Reproduce “Interaction, Process, Infrastructure: A Unified Framework for Human-Agent Collaboration” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yun Wang, Yan Lu
Researchers at Microsoft Research Asia present a unified theoretical framework for human-agent collaboration, termed "Interaction, Process, Infrastructure." It formalizes "structural adaptation" and proposes a five-module process model to enable AI systems to dynamically interpret, inspect, and adapt the underlying structure of collaborative work, thereby addressing the fragmentation and rigidity observed in current AI tools.
Key insights
- The inability of current AI systems to explicitly represent and dynamically adapt the structure of collaborative work is a fundamental limitation, leading to implicit, local, and reactive responses to change.
- A minimal, domain-agnostic five-module model can sufficiently capture the essential facets of an evolving collaborative activity, with a dedicated "Reflection" module mediating dynamic interactions and structural changes.
- A layered architecture decoupling interaction, explicit process representation, and computational infrastructure provides a unified foundation for more flexible, continuous, and coherent human-agent collaboration.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.