Complementary read to our other recent paper on complex contagion
www.pnas.org/doi/10.1073/...

New paper out in Sociological Science sociologicalscience.com/articles-v12...

Across three different analytical methods we provide evidence for complex contagion: the contagion process cannot be understood as independent cascades but rather as a process in which signals from multiple sources amplify each other through synergistic interdependence

arxiv.org/abs/2510.05174

Emergent properties have functional benefits and realign internal group structure. Results mirror collective intelligence principles: effective performance requires both alignment on shared objectives and complementary contributions across members.

We find multi-agent systems have capacity for emergence - they are real "teams" that are more than the sum of their parts. And we can steer them with clever prompts. The ToM condition in particular leads to stable specialization and goal-directed complementarity across agents

Experiments use a simple guessing game without direct agent communication and only minimal group-level feedback with three randomized interventions: plain, agents with personas, and personas with an instruction to "think about what other agents might do" (a ToM prompt)

osf.io/preprints/ps...

New paper on quantifying human-AI synergy in a large dataset with AI-alone, human-alone, and human-AI together.

aka people with higher theory of mind write better prompts and get better answers from AI. AI on the other hand is quite bad responding to moment-to-moment fluctuations in ToM

Crucially we find that solo-ability and collaboration-ability are distinct traits. And it turns out social skills (theory of mind) predict AI collaboration ability AND AI response quality.

The basic pattern is a tradeoff: reaching more people via random ties or exploit social reinforcement via clustered ties. Paper with @allisonwan.bsky.social @davidlazer.bsky.social @criedl.bsky.social (7/7)

Clustered networks are even less advantageous when individuals in the network are connected to more people, can influence their neighbors for longer periods of time, or when they require more adopting neighbors to benefit for social reinforcement (6/7)

While faster spread on clustered networks is possible it does not represent the dominant pattern. Social reinforcement is necessary but insufficient condition for clustered network to diffuse better but faster spread on clustered networks is no test for social reinforcement (5/7)

Even with strong social reinforcement random networks spread behavior better than clustered networks. Clustered networks outspread random networks in only a small region of the space we model and mostly only when behavior is near deterministic (4/7)

We develop a novel model of behavior diffusion with tunable probabilistic adoption and social reinforcement parameters that contains many prior diffusion models as special cases (3/7)

Complex contagion theory suggests socially reinforced behaviors spread more on clustered networks. But when spread is modeled with realistic probabilistic adoption, in most cases behaviors spread equally-if not better-on random networks (2/7)