Earlier this month I attended the MIT Communications Forum on Collective Intelligence. Here is a summary of the event.
David Thorburn introduced the evenings panel and explained that the idea of collective intelligence has been a topic that the Forum has covered in the past, mentioning Howard Rheingold’s session on Smart Mobs in 2002 as an example. He also pointed out that it has been an ongoing preoccupation for people in technology for a long time.
One of the missions of the Forum, he explained, is to speak to a literate citizenry – and he asked the speakers to speak in a common, literate language. With that he opened the Forum by asking Thomas Malone to provide a definition and overview of collective intelligence.
Malone started by saying the the Forum was an example of collective intelligence, in that it seeks to take advantage of the intelligence of the audience. Collective intelligence happens through conversations among a literate citizenry.
He defined collective intelligence as groups of individuals, doing things collectively that seem intelligent. By that definition, it’s clear that collective intelligence has been around for a very long time – families, armies, countries – have all provided examples throughout history. He pointed out that all of these groups (and others) have also exhibit collective stupidity. Recognizing the differences between collective intelligence and collective stupidity is important.
In the past few years, he went on, there have been some interesting examples of collective intelligence: Google – not jut the technology or the company; but also the system – the creation of Web pages, the linking of pages and the technology that harvests all of this information. It’s an amazing example because it combines people and computers in a way that never existed before.
Wikipedia is also an excellent example. Again, it’s not just the technology. What’s really amazing is the organizational design that has arisen around Wikipedia. There, the community – which is virtually entirely voluntary – has invented an organizational design that allows thousands of people from all over the world to create an intellectual product with out centralized control.
He thinks these are just the beginning of whole new classes of intelligent entities that we will see over the coming decades. In order to take full advantage of them, we will need to understand their possibilities at a much deeper level. The goal of this understanding is to connect people and computers so they can act more intelligently than any people, machines or groups ever have before.
Consider the research status of collective intelligence as compared with artificial intelligence. In AI, the goal is to create machines that are more intelligent that humans; and much work has been done on this topic so the core questions of AI are well understood. Too little time has been spent on developing the same understanding for collective intelligence and we need to figure out how to take advantage of both people and machines.
Sandy Pentland spoke next and described his general thoughts on collective intelligence. The reason people come together into groups, he explained, is to be more intelligent. There are also problems that arise when groups collect – conflict, group think, etc.; and the larger the organization, the bigger these problems become. In some ways, collective intelligence can be seen as the attempt to simply break even and avoid collective idiocy.
This break even can be achieved by developing more sensible organizations – not ad hoc but based on data, science and modeling. In modern organizations, there is a great deal of organizational data that can be analyzed – emails, memos, etc. But the most important communication happens face-to-face. This communication is the delicate, content-full discussions that really matter. Until recently though, this was unmanageable and unstorable. Most of it was invisible and couldn’t be organized and managed. Now, Pentland explained, we are able to measure face-to-face communication in real time.
Pentland and his team are tracking face-to-face interaction with a name badge the keeps track of whom the wearer is talking to and when – but not the content on the conversations. It creates a recording of what happens when people see and are animated by each other.
He went onto describe research conducted at a German bank looking at the dynamics of communications. It analyzed communications patters – both over email and face-to-face. The resulting maps were very different, and, for the first time, provided a means to view how information flows within an organization.
Pentland emphasized that neither alone was key; but that the combination of the two helped provide and understanding of the most effective modes of communication. This work has allowed an analysis of what is happening, who is over worked, the quality of group interaction, etc. It permits the identification of potential problems and provides the ability to make suggestions for improved communication and information flow.
It can, of course, also seem more than a little Big Brother; but Pentland sees a value in creating personal information tools – based on electronic communication tracking, proximity badges, etc. – that lead to more awareness and alert people to issues of group think or polarization. His hope is that with this information people will be better able to manage themselves and achieve better results for themselves and their organizations.
He also mentioned other work that could lead to more formal collective intelligence applications. These included market-based models for predicting future events and systems for identifying and addressing the problems of gossip and rumor – which can be identified by paying attention to communication patterns.
Thorburn was interested in how rumor or gossip could be identified without access to the content of the communication. Pentland explained that you do need to have some baseline understanding in order to identify patterns of communication with patterns of decision making. With this though, he said, you can see who is reacting to the same memes versus those you are thinking independently.
Malone liked Pentland’s research to Von Leeuwenhoek and the microscope: what the microscope allowed was the observation of things that have always existed but in detail never before possible. Pentland is creating an organizational microscope for observing existing communication behavior in ways not possible in the past.
Lakhani spoke next and described how he came to be involved with collective intelligence – which he described as being quite by accident. He’d been working with a client, trying to sell them some software. They wouldn’t buy because they claimed to have created a similar system themselves. Given the complexity of the problems they were trying to address, Lakhani couldn’t accept this explanation; but when he spent time looking at what they’d done, he saw they were years ahead of others. How had they done this? By tapping into the Open Source community.
While at MIT he saw examples of this again and again as people were using communities to solve tough computing challenges. How, he wondered, is it that this distributed organizational model can work better than large, centralized systems? He came to recognize that there was something real and different about the Open Source approach – despite the general skepticism of the time. These projects typically entailed lots of work – so what motivated people to get involved in what is essentially unpaid work?
Lakhani found that Open Source is a prototypical example of collective intelligence and there there is a heterogeneity in motivation. Some people simply believe in the model and so want to support it. Others are pragmatists who are trying to solve a specific problem they are facing. The fact of the matters is that the community doesn’t care about motivation as long as the work gets done.
For this to happen, communities need to have a participation infrastructure that will attract people. There also needs to be an understanding of the ownership of the intellectual property. For Open Source projects this has been solved but, he wondered, will it become an issue for collective intelligence efforts? Governance also needs to be considered. Wikipedia offers an example of a functional flat structure, but within each article there are battles to reach compromise and build consensus.
For Lakhani, the Open Source community provides an inspiration; but he sees other examples. Innocentive, for instance, takes scientific problems beyond the bounds of an organization to tap into the collective intelligence. Often, people are able to provide solutions that are outside of their domain. One of the hopes of collective intelligence is that it is able to aggregate the pockets of “sticky” intelligence that exist around the world.
Another example he provided was Threadless, a t-shirt design company. Its designs are user submitted and user judged. User demand for specific shirts is also tracked to determine how many of each design is actually produced. This model – though small and specialized – shows how the way an organization can be redefined and how much work can be done by the community.
Thorburn asked the panel to provide additional examples of collective intelligence at work, and to discuss its limitations.
Malone described the idea of collective prediction and predictive markets mentioned by Pentland. The idea is to broaden this and to predict more types on information. This will require an infrastructure that includes the participation of computational agents. In many cases, these types of software agents can do a better job of predicting things that humans with the result being faster, smoother markets. People will continue to play an important role to step in if the agents are making bad predictions.
In terms of measuring collective intelligence, Malone pointed out that we’ve been measuring human intelligence for more than a century and as a result have a precise definition of intelligence. According to the psychometric definition, how well does at one intellectual task is a good predictor of how well one will do on other types of tasks; and there are all sorts of statistically significant relationships and correlations.
The question is, can the same thing be true for groups of humans or humans and computers? Will a group that does well on some tasks do well on others? While it is currently unclear if this is true, there is a project underway to find the answer. Malone is especially interested in what causes difference in intelligence between groups and to find ways to improve the collective intelligence capabilities of groups in general.
Thorburn raised the issue of collective intelligence as a surveillance tool and asked Pentland for his thoughts, as well as about the limitations of collective intelligence.
Regarding privacy, Pentland pointed out that like it or not, we are constantly being monitored or monitoring ourselves (he pointed to cell phones as one means of this happening). The real question is determining the balance between privacy and advantage: how much information do we need to give up for what level of benefit.
His vision is the creation of feedback tools able to provide valuable insight for individuals. In terms of limits, there certainly are some; but we need to work to detect our errors and improve the way we think in groups.
Malone pointed out that behind Thorburns questions was a widespread misnotion – wisdom of crowds – that crowds are intelligent and can by their nature can solve problems. He explained that collective intelligence isn’t magic; sometimes it works and sometimes it does not. It’s also very complicated and whether it is successful or not depends on what one is trying to achieve. His goal is to put it on a firm scientific foundation so we can better judge when and how collective intelligence will work well. For this to happen, we need to know which people and computers to connect and how to connect them effectively.
Lakhani said that collective intelligence is not a universal solutions. He has worked on decision markets in the past and the data showed that the system can work. One of the biggest issue is that managers don’t want to use predictive markets or collective intelligence because they can be contradictory to the role of the manager. This is often an organizational rather than a technical limit and at this point there are no course on community management so there are few mechanisms for doing this yet.
He also pointed to legal and technical issues as potential limitations. From a legal perspective, our views on intellectual property are a major question mark. How will profits be shared the profits from that which is created by the community? And on technology, not everything can be modularized and distributed. It works well in software, but how would it work in something like drug discovery?
With these comments in mind, Thorburn asked if the best we can hope for from collective intelligence was selling more t-shirts and wondered if there were any applications that weren’t based on profit and loss.
Malone pointed to Wikipedia is an example. Another, he suggested, is the use of collective intelligence to monitor and deal with climate change. He mentioned a project using technology to allow people to propose and analyze plans for addressing climate change. It will probably focus on government policy and give people the ability to access, view and analyze massive computer climate simulations. Collective decision making around climate data – by allowing people to consider and vote the policies that make the most sense. Finally, he discussed using technology for improved argumentation by providing structure around discussion through a series of positions and structured arguments so debates can become become less chaotic.
Lakhani mentioned Open Congress as a means to enable citizens to observe what the legislature is doing. He also discussed the rise of creative commons and peoples willingness to allow their content to be remixed to create new content.
Pentland suggested collective intelligence as a tool for detecting depressions by analyzing patterns of interaction. He again mentioned the benefit of collective intelligence as a reflective aid. In the larger arena, he said that collective intelligence could be used to detect societal discord and mentioned patterns of cell phone usage in the UK and the correlation of that data with social integration. He concluded by describing the Legatum Center for Development and Entrepreneurship, which is bringing together change agents from around the world to consider problems and bring solutions back to their communities.
Overall, it was a strong and interesting Forum that raised some pretty interesting points. One of the things I found most interesting was how much the patterns and context of communication among a group could say about the group’s effectiveness. That, coupled with the difference in patterns of electronic and interpersonal communication, made me question my own communication patterns. I use the phone very infrequently now – most of my communication is either over email or Facebook, twitter, etc.
The analysis that was described didn’t appear to take these channels into account – but were focused more on email, telephone and interpersonal. How different would the patterns be if these were taken into account?
[tags]MIT, Communication Forum, Collective Intelligence, Karim R. Lakhani, Thomas W. Malone, Sandy Pentland[/tags]