{
  "title": "Shared Protentions in Multi-Agent Active Inference",
  "domain": "Active Inference",
  "type": "Paper",
  "creators": [
    {
      "name": "Mahault Albarracin"
    },
    {
      "name": "Riddhi J. Pitliya"
    },
    {
      "name": "Toby St. Clere Smithe"
    },
    {
      "name": "Daniel Ari Friedman"
    },
    {
      "name": "Karl Friston"
    },
    {
      "name": "Maxwell J. D. Ramstead"
    }
  ],
  "description": "We develop the concept of shared protentions—shared anticipatory states—in multi-agent Active Inference. Protentions are future-directed expectations that shape perception and action; here we formaliz...",
  "abstract": "In this paper, we unite concepts from Husserlian phenomenology, the active inference\nframework in theoretical biology, and category theory in mathematics to develop a comprehensive\nframework for understanding social action premised on shared goals. We begin with an overview\nof Husserlian phenomenology, focusing on aspects of inner time-consciousness, namely, retention,\nprimal impression, and protention. We then review active inference as a formal approach to modeling\nagent behavior based on variational (approximate Bayesian) inference. Expanding upon Husserl’s\nmodel of time consciousness, we consider collective goal-directed behavior, emphasizing shared\nprotentions among agents and their connection to the shared generative models of active inference.\nThis integrated framework aims to formalize shared goals in terms of shared protentions, and thereby\nshed light on the emergence of group intentionality. Building on this foundation, we incorporate\nmathematical tools from category theory, in particular, sheaf and topos theory, to furnish a mathe-\nmatical image of individual and group interactions within a stochastic environment. Specifically, we\nemploy morphisms between polynomial representations of individual agent models, allowing pre-\ndictions not only of their own behaviors but also those of other agents and environmental responses.\nSheaf and topos theory facilitates the construction of coherent agent worldviews and provides a way\nof representing consensus or shared understanding. We explore the emergence of shared protentions,\nbridging the phenomenology of temporal structure, multi-agent active inference systems, and cate-\ngory theory. Shared protentions are highlighted as pivotal for coordination and achieving common\nobjectives. We conclude by acknowledging the intricacies stemming from stochastic systems and\nuncertainties in realizing shared goals.",
  "keywords": [
    "shared protentions",
    "multi-agent",
    "Active Inference",
    "anticipation",
    "coupled generative models",
    "coordination",
    "predictive processing"
  ],
  "methods": [
    {
      "name": "Husserlian time-consciousness as conceptual basis for shared goals",
      "description": "Uses Husserl's retention, primal impression, and protention, extended to intersubjectivity, to cast shared goals as shared protentions.",
      "evidence": "we unpack the notion of shared goals in a social group by appealing to the construct of protention (or real-time, implicit anticipation) in Husserlian phenomenology."
    },
    {
      "name": "Polynomial-functor formalization of agent boundaries and generative models",
      "description": "Represents agents' Markov blankets as polynomial functors with stochastic feedback, using tensors and hom polynomials to model multi-agent prediction.",
      "evidence": "By using tools from category theory, namely, polynomial morphisms and hom polynomials, we are able to design agent architectures that implement a form of recursive cognition and prediction of other agents’ actions and environmental responses."
    },
    {
      "name": "Sheaf- and topos-theoretic gluing of agents' world models",
      "description": "Proposes gluing agents' spatial generative models into a sheaf of world models and their toposes into a consensus topos representing shared understanding.",
      "evidence": "In order to describe agents with such shared beliefs, we propose upgrading the formalism using the mathematical tools of sheaf and topos theory."
    }
  ],
  "key_findings": [
    "Replacing the trivial interface with a hom over all agents' blankets yields agents that predict their own and companions' actions plus the environment's response, a starting point for formal theory of mind.",
    "The authors argue that if agents' models can be glued into a sheaf of world models, the agents inhabit a shared universe and may be said to share protentions; failure to glue indicates disagreement.",
    "The paper proposes shared protentions as an emergent, possibly necessary, property of collective self-organization.",
    "The authors state the proposal is broad and does not yet provide a specific testable model, leaving that to future work."
  ],
  "related_papers": [
    "2018_WoodliceAndMen",
    "2020_BehaviorEngineering",
    "2021_ModelingConflict"
  ],
  "checked_at": "2026-07-01T20:50:01Z",
  "doi": "10.3390/e26040303",
  "doi_url": "https://doi.org/10.3390/e26040303",
  "key_findings_evidence": [
    "Such agents thus predict not only their individual actions but those of their companion agents, along with how the environment will respond to all of them.",
    "If it is possible, then we may say that the agents inhabit a shared universe, and thus, with appropriate generative models, may be said to share protentions.",
    "In this setting, shared protentions are an emergent property—and possibly a necessary property—of any collective scale of self-organization",
    "it is formulated quite broadly and does not provide a specific testable model yet. Moving towards such a specific model making testable hypotheses will be left for future follow-up work."
  ],
  "summary_provenance": {
    "source": "full_text.md",
    "date": "2026-10-01",
    "work_kind": "theoretical",
    "method": "Evidence-grounded extraction from full_text.md: every method and finding carries a verbatim quote from the paper text, mechanically verified to occur in full_text.md (whitespace/ligature-normalized) before merge.",
    "notes": "Purely theoretical/formal paper; no data were created or analyzed (per its Data Availability Statement).",
    "reviewed": "2026-10-01 independent review of every item against its quote and context; no corrections needed"
  },
  "citation_metadata": {
    "doi": "10.3390/e26040303",
    "title": "Shared Protentions in Multi-Agent Active Inference",
    "source_url": "https://api.crossref.org/works/10.3390%2Fe26040303",
    "checked_at": "2026-10-12T21:33:52Z",
    "resource_type": "journal-article",
    "journal_title": "Entropy",
    "volume": "26",
    "issue": "4",
    "pages": "303",
    "issn": "1099-4300",
    "publication_date": "2024-03-29"
  },
  "abstract_provenance": {
    "source": "publisher_pdf",
    "date": "2026-11-05",
    "kind": "complete_author_abstract",
    "url": "https://danielarifriedman.com/papers/2024_SharedProtentions/2024_SharedProtentions.pdf",
    "source_pdf": "2024_SharedProtentions.pdf",
    "source_pdf_sha256": "0b40c6a4bc0d1dd0d2a9a4597b557132b01187327389c05c5455df713905f7d3",
    "abstract_sha256": "666d4ac466b142c3c56b3254f60320ef6409c8fefd49bb72dbc3620dc5619fc2",
    "reviewed_by": "Codex (publication evidence review)",
    "basis": "Complete author-written abstract transcribed from the exact retained publisher PDF; publisher-page access was unavailable, not inferred."
  }
}
