The "holy grail" of artificial intelligence research for decades has been to design computers with robust, integrated, human-like intelligence. This goal has proven elusive, in spite of a massive amount of research. But another goal is just now becoming feasible, and so has been the subject of much less research: using vast computer networks to create new kinds of intelligent entities that combine the best of both human and machine intelligence. One key to designing such human-centered computing systems is better ways of measuring the collective intelligence they exhibit. That is the focus of this research, which represents a collaborative effort among researchers at MIT (lead institution), CMU and Union College. The PIs will first use analogies with what is already known about measuring individual intelligence to suggest new ways of measuring the collective intelligence of complex human-machine systems. For instance, they will determine whether the striking pattern of correlations across tasks that characterizes individual human intelligence even exists for human-machine groups. Next, a series of statistically validated tests will be developed to measure the key components of collective intelligence in human-machine groups. Then, to better understand the "active ingredients" of collective intelligence, the PIs will use what is already known about how groups of people interact effectively to measure micro-level behavior in human-machine groups. A key goal will be to find critical factors (such as group size, technological support, or individual capabilities) that contribute to a human-machine group's adaptability across a wide range of tasks.
Most people and computers today are parts of larger human-machine systems that must cope with a wide range of problems. This research will provide powerful new tools for managing and designing such systems. Imagine, for instance, that one could give a short "collective intelligence test" to a top-management team, a product development team, or a collection of Wikipedia contributors. Imagine that this test could predict the team's future performance on a wide range of important tasks. And imagine that the test could also help suggest changes to the team that would improve its flexibility. Or imagine that designers of new collaboration software tools could use a single test to predict how well their tools would improve a group's effectiveness on many different tasks. From the smallest business work groups to our largest societal challenges, there are now many new opportunities for people and computers to solve problems together, not just more efficiently, but also more intelligently. This work will help build a firmer scientific foundation for doing this.
Broader Impacts: With individual humans, it is relatively easy to measure intelligence, but it is difficult to increase that intelligence or to observe the detailed events inside the brain that give rise to it. With human-computer groups it is much easier to observe and change factors (such as group size, composition, and technological support) that are likely to determine the group's collective intelligence. Thus, there is a profound intellectual opportunity, not just to learn more about how to design intelligent human-computer systems but also to gain new insights into the very nature of intelligence in complex systems. The results of this research, therefore, will be of interest not only to researchers in computer-supported cooperative work, human-computer interaction, and artificial intelligence, but also more broadly to fields such as cognitive science, social psychology, and organization theory.
