====== The Wisdom of Crowds I ======

//Lecture on 4.10.2007//

{{hivemind:wisdom_of_crowds.ppt|Presentation}}

Under certain circumstances, groups outperform individuals, even experts. If the probability that a member of a group is correct is more than 50%, the probability of a correct answer by the majority of the group increases to 100% as the group size increases. On the other side if there is a probability that a member has a less than 50% chance of being correct, group majority answer tends towards 0% correctness as group size increases.
    * [[http://en.wikipedia.org/wiki/Condorcet's_jury_theorem|Condorcet Jury Theorem]]
  == examples ==
    * guessing the gummnibears
    * [[http://en.wikipedia.org/wiki/Kasparov_versus_The_World|Garry Kasparov vs. The World]]
    * [[http://wilsonsalmanac.blogspot.com/2006/02/francis-galton-and-ox.html| Galtons's ox experiment]]


===== Guessing the Gummi Bears =====

The class was presented with a clear plastic bag with a large number of gummi bears inside, and asked to individually write on slips of paper their guess as to how many were in the bag. The person coming the closest would win the bag.

The bag in reality contained 813 bears. There were 18 guesses, of which one was dropped as the answer was in grams and thus invalid.

The following table gives the individual guesses, sorted by distance from the correct answer.

^ Guess ^ Distance ^
| 755 | 58 |
| 900 | 87 |
| 698.5 | 114.5 |
| 1000 | 187 |
| 474 | 339 |
| 450 | 363 |
| 384 | 429 |
| 333 | 480 |
| 276 | 537 |
| 251 | 562 |
| 205 | 608 |
| 185 | 628 |
| 1500 | 687 |
| 1500 | 687 |
| 1600 | 787 |
| 1983 | 1170 |
| 3000 | 2187 |

After dropping the outliers (largest and smallest guesses, 185 and 3000), the average guess of the group was 820.63, a distance of **7.63** from the correct answer, far better than the best guess of 755/58 (congratulations, Tom!), and a great surprise to the teacher, who had expected it to work but not so amazingly well.

Many gummi bears sacrificed their lives during this experiment in the name of science.


===== kinds of problems =====




==== cognition problems ====
  * problems with an definit answer
  * they must not have a universal answer but they have one
  * its an answer you can pick
  * president election
  * gulty or not questions
  * weather forecast


==== coordination problems ====
  * Experiment: Make groups of persons born in the same month without talking (our record: 28s)
  * group discussions
  * team sports
  * car assembly



==== cooperation problems ====
  * self-interested individuals working togheter
  * TV show "Survivor" or "Big Brother"
  * Ultimatum Game
  * eployees in a company
  * [[http://www.upi.com/NewsTrack/Science/2007/10/08/chimps_choose_more_rationally_than_humans/8872/|Chimps  play the Ultimatum Game more "rationally" than humans (Max Planck Institute Leipzig!)]]

==== Prediction Markets ====

[[http://hsxresearch.com/assets/LL_forecast.pdf|HSX.com Accuracy]]

[[http://googleblog.blogspot.com/2005/09/putting-crowd-wisdom-to-work.html|Prediction Markets at Google]]

[[http://blog.commerce.net/?p=197|Fortune Article about InTrade]]


===== Links =====
[[http://www.spiegel.de/international/spiegel/0,1518,448747,00.html|European cities do away with traffic signs]]

[[http://www.youtube.com/watch?v=Ka9IwHNvkfU|Lecture hold by Sergey Brin with some information about  Pagerank]]

[[http://infolab.stanford.edu/~backrub/google.html|The Anatomy of a Large-Scale Hypertextual Web Search Engine by Sergey Brin and Lawrence Page]]

[[http://www.crowdchess.com/|CrowdChess, a site where groups of people can play chess against each other]]

[[http://en.wikipedia.org/wiki/PageRank|PageRank]]

[[http://www.rtl.de/tv/tv_953245.php|Live Experiment regarding The Wisdom of Crowds on RTL - 20.01.2008 - 20:15]]