In the coming decades, a defining task for humanity will be to solve global challenges through mass coordination. To avert the most catastrophic consequences of rapid climate change, humankind needs to adopt a rapid and widespread change of lifestyle in favor of carbon-neutral practices. To curb the diffusion of pandemics, mass adherence to preventive health measures and to vaccination campaigns is essential. To make societies resilient to the massive exodus of climate refugees that is on the horizon, citizens will need to share part of their resources to avoid con ict and to grow thriving communities. These challenges can be understood as social dilemmas: the individual benefits are at odds with collective interests, and only concerted action can attain shared gains that outmatch individual costs. To resolve societal dilemmas, a key question is how to achieve the necessary collective coordination within a short time span.

The Social Web is an ideal candidate to achieve grassroots coordination: it is the largest and most pervasive network for the diffusion of culture, it allows rapid participation to the public debate at low cost, and it proved to be a fertile ground for collective action even in absence of tangible rewards. However, the key determinants of cooperation in online communities are largely unknown, and there is little experimental evidence to inform how to build online communities that facilitate cooperation rather than exacerbating polarization. Social scientists have formulated theories that link social dimensions such as knowledge exchange, group identity, and trust to successful coordination. These theories have never been verified at scale because the social dimensions they are based on were traditionally hard to quantify. However, that recently changed with the development of AI algorithms for Natural Language Processing that can infer high-level properties of social interactions from conversational text. This project proposal unfolds in the wake of this opportunity.

The goal of COCOONS is to unveil the prime elements of social interactions that enable spontaneous coordination in the face of social dilemmas. It will do so by quantifying fundamental dimensions of social interactions with cutting-edge Natural Language Processing algorithms applied to online social media conversations, and by studying how these dimensions are linked to cooperation outcomes in complex social networks.

To achieve this goal, the project will tackle four overarching research tasks:

1. Develop deep-learning models that process natural language to detect dimensions of social interactions that are theoretically relevant to the process of coordination.
2. Study which social dimensions are more effective in triggering opinion change in social media conversations.
3. Design models that can predict coordination dynamics in large groups based on their social network structure and on the social dimensions they voice.
4. Create a set of guidelines on how to design online communities for promoting grassroots cooperation and agreement, and validate them with controlled experiments.