Social Machines
Understanding and empowering human networks.
The Laboratory for Social Machines is a new initiative, based at the Media Lab, focusing on the development of new technologies to make sense of semantic and social patterns across the broad span of public mass media, social media, data streams, and digital content. Pattern discovery and data visualization will be explored to reveal interaction patterns and shared interests in relevant social systems, while collaborative tools and mobile apps will be developed to enable new forms of public communication and social organization. A main goal for Social Machines will be to create new platforms for both individuals and institutions to identify, discuss, and act on pressing societal problems.

Research Projects

  • AINA: Aerial Imaging and Network Analysis

    Deb Roy and Neo (Mostafa) Mohsenvand

    This project is aimed at building a machine learning pipeline that will discover and predict links between the visible structure of villages and cities (using satellite and aerial imaging) and their inhabiting social networks. The goal is to estimate digitally invisible villages in India and Sub-Saharan Africa. By estimating the social structure of these communities, our goal is to enable targeted intervention and optimized distribution of information, education technologies, goods, and medical aid. Currently, this pipeline is implemented using a GPU-powered Deep Learning system. It is able to detect buildings and roads and provide detailed information about the organization of the villages. The output will be used to construct probabilistic models of the underlying social network of the village. Moreover, it will provide information on the population, distribution of wealth, rate and direction of development (when longitudinal imaging data is available), and disaster profile of the village.

  • Human Atlas

    Deb Roy, Martin Saveski, Soroush Vosoughi and Eric Chu

    This project aims to map and analyze the publicly knowable social connections of various communities, allowing us to gain unprecedented insights about the social dynamics in such communities. Most analyses of this sort map online social networks, such as Twitter, Facebook, or LinkedIn. While these networks encode important aspects of our lives (e.g., our professional connections) they fail to capture many real-world relationships. Most of these relationships are, in fact, public and known to the community members. By mapping this publicly knowable graph, we get a unique view of the community that allows us to gain deeper understanding of its social dynamics. To this end, we built a web-based tool that is simple, easy to use, and allows the community to map itself. Our goal is to deploy this tool in communities of different sizes, including the Media Lab community and the Spanish town of Jun.

  • Journalism Mapping and Analytics Project (JMAP)

    Deb Roy, Sophie Chou, Pau Kung, Neo Mohsenvand, William Powers

    Over the last two decades, digital technologies have flattened old hierarchies in the news business and opened the conversation to a multitude of new voices. To help comprehend this promising but chaotic new public sphere, we're building a "social news machine" that will provide a structured view of the place where journalism meets social media. The basis of our project is a two-headed data ingest. On one side, all the news published online 24/7 by a sample group of influential US media outlets. On the other, all Twitter comments of the journalists who produced the stories. The two streams will be joined through network analysis and algorithmic inference. In future work we plan to expand the analysis to include all the journalism produced by major news outlets and the overall public response on Twitter, shedding new light on such issues as bias, originality, credibility, and impact.

  • Playful Words

    Ivan Sysoev, Anneli Hershman, Mina Soltangheis and Deb Roy

    While there are a number of literacy technology solutions developed for individuals, the role of social--or networked--literacy learning is less explored. We believe that literacy is inherently a social activity that is best learned within a supportive community network, including peers, teachers, and parents. By designing an approach that is child-driven and machine-guided, we hope to empower human learning networks in order to establish an engaging and effective medium for literacy development while enhancing personal, creative, and expressive interactions within communities. We are planning to pilot and deploy our system initially in the Boston area, with eventual focus on low-income families where the need for literacy support is particularly acute. We aim to create a cross-age peer-tutoring program to engage students from different communities in socially collaborative, self-expressive, and playful literacy learning opportunities via mobile devices.

  • Responsive Communities: Pilot Project in Jun, Spain

    Deb Roy, Martin Saveski, William Powers

    To gain insights into how digital technologies can make local governments more responsive and deepen citizen engagement, we are studying the Spanish town of Jun (population 3,500). For the last four years, Jun has been using Twitter as its principal medium for citizen-government communication. We are mapping the resulting social networks and analyzing the dynamics of the Twitter interactions, in order to better understand the initiative's impact on the town. Our long-term goal is to determine whether the system can be replicated at scale in larger communities, perhaps even major cities.

  • Rumor Gauge: Automatic Detection and Verification of Rumors in Twitter

    Soroush Vosoughi and Deb Roy

    The spread of malicious or accidental misinformation in social media, especially in time-sensitive situations such as real-world emergencies, can have harmful effects on individuals and society. Motivated by this, we are creating computational models of false and true information on Twitter to investigate the nature of rumors surrounding real-world events. These models take into account the content, characteristics of the people involved, and virality of information to predict veracity. The models have been trained and evaluated on several real-world events, such as the 2013 Boston Marathon bombings, the 2014 Ferguson riots, and the Ebola epidemic, with promising results. We believe our system will have immediate real-world applications for consumers of news, journalists, and emergency services, and that it can help minimize and dampen the impact of misinformation.

  • The Electome: Measuring Responsiveness in the 2016 Election

    Deb Roy, Russell Stevens, Soroush Vosoughi, William Powers, Sophie Chou, Perng-Hwa Kung, Neo (Mostafa) Mohsenvand, Raphael Schaad and Prashanth Vijayaraghavan

    The Electome project is a comprehensive mapping of the content and network connections among the campaign's three core public sphere voices: candidates (and their advocates), media (journalists and other mainstream media sources), and the public. This mapping is used to trace the election's narratives as they form, spread, morph, and decline among these three groups–-identifying who and what influences these dynamics. We are also developing metrics that measure narrative alignment and misalignment among the groups, sub-groups (political party, media channel, advocacy group, etc.), and specific individuals/organizationss (officials, outlets, journalists, influencers, sources, etc.). The Electome can be used to promote more responsive elections by deploying analyses, metrics, and data samples that improve the exchange of ideas among candidates, the media, and/or the public in the public sphere of an election.

  • The Foodome: Building a Comprehensive Knowledge Graph of Food

    Deb Roy, Russell Stevens, Neo (Mostafa) Mohsenvand, Prashanth Vijayaraghavan, Soroush Vosoughi and Guolong Wang

    The Foodome addresses how to create deeper understanding and predictive intelligence about the relationships between how we talk and learn about food, and what we actually eat. Our aim is to build a food learning machine that comprehensively maps, for any given food, its form, function, production, distribution, marketing, science, policy, history, and culture (as well as the connections among all of these aspects). We are gathering and organizing a wide variety of data, including news/social content, recipes and menus, and sourcing and purchase information. We then use human-machine learning to uncover patterns within and among the heterogeneous food-related data. Long term, the Foodome is meant to help improve our understanding of, access to, and trust in food that is good for us; find new connections between food and health; and even predict impacts of local and global events on food.

  • Visible Communities

    Raphael Schaad, Michael Koehrsen and Deb Roy

    Better maps and local knowledge increase the efficiency and effectiveness of community health workers (CHW) in developing countries. In a post-Ebola world, the World Health Organization and the United Nations have elevated the priority of developing such CHW systems. However, commercial map services do not always reach these regions of the world. We are developing mapping based on satellite analytics of household- and road-level data in rural areas, coupled with a mobile app with a navigation interface specifically designed for CHWs to help them work better. Their bottom-up annotations will continuously improve the machine-learning analytics of the top-down satellite maps.