Iterated giving is a way to quantify the value contributed by participants in a project according to peer review.
In other words, it is a trust metric method for dynamically determining payments to participants in a joint venture. It can also be used to periodically (re)distribute other things, such as ownership/control. The system can be used in open groups, that is, groups that anyone can join. The method scales reasonably well to large groups. The method is resistant to attack. Optionally, members can be rewarded for staying with the project a long time.
There is a group of people; maybe anyone can join this group. A group within this group is called "the seed group". Maybe just one person makes up "the seed group"; maybe everyone in the whole group is also in the seed group; or maybe some people are in the seed group and some are not.
Each day, the (whole) group earns a pool of dollars which must be distributed.
At breakfast, each person in the seed group is given an equal amount of an imaginary currency (not dollars), called "influence". During the morning, each person must give away all of the influence that they had at the end of breakfast to other people; they can give it away however they want. At lunchtime, the pool of dollars is divided up and temporarily given to people in proportion to how much influence they received. After lunch, each person must give away all of the dollars that they got at lunchtime; they can give them away however they want. At the end of the day, people get to take home any influence that they were given after breakfast, and any dollars that they were given after lunch. A record is kept of who gave what to who, and made public.
Tomorrow, many people, not just those in the seed group, may start out with some influence. They will have to give it away tomorrow morning. People in the seed group may start out with some influence, and also get more at tomorrow's breakfast; they will have to give away all of the influence from both of these sources.
The idea is to run a simulation of the procedure that I just described. Each person submits two lists of ratings; first, a list that says what percentage of the influence they would like each person (except themself) to get; second, a list that says what percentage of the money they would like each person (except themself) to get. A person can login to the system and change their ratings from time to time. All ratings are public.
Each day, a computer simulates the described procedure, using the two ratings lists as a guide to tell it how each person gives away their influence and money. The computer keeps track of how much influence a person has; the amount of money that a person "takes home" each night is actually paid out to the person.
The result of the procedure is different from just taking the average over every person's lists. Over time, a person tends to get more influence if they are highly rated by people who are themselves highly rated by others.
The procedure has some resistance to attacks. If a bunch of freeloaders who are not in the seed group sign up and rate each other highly, but don't get rated by anyone in the seed group (or anyone who themselves was rated by the seed group, or anyone who was rated by someone who was rated by the seed group, etc), then they do not get any money, even if the freeloaders form the vast majority of participants.
Even if the freeloaders manage to get non-zero ratings from others who have influence, the total amount of money per day that is given to all of the freeloaders combined is bounded by the sum of the ratings that people with influence gave the freeloaders, times the influence of the raters, times the number of days that the freeloaders had those ratings. When those with influence catch on and cut off the freeloaders (either by giving them zeros, or, if you did not rate the freeloaders themselves but rated someone who did, by giving THEM zeros in their influence ratings, or, ... etc), then after that point, the amount of money that the freeloaders can get per day decreases geometrically.
What if a malicious clique of people in the seed group try to monopolize influence? A clique could decide to only give influence and money to other people in the clique. In this way, any influence that comes into the clique never leaves. In this case, people outside of the clique can "cut them off" once they notice (by giving them zero influence ratings), just as with the freeloaders. Just as with the freeloaders, there is a bound on how much damage is done, and after the clique is cut off, the amount of money that the clique gets per day from the scheme decays geometrically. The only difference is that the people in the seed group can't be "cut off" from the money that they were given at breakfast; so they can always ensure that they get almost the same proportion of money as the fraction of the seed group that they represent (for example, if one third of the seed group is in the clique, then the clique can guarantee that it gets about one third of each day's money).
One presentation detail is that instead of actually creating more "influence currency" each day and giving it to the seed group (which would have the implicit effect of currency inflation; the value of all of the existing influence would be reduced), the amount of influence currency is fixed, and each day some currency is taken away from everyone, and then redistributed to the seed group. The results given by the system are identical whether this influence is created each day, or just recycled.
What is the meaning, or role, of influence? Influence means how much influence a person's ratings have. Each person gives two types of ratings; influence ratings and money ratings. The weight of both of these types of ratings within the system is determined by the influence of the rater.
Why do you give each person separate ratings for influence and for money? An example of someone who would get lots of influence but not much money would be someone who is thought to not contribute very much themself, but to be an excellent judge of how much others contribute. Conversely, someone might be thought to be a valuable contributor to the project, yet not very good at giving accurate ratings to other people, in which case they would get lots of money but not much influence.
Raising someone's rating can never hurt them (unless other people lower their ratings because they saw that you raised yours; but I just meant that raising your ratings won't cause the system to injure the person, I'm not saying what other people will or won't choose to do). It's better to get a small rating from someone than a zero rating.
Things add linearly in this system. For example, receiving two ratings (of either type) of .25 are just as good as receiving one rating of .5 (assuming that all the raters were of equal influence). Receiving a rating of .25 from a person with an influence of .6 is just as good as receiving a rating of .5 from a person with an influence of .3. If you receive one rating of .1 from a person with an influence of .2, and also one rating of .3 from a person with an influence of .4, then that's just as good as if you had only received one rating of .2 from a person with an influence of .7 (because .1*.2 + .3*.4 = .2*.7).
How should you decide what money ratings to give people? The higher money ratings you give people, the more money they will get. But ratings are percentages; if you raise one person's money rating, then you must lower another person's, because the list of all of your money ratings must sum to 100%. Bear in mind that ratings are public.
How should you decide what influence ratings to give people? Raising a person's influence means that their opinion of other people is given more weight. Therefore, you should give higher influence ratings to people that you think will do a good job of rating other people. A well-meaning but naive person who gives high ratings to people who don't deserve them should not be given a high influence rating; but neither should a hothead who witholds high ratings from people for bad reasons. Like money ratings, influence ratings are percentages; in order to give one person more influence, you must give another person less.
Influence is separate from money; if you give someone low influence ratings, they won't get paid any less (except to the extent that they cause their influence to flow to people who in turn give them high money ratings).
Each day, in addition to reporting current influence, the software will make a projection that tells you, if no one ever changed any of their ratings ever again, what people's influence would eventually become, and how much money they would eventually make per day. Try to consult this periodically and see if you agree with the outcome. If someone or some group has much more or much less influence than you would like, or if they are paid much more or much less than you would like, then look at who gave that person the low or high ratings that you disagree with, and lower your influence rating for the ratings-giver(s) whom you disagree with.
The software will also tell you which subgroups seem to exist within the total group, in terms of which groups tend to attract influence and then circulate it within themselves rather than giving it back to people outside the subgroup. This might be perfectly innocent, or it might be malicious. If you don't want those subgroups getting so much influence, then decrease your influence ratings for them and for people who rate them highly.
Your group will have to decide for itself whether it is to be considered unethical for a group of people to scheme to rate each other highly, and whether it is to be considered unethical to trade ratings for favors (or even to use it as currency exchanged in return for other people to do something that you need in order to do your job, similar to authorship on academic papers). My best guess is that the system will work best if these sorts of things are openly allowed, at least to some extent; but I'm not really sure, so feel free to ban them if that's what your group wants to do. But either way, I recommend having an explicit policy on these things. It would be too bad if, due to the lack of a group-wide discussion, different departments within the group evolved different cultural norms regarding these things, and if that led to one department thinking that the people in a different department were unethical.
This is the end of the "overview" section. Users of the system can stop reading here. If you are interested in reading more details, analysis, and discussion about the system, see http://bayleshanks.com/wiki.pl?ideas-trustMetrics-iteratedGiving.