody figured out how it might be done.

The key is in separating the function of intelligence (advice, recommendation, knowledge) from control (the executive, control over individual actions and concentrated power -- property, specifically). An FA cannot, by tradition, own more property than necessary for its basic functions, which, by design, are very light. AA meetings are supported by very small contributions tossed in the basket at meetings, and they are used to pay rent and buy publications, typically for free distribution or by donation. The contributions are entirely voluntary, there is no disapproval of those who don't contribute. But almost nobody fails to toss something in, and it is almost always enough. Meetings don't accept free meeting space, generally, and they don't accept large donations. If they accumulate funds not needed for a few month's worth of expenses, they will generally pass it on to intergroup for intergroup expenses and intergroup will pass on what they don't need to the national office. Quite simply, they don't have financial problems, ever.

Central decisions in AA are either about the office operation, those decisions being made entirely by AAWS, Inc, as members don't really care about that, or about publication content or recommendations back to meetings over policy issues. There is a delegate conference, each delegate elected either by supermajority from areas (2/3 vote necessary) or, if, after so many ballots, a 2/3 majority cannnot be obtained, by lot fro the top two candidates. The theory of this is that it does provide for some minority representation. Due to the nature of the organization and the kinds of decisions that need to be made, this works very well.

But if we really want to accurately and efficiently measure consensus, we need something more efficient and more representative. That's delegable proxy.

So the community uses DP to estimate consensus. These estimates are made by anyone, using a proxy table, and additional analysis as determined appropriate by the one doing the analysis, most notably by someone who needs to make a decision, such as an administrator closing some decision-making process.

We don't have DP in place (all it takes, though, is a proxy table, and if not many members assign proxies, no harm, it simply isn't measuring as broad a community), but other forms of vote analysis can be done.

With my current RfA, almost certainly likely to close in "failure," I plan to take the results and anaylze (and report) them in various ways. For example for every voter, I could take the edit count and weight the votes according to that. I could look at only administrator votes, or only non-admin votes, or time since registration. Each of these analyses would generate some different perspective on what happened. Using this for *decision*, through some fixed rule, would be impossibly chaotic (though it's possible that after substantial history of use, standards could develop that could make it possible -- but then it becomes vulnerable to new forms of manipulation, long-term sleeper sock puppets, etc.), but as "advice," there is no problem.

The responsibility for the decision remains quite the same as it is now. I'm not doing the vote analysis to try to claim that the result should be different, as Sarsaparilla knows, I'm quite unattached to the outcome, and starting some dispute would be the last thing I'd want. I'm doing it just to see how it looks. In particular, edit count weighting, I expect will be interesting. But voluntarily chosen proxy weighting, considering proxies delegable, would be the most interesting of all, by itself, but also very interesting if combined with edit count. (I.e., each individual vote, whether direct or by proxy, would be weighted by an edit count).

(The RfA is running close to 50-50 support-oppose, after the sock-puppet-solicited votes are subtracted. I'm being accused of thinking that it's supposed to be "majority vote," which is preposterous, I know the guidelines and traditions, but I expressed satisfaction at the support being that high, considering my low edit count and my rather pugnacious history (as it would be seen, all too easily). But that's the kind of stuff I've faced over the few months I've been active here, and I'm utterly unsurprised. Little do they know that I'm smuggling donkeys. I'll tell the story if asked, or someone else can if they know it. gotta get to bed.)

Now, as to a "legislature," the question is whether or not it should be delegable proxy or Asset Voting. Asset Voting was invented, turns out, by Lewis Carroll in the 19th century.... Asset is used to create peer assemblies, where every member has the same voting power. It can be combined with DP to make it more flexible, and it is entirely possible to combine it with direct democracy. The problem with direct democracy is that if the people actually participate on a large scale, the noise makes it impossible to function. So what Asset (or DP) would do is to create a representative class that represents the entire group in deliberation. Voting remains open to everyone, but ... whoever doesn't directly vote, if they have a representative, the representative's vote counts for them. In Asset, it's possible that someone chooses a representative, and the representative they choose can't get a seat in the assmbly, due to lack of available compromises or intransigence of those holding the votes (but not enough votes individually to gain a seat). With Delegable Proxy, because of the variable voting power, the proportionality of representation becomes almost perfect. These things have been worked on for years, there are a lot of details that could be explained or proposed.

Summary: DP creates, all by itself, a representative hierarchy. That hierarchy can, if it is decided to do so, elect a peer assembly, or a proxy assembly; both of them can be almost fully representative. In a peer assembly, every member may address the assembly. In a DP assembly, the assembly sets rules for floor access, which might be representing a certain number of members. Beause of the variability, and with a fixed assembly size (chosen for efficiency in finding consensus), more people can be represented by members with a "seat," i.e., the right to address, with DP rather than Asset.

Yes, this is another new discovery: the possibility of separating, for an assembly, the right to vote and the right to participate directly in deliberation. The problem of scale in democracy has only involved, really, the last problem, it would always have been possible to have substantial votes be open to every elector. Wikipedia process is completely open, but mixes deliberation and voting; really, the two should be separated. Votes are votes, and evidence and arguments are evidence and arguments. ArbComm functions that way. (Though small comments are often made with votes). --Abd (talk) 05:28, 14 February 2008 (UTC)

By the way, our right of association can be rejected, or at least the "community" can try. See Esperanza and WP:AMA. In both cases the right of members to freely associate was rather violently rejected by "consensus." Not only non-democratic, but also a dicatorship of the "majority." Repressive,with little expressed justification that made sense. So ... the proxy table can be here on Wikipedia, initially, but it should also be copied elsewhere. And the basic rule that I would make for the proxy table is that naming a proxy there is granting consent for the proxy to contact the client, and accepting is granting consent for the client to directly contact the proxy, with both contacts being, preferably, off-wiki (at least they should know each other's email addresses and preferably phone numbers as well). This is one measure that would be taken, among other things, to dampen enthusiasm for collecting large numbers of proxies (do you really want to give 10,000 people your phone number?). This would, then, make the system robust, unable to be disrupted by deletion of the tools that structure it. The key to DP is the communications structure itself, not vote-counting, which is merely one application. --Abd (talk) 05:37, 14 February 2008 (UTC)

Rejecting very firmly

I would like to take the floor and thank the honorable gentleperson Abd for providing a long list of reasons to reject this concept, to draw and quarter it, burn it at the stake, and finally bury it. --Kim Bruning (talk) 11:24, 14 February 2008 (UTC)