Interdisciplinary Case Study: Understanding the Cooperation of Humans and Robots through the Collaboration of Social and Computer Scientists

摘要

In our research, we have been obsessed with the idea that it should be possible to use artificial intelligence (AI), to support not only individual humans, such as physicians making difficult decisions or athletes perfecting their game, but also groups of humans trying to work together better. Can AI be used to help humans to help themselves? So, for over 5 years, we have been collaborating on a deepening set of projects with this objective. The article published recently in iScience, “Network Engineering Using Autonomous Agents Increases Cooperation in Human Groups” (https://www.cell.com/iscience/fulltext/S2589-0042(20)30630-1#%20) represents a culmination, in some ways, of this effort—because, among other things, we are able to show that certain kinds of “simple AI,” when introduced into human groups, can actually enhance levels of cooperation in human groups. Cooperation is challenging because it creates a social dilemma: the group does best if people cooperate, but each individual is always templated to keep their resources given their self-interest. It is also clear that individual people, if they do not have some overarching power (like being the “boss”), can find it very difficult to facilitate group cooperation. Hence, in our view, as the authors of this work, new, decentralized approaches to the challenge of cooperation would be very helpful—especially if they could involve the deployment of AI. How might individuals we can call “reformers”—and, in particular, bots (i.e., AI agents)—embedded within a group and simply acting locally facilitate cooperation in a wider population? Using online experiments with hundreds of people, we find that bots can improve (and not just maintain) cooperation in human groups when the bots are allowed to adjust social ties locally and engage in a kind of “social network engineering” to address the challenges faced by reformers attempting to make the world better. In this work, we presented what is, to our knowledge, the first intervention in human groups that can actually increase, and not merely preserve, levels of cooperation found at baseline. Deploying these decentralized, simply programmed AI agents helped. Ironically, cooperation between several disciplines was needed to come to such a conclusion. This project was done by Hirokazu Shirado (Assistant Professor; the School of Computer Science at Carnegie Mellon University, http://www.shirado.net) and by Nicholas A. Christakis (Professor; Yale Institute for Network Science, with appointments in the Departments of Sociology, Medicine, and Data Science). Hirokazu was a former graduate student of Nicholas, and we worked together on the project in Human Nature Lab (Nicholas's lab, www.humannaturelab.net). We also had technical support for the requisite software for our experiments (available at breadboard.yale.edu) by software engineers in the lab, principally Mark McKnight, and Hirokazu's specifications about what the software needed to be able to do helped the software engineers to perfect their product, too.For one, each of us has made efforts to learn other disciplines. We have benefitted a lot from our interdisciplinary collaboration. Hirokazu brings engineering ideas and technological skills to the challenge of designing and performing the online social experiments. Nicholas brings theoretical and analytical frames in social sciences and biology to make our work more meaningful and relevant to the workings of our society. Together, we are interested in what we have come to call “hybrid systems” of humans and machines. We have not encountered difficulties from this interdisciplinary collaboration. For one, each of us has made efforts to learn other disciplines. Hirokazu, who was a robotic engineer, completed a PhD program in Sociology. Nicholas, who was a medical doctor and is a professor in sociology, has many collaborations, including with engineering and computer science researchers. Moreover, we work in the same field of network