更改

=== Evidence 证据 ===

=== Evidence 证据 ===

[[File:Standardized Regression Coefficients.png|alt=Standardized Regression Coefficients for the collective intelligence factor ''c'' and group member intelligence regressed on the two criterion tasks as found in Woolley et al.'s (2010) two original studies.|thumb|Standardized Regression Coefficients for the collective intelligence factor ''c'' as found in Woolley et al.'s<ref name=":0"/> (2010) two original studies. ''c'' and average (maximum) member intelligence  scores are regressed on the criterion tasks.]]

[[文件:Standardized Regression Coefficients|缩略图|集体智力因子c的标准化回归系数]]

Woolley, Chabris, Pentland, Hashmi, & Malone (2010), the originators of this scientific understanding of collective intelligence, found a single statistical factor for collective intelligence in their research across 192 groups with people randomly recruited from the public. In Woolley et al.'s two initial studies, groups worked together on different tasks from the [[The Circumplex Model of Group Tasks|McGrath Task Circumplex]], a well-established taxonomy of group tasks. Tasks were chosen from all four quadrants of the circumplex and included visual puzzles, brainstorming, making collective moral judgments, and negotiating over limited resources. The results in these tasks were taken to conduct a [[factor analysis]]. Both studies showed support for a general collective intelligence factor ''c'' underlying differences in group performance with an initial eigenvalue accounting for 43% (44% in study 2) of the variance, whereas the next factor accounted for only 18% (20%). That fits the range normally found in research regarding a [[G factor (psychometrics)|general individual intelligence factor ''g'']] typically accounting for 40% to 50% percent of between-individual performance differences on cognitive tests.

 

 

Woolley, Chabris, Pentland, Hashmi, & Malone (2010),<ref name=":0" /> the originators of this scientific understanding of collective intelligence, found a single statistical factor for collective intelligence in their research across 192 groups with people randomly recruited from the public. In Woolley et al.'s two initial studies, groups worked together on different tasks from the [[The Circumplex Model of Group Tasks|McGrath Task Circumplex]],<ref>{{Cite book|title=Groups: Interaction and Performance|last=McGrath, J. E.|publisher=Prentice-Hall|year=1984|isbn=|location=Englewood Cliffs, NJ|pages=}}</ref> a well-established taxonomy of group tasks. Tasks were chosen from all four quadrants of the circumplex and included visual puzzles, brainstorming, making collective moral judgments, and negotiating over limited resources. The results in these tasks were taken to conduct a [[factor analysis]]. Both studies showed support for a general collective intelligence factor ''c'' underlying differences in group performance with an initial eigenvalue accounting for 43% (44% in study 2) of the variance, whereas the next factor accounted for only 18% (20%). That fits the range normally found in research regarding a [[G factor (psychometrics)|general individual intelligence factor ''g'']] typically accounting for 40% to 50% percent of between-individual performance differences on cognitive tests.<ref name=":5" />

Woolley, Chabris, Pentland, Hashmi, & Malone (2010), a well-established taxonomy of group tasks. Tasks were chosen from all four quadrants of the circumplex and included visual puzzles, brainstorming, making collective moral judgments, and negotiating over limited resources. The results in these tasks were taken to conduct a factor analysis. Both studies showed support for a general collective intelligence factor c underlying differences in group performance with an initial eigenvalue accounting for 43% (44% in study 2) of the variance, whereas the next factor accounted for only 18% (20%). That fits the range normally found in research regarding a general individual intelligence factor g typically accounting for 40% to 50% percent of between-individual performance differences on cognitive tests.

Woolley, Chabris, Pentland, Hashmi, & Malone (2010), a well-established taxonomy of group tasks. Tasks were chosen from all four quadrants of the circumplex and included visual puzzles, brainstorming, making collective moral judgments, and negotiating over limited resources. The results in these tasks were taken to conduct a factor analysis. Both studies showed support for a general collective intelligence factor c underlying differences in group performance with an initial eigenvalue accounting for 43% (44% in study 2) of the variance, whereas the next factor accounted for only 18% (20%). That fits the range normally found in research regarding a general individual intelligence factor g typically accounting for 40% to 50% percent of between-individual performance differences on cognitive tests.

Afterwards, a more complex criterion task was absolved by each group measuring whether the extracted ''c'' factor had predictive power for performance outside the original task batteries. Criterion tasks were playing [[Draughts|checkers (draughts)]] against a standardized computer in the first and a complex architectural design task in the second study. In a [[regression analysis]] using both individual intelligence of group members and ''c'' to predict performance on the criterion tasks, ''c'' had a significant effect, but average and maximum individual intelligence had not. While average (r=0.15, P=0.04) and maximum intelligence (r=0.19, P=0.008) of individual group members were moderately correlated with ''c'', ''c'' was still a much better predictor of the criterion tasks. According to Woolley et al., this supports the existence of a collective intelligence factor ''c,'' because it demonstrates an effect over and beyond group members' individual intelligence and thus that ''c'' is more than just the aggregation of the individual IQs or the influence of the group member with the highest IQ.<ref name=":0" />

Afterwards, a more complex criterion task was absolved by each group measuring whether the extracted ''c'' factor had predictive power for performance outside the original task batteries. Criterion tasks were playing [[Draughts|checkers (draughts)]] against a standardized computer in the first and a complex architectural design task in the second study. In a [[regression analysis]] using both individual intelligence of group members and ''c'' to predict performance on the criterion tasks, ''c'' had a significant effect, but average and maximum individual intelligence had not. While average (r=0.15, P=0.04) and maximum intelligence (r=0.19, P=0.008) of individual group members were moderately correlated with ''c'', ''c'' was still a much better predictor of the criterion tasks. According to Woolley et al., this supports the existence of a collective intelligence factor ''c,'' because it demonstrates an effect over and beyond group members' individual intelligence and thus that ''c'' is more than just the aggregation of the individual IQs or the influence of the group member with the highest IQ.

Afterwards, a more complex criterion task was absolved by each group measuring whether the extracted c factor had predictive power for performance outside the original task batteries. Criterion tasks were playing checkers (draughts) against a standardized computer in the first and a complex architectural design task in the second study. In a regression analysis using both individual intelligence of group members and c to predict performance on the criterion tasks, c had a significant effect, but average and maximum individual intelligence had not. While average (r=0.15, P=0.04) and maximum intelligence (r=0.19, P=0.008) of individual group members were moderately correlated with c, c was still a much better predictor of the criterion tasks. According to Woolley et al., this supports the existence of a collective intelligence factor c, because it demonstrates an effect over and beyond group members' individual intelligence and thus that c is more than just the aggregation of the individual IQs or the influence of the group member with the highest IQ.

Afterwards, a more complex criterion task was absolved by each group measuring whether the extracted c factor had predictive power for performance outside the original task batteries. Criterion tasks were playing checkers (draughts) against a standardized computer in the first and a complex architectural design task in the second study. In a regression analysis using both individual intelligence of group members and c to predict performance on the criterion tasks, c had a significant effect, but average and maximum individual intelligence had not. While average (r=0.15, P=0.04) and maximum intelligence (r=0.19, P=0.008) of individual group members were moderately correlated with c, c was still a much better predictor of the criterion tasks. According to Woolley et al., this supports the existence of a collective intelligence factor c, because it demonstrates an effect over and beyond group members' individual intelligence and thus that c is more than just the aggregation of the individual IQs or the influence of the group member with the highest IQ.

然后，一个更复杂的标准任务被免除，每组测量提取的 c 因子是否对原任务电池之外的性能有预测能力。标准任务是下棋(跳棋)对标准计算机在第一个和复杂的建筑设计任务在第二个研究。在同时使用小组成员的个人智力和 c 来预测标准任务的表现的回归分析中，c 有显著的效果，但是平均和最大的个人智力没有。各组成员的平均智力(r0.15，p0.04)和最高智力(r0.19，p0.008)与 c 有中度相关，但 c 仍是标准任务的较好预测因子。根据 Woolley 等人的研究，这支持了集体智力因素 c 的存在，因为它证明了一种超越团队成员个人智力的影响，因此 c 不仅仅是个人智商的集合或者智商最高的团队成员的影响。

后来每个小组进行测试，验证提取c因子是否具有预测原始任务以外的能力，进而解决了更为复杂的判据任务。在第一个研究中，判据任务是在标准计算机上玩跳棋（国际跳棋），在第二个研究中则是复杂的建筑设计任务。在使用组员个人智力和c因子来预测判据任务执行情况的回归分析中，c具有显著作用，而平均和最大的个人智力则没有。虽然单个组成员的平均智力（r = 0.15，P = 0.04）和最高智力（r = 0.19，P = 0.008）与c有中等程度的相关性，但是c仍然是判据任务更好的预测指标。根据伍利等人的说法，该结果支持了集群智力因子c的存在，因为它证明了超出小组成员个人智力外的影响，因此c不仅仅是个人智商的累加，或单纯受到智商最高组员的影响。

Engel et al.<ref name=":4" /> (2014) replicated Woolley et al.'s findings applying an accelerated battery of tasks with a first factor in the factor analysis explaining 49% of the between-group variance in performance with the following factors explaining less than half of this amount. Moreover, they found a similar result for groups working together online communicating only via text and confirmed the role of female proportion and social sensitivity in causing collective intelligence in both cases. Similarly to Wolley et al.,<ref name=":0" /> they also measured social sensitivity with the RME which is actually meant to measure people's ability to detect mental states in other peoples' eyes. The online collaborating participants, however, did neither know nor see each other at all. The authors conclude that scores on the RME must be related to a broader set of abilities of social reasoning than only drawing inferences from other people's eye expressions.<ref name=":13">{{Cite book|last=Engel|first=David|last2=Woolley|first2=Anita Williams|last3=Aggarwal|first3=Ishani|last4=Chabris|first4=Christopher F.|last5=Takahashi|first5=Masamichi|last6=Nemoto|first6=Keiichi|last7=Kaiser|first7=Carolin|last8=Kim|first8=Young Ji|last9=Malone|first9=Thomas W.|date=2015-01-01|title=Collective Intelligence in Computer-Mediated Collaboration Emerges in Different Contexts and Cultures|journal=Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems|series=CHI '15|location=New York, NY, USA|publisher=ACM|pages=3769–3778|doi=10.1145/2702123.2702259|isbn=9781450331456}}</ref>

Engel et al. (2014) replicated Woolley et al.'s findings applying an accelerated battery of tasks with a first factor in the factor analysis explaining 49% of the between-group variance in performance with the following factors explaining less than half of this amount. Moreover, they found a similar result for groups working together online communicating only via text and confirmed the role of female proportion and social sensitivity in causing collective intelligence in both cases. Similarly to Wolley et al., they also measured social sensitivity with the RME which is actually meant to measure people's ability to detect mental states in other peoples' eyes. The online collaborating participants, however, did neither know nor see each other at all. The authors conclude that scores on the RME must be related to a broader set of abilities of social reasoning than only drawing inferences from other people's eye expressions.

Engel et al.

Engel et al. (2014) replicated Woolley et al.'s findings applying an accelerated battery of tasks with a first factor in the factor analysis explaining 49% of the between-group variance in performance with the following factors explaining less than half of this amount. Moreover, they found a similar result for groups working together online communicating only via text and confirmed the role of female proportion and social sensitivity in causing collective intelligence in both cases. Similarly to Wolley et al., they also measured social sensitivity with the RME which is actually meant to measure people's ability to detect mental states in other peoples' eyes. The online collaborating participants, however, did neither know nor see each other at all. The authors conclude that scores on the RME must be related to a broader set of abilities of social reasoning than only drawing inferences from other people's eye expressions.

A collective intelligence factor ''c'' in the sense of Woolley et al.<ref name=":0" /> was further found in groups of MBA students working together over the course of a semester,<ref name=":8">{{Cite journal|author1=Aggarwal, I. |author2= Woolley, A.W. |last-author-amp=yes |date=2014|title=The effects of cognitive diversity on collective intelligence and team learning.|url=|journal=Symposium Presented at the 50th Meeting of the Society of Experimental Social Psychology, Columbus, OH.|doi=|pmid=}}</ref> in online gaming groups<ref name=":9">{{Cite journal|author1=Kim, Y. J. |author2=Engel, D. |author3=Woolley, A. W. |author4=Lin, J. |author5=McArthur, N. |author6= Malone, T. W. |last-author-amp=yes |date=2015|title=Work together, play smart: Collective intelligence in League of Legends teams|url=|journal=Paper Presented at the 2015 Collective Intelligence Conference, Santa Clara, CA.|doi=|pmid=}}</ref> as well as in groups from different cultures<ref name=":10">{{Cite journal|author1=Engel, D. |author2=Woolley, A. W. |author3=Aggarwal, I. |author4=Chabris, C. F. |author5=Takahashi, M. |author6=Nemoto, K. |author7=Malone, T. W. |date=2015|title=Collective intelligence in computer-mediates collaboration emerges in different contexts and cultures.|url=https://dl.acm.org/ft_gateway.cfm?id=2702259&type=pdf|journal=In Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems (CHI '15) (Pp. 3769–3778). New York, NY: ACM|doi=|pmid=}}</ref> and groups in different contexts in terms of short-term versus long-term groups.<ref name=":10" /> None of these investigations considered team members' individual intelligence scores as control variables.<ref name=":9" /><ref name=":8" /><ref name=":10" />

A collective intelligence factor ''c'' in the sense of Woolley et al. was further found in groups of MBA students working together over the course of a semester, as well as in groups from different cultures and groups in different contexts in terms of short-term versus long-term groups. None of these investigations considered team members' individual intelligence scores as control variables.

A collective intelligence factor c in the sense of Woolley et al. in online gaming groups as well as in groups from different cultures and groups in different contexts in terms of short-term versus long-term groups. None of these investigations considered team members' individual intelligence scores as control variables.

A collective intelligence factor c in the sense of Woolley et al. in online gaming groups as well as in groups from different cultures and groups in different contexts in terms of short-term versus long-term groups. None of these investigations considered team members' individual intelligence scores as control variables.

Note as well that the field of collective intelligence research is quite young and published empirical evidence is relatively rare yet. However, various proposals and working papers are in progress or already completed but (supposedly) still in a [[scholarly peer review]]ing publication process.<ref>{{Cite web|url=https://sites.google.com/a/stern.nyu.edu/collective-intelligence-conference/|title=Collective Intelligence 2016|website=sites.google.com|access-date=2016-04-27}}</ref><ref>{{Cite web|url=https://sites.lsa.umich.edu/collectiveintelligence/posters/|title=Posters {{!}} Collective Intelligence 2015|website=sites.lsa.umich.edu|access-date=2016-04-27}}</ref><ref>{{Cite web|url=http://collective.mech.northwestern.edu/?page_id=217|title=Proceedings {{!}} Collective Intelligence 2014|website=collective.mech.northwestern.edu|access-date=2016-04-27}}</ref><ref>{{Cite arxiv|eprint=1204.2991|last1= Malone|first1= Thomas W.|title= Collective Intelligence 2012: Proceedings|author2= Luis von Ahn|class= cs.SI|year= 2012}}</ref>

Note as well that the field of collective intelligence research is quite young and published empirical evidence is relatively rare yet. However, various proposals and working papers are in progress or already completed but (supposedly) still in a [[scholarly peer review]]ing publication process.

Note as well that the field of collective intelligence research is quite young and published empirical evidence is relatively rare yet. However, various proposals and working papers are in progress or already completed but (supposedly) still in a scholarly peer reviewing publication process.

Note as well that the field of collective intelligence research is quite young and published empirical evidence is relatively rare yet. However, various proposals and working papers are in progress or already completed but (supposedly) still in a scholarly peer reviewing publication process.

 

 

=== Predictive validity ===

=== Predictive validity ===