The MIT Center for Collective Intelligence recently published an important overview of the theory and mechanisms behind successful crowdsourcing efforts. Their report, called “Harnessing Crowds: Mapping the Genome of Collective Intelligence“, can be found here.
Their research reveals similarities behind many high-profile collective intelligence (CI) systems, including Threadless, Wikipedia and InnoCentive. It then describes how these lesson can be applied to the design of other successful CI platforms.
I call this work the MIT Approach to Collective Intelligence, which is a generic approach applicable to a wide range of problems and circumstances. This post explores how I have used the MIT approach in my own work and how you can use it to build your CI system. I also offer a slightly reformatted version of their content – in the form of a detailed process flowchart – which I hope will make their work more accessible to a wider audience.
The MIT approach to collective intelligence
According to the Center for Collective Intelligence, a good collective intelligence platform (CI) must address the following themes:
- Goals, referring to the desired outcome;
- Incentives, referring to the motivational factors;
- Structure/process, referring to the business model and organizational structure to complete the task; and
- Staffing, referring to the people required to support the business model and sustainability of CI within the organization.
These four themes then translate into the following four questions:
- What is to be accomplished?
- Why should anyone help out?
- How are they meant to contribute?
- Who will perform the necessary work?
, below, illustrates how these four themes and questions interact to form the building blocks of any collective intelligence system.
[caption id="attachment_734" align="alignnone" width="495" caption="Figure 1, the basic building blocks of a CI system"]
Developing a detailed decision tree
This approach then asks a series of sequential, logical questions, the answers of which form specific guidelines for all CI systems:
- Can activities be divided into pieces? Are necessary resources widely distributed or in unknown locations?
- Are there adequate incentives to participate?
- What kind of activity needs to be done?
- Can the activity be divided into small, independent pieces?
- Are only a few good (best) solutions needed?
- Does the entire group need to abide by the same decision?
- Are money or resources required to exchange hands or motivate decision?
The answer to these questions comes in the form of specific “genetic” building blocks, such as the “Create” gene, the “Crowd” gene, or the “Decide” gene. The paper concludes with a detailed table listing these genes and how they interact with the questions above.
In my own work developing online scenario planning systems
, I have found it useful to translate these questions into a flowchart that can be used to help navigate this process. This chart is presented in Figure 2
, below, which presents each of these questions and possible answers in the form of a decision tree (full PDF by clicking on the image or downloading here
[caption id="attachment_710" align="alignnone" width="658" caption="Figure 2, a flowchart for the design of any CI system"]
Applying this method to the real world
Answering these questions in turn, using this flowchart as your guide, can help you develop your own strategy for setting up a successful collective intelligence system.
Lets see how this works in two real world examples taken from the original paper. The first is InnoCentive
, a platform for crowdsourcing novel solutions to complex scientific problems, and the second is the t-shirt design company Threadless
below, illustrates the core “collective intelligence genes” for each example.
[caption id="attachment_714" align="alignnone" width="625" caption="Figure 3, the MIT Approach applied to two real world examples"]
InnoCentive has two core CI functions. These are “create” and “decide”. Scientific solutions are created by the crowd, who do it for the promise of money in a contest-like setting. The second function, “decide”, is used to determine which solution gets the reward. This is decided by the management of the sponsoring company, who also does it for money, in a hierarchical decision process. This example is important because it illustrates that not all crowdsourced efforts rely upon “the wisdom of the crowds” in all circumstances. This is especially true in scientific ventures and other high-precision efforts, where technical rigour and excellence are more important than popularity or attractiveness to the majority of participants.
Threadless is another interesting example. Their core platform has three functions, “create” and two different versions of “decide”. First, T-shirt designs are created by the crowd for both the love of design and t-shirt culture as well as the promise of potential money. Their creations are submitted in a contest fashion. Next, the crowd also
picks which designs it likes best. They do this by scoring each t-shirt they like and the shirt with the highest average score, wins. Finally, it is up to the management of Threadless to decide which of the most popular designs it will use to put into production. They do this through a traditional management hierarchy which is primarily motivated by money (even though they might still love their jobs).
In my experience, the MIT approach to collective intelligence offers a robust framework for thinking about crowdsourcing in a rigourous, pragmatic way. It is informed by specific examples, backed up by a theory of social interaction online, and when combined with my decision-tree flow chart, can be readily applied to a variety of other circumstances.
I hope that this summary of their approach helps you think about setting up your own collective intelligence platform, in wherever context that might be.]]>