Crowd wisdom and social networks
“Wisdom of the crowd” (“MT”) is one of the most important concepts of the web today, perhaps the most important for social media, but it is also one of the least understandable phenomena. When James Shurovyeski wrote the book “The Wisdom of the Crowd” in 2004, he explored the stock market and other classic examples of social psychology, but the “Web 2.0” environment was just beginning to take shape. The time has come to find an application to his ideas on social networks, where they can fully reveal themselves.
The theory of MT does not mean that people are getting smarter in groups, not at all. Anyone who has seen an angry mob will confirm this. But the team, in which there are the correct formulation of the problem and the correct internal communication, can become wiser. Under these conditions, a group of people is wiser than any of its only participants.
The standard example (in Habré 3 similar tests 1 , 2 and 3 were already conducted last year): suppose that you have a box with coins. Ask a few hundred people how many coins are inside. When you summarize the survey, it is possible that all assumptions will be wrong. But if you find the arithmetic mean of all the answers, the result will be close enough to the correct one.
')
The network, with its low barrier to entry and blurred social boundaries, is a unique medium for exploring the subtleties of the collective mind. There are many examples on the web: google search results, torrents, habratopics on the main page (there is another example in the article). In each example, the result is the average opinion of a large group of people.
You only need to know a few things for the collective mind to start working.
Simplicity
In the example with the box of coins, you ask each participant a number. Google does not ask anyone to generate search results, it uses the importance parameter of each page in its index, considering how often it is referred to.
Note that in each of the above cases: with google, torrents, habratopics on the main and a box of coins, input is not discussed. Interactive input is too complex for MT systems. Online discussion systems do not lead to wisdom by themselves.
The simplicity of each individual task is also important. Collective intelligence systems can be busy with fairly complex projects, but each project must first be decomposed into its simplest components.
Interface
A complex interface can be ideal for complex tasks, but not for MT systems. The more complex the interface, the less people will be involved and the more confusing it will be. Participation in collective intelligence systems should not be time consuming. No need to force members of the group to think about something for too long.
The MT interfaces are most often some kind of “like / dislike” voting mechanisms, but it can also be, for example, choosing a point on a map or drawing a shape.
Data collection
One of the reasons why discussions often do not lead to any results is that there is no information gathering, but only a conversation. But collective intelligence systems must produce results. This requires a drive and algorithm.
In the example with the box of coins, the drive is a person who summarizes all the guesses, the algorithm is an arithmetic average. In the example with Google search results, Google is the storage engine, and the algorithm, known as PageRank, is its constantly developing, well-protected secret.
Members of the group
A characteristic feature of the MT system is that the more participants it has, the better the result will be. In forums and chat rooms discussion falls apart when there are too many people present. If your community gets worse with new members, you cannot call it a collective wisdom system.
Also the group should be heterogeneous. The collective mind works because people with opposing views balance each other. If there are too many participants representing a particular group or society, the results may suffer.
Selfishness
Contrary to possible assumptions, the most loyal collective decisions are made in groups made up of people who think about their own needs, and not about the needs of the group. In the stock market, participants are all motivated to buy cheaper and sell more expensive. The value of shares reflects (imperfectly, but fairly effectively) the judgment of investors about the value of companies in the market. All market participants think about their own benefit, not about a single company or the entire market, but this is what works.
Similarly, the creators of the site may not consciously participate in the voting in their project, but the collective decisions of visitors will still lead to good results. The mechanism of functioning of Google is also similar to the stock market. There may be some manipulation of collective opinion (for example, the creation of reference farms), but they are trying to counteract them, since in both cases there is an interest only in honest results.
Egoism also struggles with a bigger problem - group thinking, when each member of a collective puts the interests of the collective above their own. As soon as the collective opinion begins to prevail, the group is endangered. The collapse of the stock market, the fall of Challenger and many other examples may arise from group decisions.
In the example with the box of coins, participants were told that the correct correct answer would win all the money. Their participation was completely selfish - they wanted to win money. The fact that their responses were averaged to get a result was only a good side effect.
But for the manifestation of the egoism of people, it is not necessary to pay cash. For example, there are a lot of news sites that feature the most frequently mailed articles. No one pays people to choose their favorite stories, newsletter information is easily generated on the server. The key factor is user motivation to participate. People participate for their own reasons. They are deliberately not allowed to vote for news, voting is just a by-product.
This method can be contrasted with the method of voting on a site like digg (or habr), where users obviously vote for the most relevant article. On such sites there is a constant "war" against people trying to play against the system. There are many companies that promise to post articles on the main page of digg.
Counting system
Saving scores (points, points, etc.) is part of any game, and any collective site can also be considered as a game. So you need to very carefully consider a system for assessing the participation of each member of the group.
For example, on the site slashdot, the “karma” of each participant was calculated depending on the degree of their participation. The karma score influenced some of the features (such as the ability to moderate other users' comments and the number of default comments available).
And everything seemed to be great. But they revealed the indicator of "karma" to users. At that moment, when the creators of the site did this, they, without knowing it, invented the “karma-drocherov” (interestingly, we translated this expression as “whores of fate”) - users who sent comments, knowing that they would be well appreciated by the community. Thus a unique type of group thinking was created.
The games are great as long as the players benefit the site itself. Now slashdot is dominated by either group thinking or trolling. Both phenomena probably existed before the disclosure of karma, but it certainly contributed to their rapid development.
Majority opinion
The majority opinion is a problem for collective intelligence systems. On the one hand, a well-tuned MT system allows you to see the opinion of most people. User reviews are collected, the algorithm sums up, and the result - here it is, in front of us in the form of a list of items in order from the best to the worst (such as, for example, google search results).
But there is one unpleasant moment: disclosing the estimated list to the community strengthens group thinking. Well-rated articles are rated even better, low-rated ones immediately disappear from view. Displaying a list destroys the accuracy of collective wisdom.
So what can be done? Here are a few suggestions:
Explicit and implicit feedback
In many of the examples cited, collective wisdom was shaped according to people's behavior. In these cases, feedback was implicit. In other examples, users were asked to directly rate something like a voting system. This was a clear feedback. The use of explicit or implicit feedback, or some combination thereof, is an important decision in the design of any MT system.
In working with your collective intelligence system, pay more attention to the possibility of using implicit feedback when you do not have to directly ask people for their opinions. Implicit feedback is usually more honest and less manipulative. There are also ways to combine both types of communication, for example, by applying a vote, but taking into account when summarizing and other data (such as page views, comments or other user actions).
Voting
This will seem undemocratic, but only one vote should not be trusted to manage all processes in collective wisdom systems. In many cases, it should not do that.
You should not declare the undisputed winner the news that collected the most votes. With the same ease it is possible to draw conclusions about controversial news (a high percentage of both good and bad votes) or unrevealed (low number of votes).
It must be remembered that all user voices do not have to be equal. Voices from “good” participants (in your understanding of the word “good”) can have a higher impact.
Studies show that, by evaluating a number of points, users are more likely to vote negatively in bursts. In other words, as soon as a person begins to vote negatively, he is more likely to continue to do so further. Thus, it would be fair to give a negative vote less weight if it happens after a negative vote of the same user. Or, you can save the duration of a voting session as a variable, and estimate after a while your early votes in comparison with the later ones during this period of time. If someone votes for an hour, does that make the voice more valuable or less? You can conduct an experiment to find out.
One head is good ...
These aspects of the theory of MT - only the basics. Much more detail these and many other concepts are disclosed in the book Shuroveski. The most important thing is to remember that collective intelligence systems must constantly evolve. If done correctly, they can change our online life, and perhaps make us all a little wiser.
The theory of MT does not mean that people are getting smarter in groups, not at all. Anyone who has seen an angry mob will confirm this. But the team, in which there are the correct formulation of the problem and the correct internal communication, can become wiser. Under these conditions, a group of people is wiser than any of its only participants.
The standard example (in Habré 3 similar tests 1 , 2 and 3 were already conducted last year): suppose that you have a box with coins. Ask a few hundred people how many coins are inside. When you summarize the survey, it is possible that all assumptions will be wrong. But if you find the arithmetic mean of all the answers, the result will be close enough to the correct one.
')
The network, with its low barrier to entry and blurred social boundaries, is a unique medium for exploring the subtleties of the collective mind. There are many examples on the web: google search results, torrents, habratopics on the main page (there is another example in the article). In each example, the result is the average opinion of a large group of people.
You only need to know a few things for the collective mind to start working.
Simplicity
In the example with the box of coins, you ask each participant a number. Google does not ask anyone to generate search results, it uses the importance parameter of each page in its index, considering how often it is referred to.
Note that in each of the above cases: with google, torrents, habratopics on the main and a box of coins, input is not discussed. Interactive input is too complex for MT systems. Online discussion systems do not lead to wisdom by themselves.
The simplicity of each individual task is also important. Collective intelligence systems can be busy with fairly complex projects, but each project must first be decomposed into its simplest components.
Interface
A complex interface can be ideal for complex tasks, but not for MT systems. The more complex the interface, the less people will be involved and the more confusing it will be. Participation in collective intelligence systems should not be time consuming. No need to force members of the group to think about something for too long.
The MT interfaces are most often some kind of “like / dislike” voting mechanisms, but it can also be, for example, choosing a point on a map or drawing a shape.
Data collection
One of the reasons why discussions often do not lead to any results is that there is no information gathering, but only a conversation. But collective intelligence systems must produce results. This requires a drive and algorithm.
In the example with the box of coins, the drive is a person who summarizes all the guesses, the algorithm is an arithmetic average. In the example with Google search results, Google is the storage engine, and the algorithm, known as PageRank, is its constantly developing, well-protected secret.
Members of the group
A characteristic feature of the MT system is that the more participants it has, the better the result will be. In forums and chat rooms discussion falls apart when there are too many people present. If your community gets worse with new members, you cannot call it a collective wisdom system.
Also the group should be heterogeneous. The collective mind works because people with opposing views balance each other. If there are too many participants representing a particular group or society, the results may suffer.
Selfishness
Contrary to possible assumptions, the most loyal collective decisions are made in groups made up of people who think about their own needs, and not about the needs of the group. In the stock market, participants are all motivated to buy cheaper and sell more expensive. The value of shares reflects (imperfectly, but fairly effectively) the judgment of investors about the value of companies in the market. All market participants think about their own benefit, not about a single company or the entire market, but this is what works.
Similarly, the creators of the site may not consciously participate in the voting in their project, but the collective decisions of visitors will still lead to good results. The mechanism of functioning of Google is also similar to the stock market. There may be some manipulation of collective opinion (for example, the creation of reference farms), but they are trying to counteract them, since in both cases there is an interest only in honest results.
Egoism also struggles with a bigger problem - group thinking, when each member of a collective puts the interests of the collective above their own. As soon as the collective opinion begins to prevail, the group is endangered. The collapse of the stock market, the fall of Challenger and many other examples may arise from group decisions.
In the example with the box of coins, participants were told that the correct correct answer would win all the money. Their participation was completely selfish - they wanted to win money. The fact that their responses were averaged to get a result was only a good side effect.
But for the manifestation of the egoism of people, it is not necessary to pay cash. For example, there are a lot of news sites that feature the most frequently mailed articles. No one pays people to choose their favorite stories, newsletter information is easily generated on the server. The key factor is user motivation to participate. People participate for their own reasons. They are deliberately not allowed to vote for news, voting is just a by-product.
This method can be contrasted with the method of voting on a site like digg (or habr), where users obviously vote for the most relevant article. On such sites there is a constant "war" against people trying to play against the system. There are many companies that promise to post articles on the main page of digg.
Counting system
Saving scores (points, points, etc.) is part of any game, and any collective site can also be considered as a game. So you need to very carefully consider a system for assessing the participation of each member of the group.
For example, on the site slashdot, the “karma” of each participant was calculated depending on the degree of their participation. The karma score influenced some of the features (such as the ability to moderate other users' comments and the number of default comments available).
And everything seemed to be great. But they revealed the indicator of "karma" to users. At that moment, when the creators of the site did this, they, without knowing it, invented the “karma-drocherov” (interestingly, we translated this expression as “whores of fate”) - users who sent comments, knowing that they would be well appreciated by the community. Thus a unique type of group thinking was created.
The games are great as long as the players benefit the site itself. Now slashdot is dominated by either group thinking or trolling. Both phenomena probably existed before the disclosure of karma, but it certainly contributed to their rapid development.
Majority opinion
The majority opinion is a problem for collective intelligence systems. On the one hand, a well-tuned MT system allows you to see the opinion of most people. User reviews are collected, the algorithm sums up, and the result - here it is, in front of us in the form of a list of items in order from the best to the worst (such as, for example, google search results).
But there is one unpleasant moment: disclosing the estimated list to the community strengthens group thinking. Well-rated articles are rated even better, low-rated ones immediately disappear from view. Displaying a list destroys the accuracy of collective wisdom.
So what can be done? Here are a few suggestions:
- The use of temporary variables. You can allow to vote only a certain period of time. When the time is up, the voting ends and the results are shown.
- Mixing display results. Instead of displaying an ordered list of values (or most opinions), you can display a list of highly rated items, but in random order. It uses flickr in its “fun” show.
- Make users earn disclosure ratings. You can show community ratings only after a vote has been cast. This is used in many survey systems to avoid the impact of current results on the choice of the voter.
- The use of algorithms. When an estimated list is displayed, complex formulas are used, taking into account a variety of data, in addition to voices. So search engines do when generating search results. An algorithm that is constantly subject to all sorts of optimizations is a Google tool in a constant struggle with people trying to uncover the PageRank counting system.
Explicit and implicit feedback
In many of the examples cited, collective wisdom was shaped according to people's behavior. In these cases, feedback was implicit. In other examples, users were asked to directly rate something like a voting system. This was a clear feedback. The use of explicit or implicit feedback, or some combination thereof, is an important decision in the design of any MT system.
In working with your collective intelligence system, pay more attention to the possibility of using implicit feedback when you do not have to directly ask people for their opinions. Implicit feedback is usually more honest and less manipulative. There are also ways to combine both types of communication, for example, by applying a vote, but taking into account when summarizing and other data (such as page views, comments or other user actions).
Voting
This will seem undemocratic, but only one vote should not be trusted to manage all processes in collective wisdom systems. In many cases, it should not do that.
You should not declare the undisputed winner the news that collected the most votes. With the same ease it is possible to draw conclusions about controversial news (a high percentage of both good and bad votes) or unrevealed (low number of votes).
It must be remembered that all user voices do not have to be equal. Voices from “good” participants (in your understanding of the word “good”) can have a higher impact.
Studies show that, by evaluating a number of points, users are more likely to vote negatively in bursts. In other words, as soon as a person begins to vote negatively, he is more likely to continue to do so further. Thus, it would be fair to give a negative vote less weight if it happens after a negative vote of the same user. Or, you can save the duration of a voting session as a variable, and estimate after a while your early votes in comparison with the later ones during this period of time. If someone votes for an hour, does that make the voice more valuable or less? You can conduct an experiment to find out.
One head is good ...
These aspects of the theory of MT - only the basics. Much more detail these and many other concepts are disclosed in the book Shuroveski. The most important thing is to remember that collective intelligence systems must constantly evolve. If done correctly, they can change our online life, and perhaps make us all a little wiser.
Source: https://habr.com/ru/post/62276/
All Articles