Sunday 8 June 2014

Collective Intelligence: 
What is it?  
How can we measure it?  
And increase it?


Sloan School of Management

VIDEO



OVERVIEW: This talk will describe how the same statistical techniques used to measure intelligence in individuals can be used to measure the "collective intelligence" of groups.  We find that, just as with individuals, a single statistical factor can predict the performance of a group on a wide range of different tasks.  This factor is only weakly correlated with the group members' individual intelligence.  It is, however, correlated with the group members' social perceptiveness, conversational behavior, and gender.
The talk will also include brief overviews of other work to increase collective intelligence by: (a) combining predictions from humans and computers, (b) mapping the "genome" of collective intelligence, and (c) harnessing the collective intelligence of thousands of people around the world to develop proposals for what to do about global climate change.

READINGS:
    Bernstein, A., Klein, M., & Malone, T. W. (2012). Programming the global brainCommunications of the ACM55(5), 41-43.
    Malone, T. W., Laubacher, R., & Dellarocas, C. (2010). 
The collective intelligence genomeIEEE Engineering Management Review38(3), 38.

    Woolley, A. W., Chabris, C. F., Pentland, A., Hashmi, N., & Malone, T. W. (2010)  Evidence for a collective intelligence factor in the performance of human groups, Science, 330 (6004), 686-688
    Malone, T. W., Laubacher, R., & Dellarocas, C. (2010) The Collective Intelligence Genome, Sloan Management Review, Spring 2010, 51, 3, 21-31 (Reprint No. 51303).
    Bernstein, A., Klein, M., & Malone, T. W.  Programming the global brainCommunications of the ACM, May 2012, 55 (5): 41-43. 

37 comments:

  1. Here is a version of the social intelligence test "Reading the Mind in the Eyes:" http://kgajos.eecs.harvard.edu/mite/

    ReplyDelete
  2. Does the variance of individual intelligence within a group affects the group score when compared among groups with the same average intelligence?

    ReplyDelete
    Replies
    1. We didn't specifically look at that, and I think it is an interesting question. If I were to speculate, I would guess that groups with higher variance and the same average would be more collectively intelligent because they would have a higher maximum intelligence and we found that correlated with collective intelligence.

      Delete
  3. You defined intelligence as to be good, to perform well at a cognitive task. But how could a computer be good in performing cognitive intelligence? What is your definition of cognitive if it could be apply to computer?

    ReplyDelete
    Replies
    1. My definition of intelligence depends on the actions an entity takes, so from that point of view, an entity that acts "intelligently" could be a person, a computer, an animal, a group of people, a group of animals, a group of people and computers, etc...

      Delete
  4. How did you find that people are as good as reading between the lines through a computer than to see emotions in looking into the eye of others?

    I guess, Steven Harnad, this could ruin my theory of difficulty to mind-read through computer.

    ReplyDelete
    Replies
    1. People that are used to communicate via extensive computer chats develop in time sensitivities and skills that are adapted to this medium. These do not replace eye contact and body language sensitivities but augment them. After all the medium does affect the overall nature of the interaction. I would say that certain computer mediated interactions can expose aspects of personality that otherwise remain unexpressed.

      Delete
    2. Thank you! Do you know some references about this topic? If you do could you give some please?

      Delete
    3. Well, I read a discussion on this very point in some web forum a couple of years ago. I will try to recall the reference :-)

      Delete
    4. I think that the skill of Reading the Mind in the Eyes must also be correlated with a much broader range of other skills, including the ability to understand what other people are thinking and feeling based on other cues (such as what they type in a computer chat).

      Delete
  5. Thank you for the talk. I have two questions for Professor Malone. First, are there or aren’t there many different kinds of intelligence? Second, does g correlates with non-linguistic and non-mathematic skills, for instance playing the piano or painting?

    ReplyDelete
    Replies
    1. Some people (like Howard Gardner) argue that we should call a number of other kinds of abilities “intelligence.” This is just a question of what we call things. I don’t know specifically whether g correlates with piano playing and painting. But there are certainly some skills that don’t correlate with g. For instance, I learned recently that the ability to recognize faces is not correlated with g.

      Delete
    2. Thank you for the response, Professor Malone!

      Delete
    3. I wonder if something like cultural intelligence, which I believe Jacqueline is studying, becomes more relevant in groups of individuals from diverse origins. One would think that social sensitivity, assuming that it is context specific, would only be useful in such a circumstance if individuals were able to adapt to the various cultural norms that they were presented with.

      Delete
  6. Nous avons vu que plusieurs facteurs affectent la variable c. Ces facteurs peuvent être la moyenne de la perception social de chaque membre d'un groupe, l'égalité de la parole entre chaque membre et la proportion de femmes dans le groupe.
    Entre ces différents facteurs, lequel a le plus d'impact sur l'intelligence collective d'un groupe? (Entre les facteurs discutés et les facteurs non-discutés).

    Translation : We have seen that different factors influence the variable c. These factors might be : the average social perceptiveness of group members, equality of distribution of turn-taking and proportion of female in the group.
    Which of theses factors influence the most the collective intelligence of group? (Between theses 3 factors or others factors not discussed)

    ReplyDelete
    Replies
    1. When all three of these factors are included in a regression at the same time, the only one that significantly predicts c is social perceptiveness.

      Delete
  7. Dear Thomas, Thank you very for this luminous presentation! I have one question : Do you think that it is possible to define and measure a general cognitive complexity of scientific content into scientific articles to understand and a general intelligence score or ability able to understand it?

    ReplyDelete
    Replies
    1. I don’t know exactly how to do this, but it certainly seems to me that it might be possible.

      Delete
  8. In the last part of your talk, you said you hoped the global brain (made of humans and machines) would learn to be wise and not just smart. Do you think that machines can supply such wisdom? Or is it humans who make the wise decisions based on the information and intelligence they gain from using machines?

    ReplyDelete
    Replies
    1. Of course, the answer to this question depends on what we mean by “wise,” but I would say that in the foreseeable future we will need some people involved to arrive at decisions I would want to call wise. Could computers, alone, someday acquire the abilities to act “wisely”? I think that’s possible in the long run, but I don’t think it will happen anytime soon.

      Delete
  9. This was a very wonderful talk thank you, my question is about a future of the combination of human-computer to reach a high level of collective intelligence, in this future computer intelligence are going to be more and more smarter this collective intelligence are going to be bow in the side of computers. What is your opinion on this !

    ReplyDelete
    Replies
    1. I’m not sure I understand this question, but I’ll assume the question is: Will people or computers be more important in the collective intelligences of the future? The answer to that question, I think, is that both will be critical.

      Delete
    2. This comment has been removed by the author.

      Delete
    3. What exactly I mean is: do you think that computers are going to be more critical and more important than the human been in this collective intelligence and if eventually it will be, do you think the collective intelligence will be only the interaction between machines

      Delete
  10. Thank you for the interesting talk. I think your Climate CoLab showed that collective intelligence works so well because it allows people to dedicate all of their individual intelligence (g) to particular issues and then combine their work and have a collaborative intelligence of (g+g+g...). It seems to me that machines act more as a lubricant to this social machine rather than a real source of intelligence.

    ReplyDelete
    Replies
    1. Yes, one important kind of collective intelligence is that in which the role of computers is to help humans communicate and collaborate more effectively, and I think the Climate CoLab is an example of this.

      Delete
  11. Thank you for a very interesting talk. I have a philosophical question regarding collective intelligence. Do you think that collective intelligence is qualitatively different from individual intelligence or does it display mostly quantitative differences. Understandably the question is not very well formed because also quantitative differences may give rise to, emergent, qualitatively differentiated kinds of intelligence. Still, even in small groups, is there something in group intelligence which is substantially different? For example: depth vision requires a minimum of two points of view to be combined together something that seemingly cannot be accomplished by a single agent. It seems that this would be a good example of a qualitative difference but it is not because a single agent can sample multiple views while moving. So the question seems to be much more subtle.

    ReplyDelete
    Replies
    1. I think that whether you call something “collective intelligence” or “individual intelligence” is a matter of perspective, not of reality. For instance, you could view a single brain as an example of collective intelligence because the brain’s intelligence arises from the interactions of many different neurons and brain regions. In this sense, then, I would say that there is no qualitative difference between individual and collective intelligence.

      Delete
  12. Thanks for the accessible and stimulating presentation.

    I have a few comments and questions:

    1) Your football example suggested that human-computer teams perform best. However, a very well-programmed computer algorithm might outperform the human-computer team. Also, if we employ different tasks, humans or computer may perform best alone. For example, unless you’re a grand master at chess, the computer is better off without human influence. Alternatively, take the ancient Chinese board game ‘Go’ where an average human player will not benefit from a computer aid. I think it’s important to consider the type of social combination which Steiner outlined (e.g., compensatory, conjunctive, additive,) when comparing human, computer, and mixed collaborative teams.

    2) Are companies and government after this information? It seems to me that a further understanding of ‘c’ and related phenomenon may have huge implications across various social groups (government, industry, academia, etc.).

    3) The climate co-lab may provide motivation and diminish the ‘what can one person do’ mentality. Have researchers analyzed similar large-scale projects?

    4) In your experiments, did hierarchical structure affect ‘c’? While even turn-taking resulted in improved ‘c’, turn-taking can occur well within a hierarchical structure. I think elucidating the influence of a hierarchical structure on collective intelligence will increase in importance as researchers study larger group collaboration.

    ReplyDelete
    Replies
    1. My answers to your questions, in order:

      (1) Yes, I agree. Our results in this experiment do not prove that the combination of people and computers is always better. They only show that it was in this case. Another very interesting question is: What are the conditions under which the combination of people and computer is better than either alone?
      (2) Yes, I agree that a better understanding of c and related phenomena could have huge implications for many kinds of organizations. That’s one reason we’re working on this!
      (3) Yes, I agree with you about this potential benefit of the Climate CoLab. I can’t think of other examples where researchers have analyzed similar large scale projects.
      (4) In our experiments so far, we didn’t study the effect of group structure on c, but are doing some early work along those lines now.

      Delete
  13. Maybe I'll just post a pet explanation for how women promote group intelligence: there are social norms about the roles people should have in conversations (FISHMAN, P. (1983). "Interaction : the work women do"): they contribute to discussion subjects brought about by others, cooperate in group endeavours, encourage input from others, etc. There are strong mechanisms to enforce those social roles (cf. Monnet – in French). Thus, it's probably not about how men behave – it's about the work women do to make conversations more efficient.

    As for the social perceptiveness thing, Fine's Delusions of Gender make a strong argument for the idea that men and women are more or less equal at it. If I remember well, a bit in the same way girls usually do bad at math unless they're primed to think they're good at it, men will do as well as women if they're primed to think they're good at social perceptiveness.

    ReplyDelete
    Replies
    1. This comment has been removed by the author.

      Delete
  14. These are very interesting hypotheses. Maybe you can do some studies to test them!

    ReplyDelete
  15. I think that way Robert Rupert's definition of cognition is written, it could also mean that groups can be in a cognitive state:

    "A state is cognitive only if it is the state of a component of a cognitive system - a persisting collection of mechanisms that contribute, in overlapping subsets, to the production of a variety of forms of intelligent behaviour."

    What do you think?

    ReplyDelete
  16. I was thinking of a thread (if there exist any) or a general principle that can connect three seemingly unrelated parts of the talk. Part (a) introduced the measure of collective intelligence c and simulation with predictive markets. Predictive markets were particularly interesting to me because they they demonstrated a simple language between humans and machines (market / money) which is effective at increasing their joint performance.

    But this language is a one-dimensional signal - a price of a kind of simulated security (or an event). In part (b) an intuitive concept of 'genome' of collective intelligence was introduced, but, as pointed out by Steven Harnad, it is difficult to interpret the success and application possibilities of this concept. It seems that Climate CoLab (Part c) diverges even further from the clean and testable proposition of collective intelligence factor c. Actually it is difficult to see collective intelligence in Climate CoLab - it looks more like a collaboration tool for individual people (this is by no means underestimation - it looks very important and useful).

    It may be that collective intelligence is much more easily achievable in settings where communication between humans and computers is simple an one dimentional (as in case of predictive markets). In more complicated situations (e.g. Climate CoLab or global warming) some sort of a language which would allow computers and humans to share meanings (e.g. something like Information Economy Meta Language (IEML): http://en.wikipedia.org/wiki/Pierre_L%C3%A9vy#Current_project) is needed for collective intelligence to emerge.

    ReplyDelete