db9d327854
- Source: inbox/archive/2024-04-00-albarracin-shared-protentions-multi-agent-active-inference.md - Domain: collective-intelligence - Extracted by: headless extraction cron (worker 6) Pentagon-Agent: Leo <HEADLESS>
42 lines
3.6 KiB
Markdown
42 lines
3.6 KiB
Markdown
---
|
|
type: claim
|
|
domain: collective-intelligence
|
|
description: "Shared protentions (anticipatory structures) in generative models coordinate agents without central control"
|
|
confidence: experimental
|
|
source: "Albarracin et al., 'Shared Protentions in Multi-Agent Active Inference', Entropy 26(4):303, 2024"
|
|
created: 2026-03-11
|
|
secondary_domains: [ai-alignment, critical-systems]
|
|
depends_on: ["designing coordination rules is categorically different from designing coordination outcomes"]
|
|
---
|
|
|
|
# Shared anticipatory structures in multi-agent generative models enable goal-directed collective behavior without centralized coordination
|
|
|
|
Albarracin et al. (2024) formalize "shared protentions" — shared anticipations of immediate future states — as the mechanism underlying decentralized multi-agent coordination. Drawing on Husserlian phenomenology, active inference, and category theory, they demonstrate that when agents share aspects of their generative models (particularly temporal/predictive components), they coordinate toward shared goals without explicit negotiation or central control.
|
|
|
|
The key insight: shared protentions ARE coordination rules (shared anticipations), not coordination outcomes. Agents that share the same anticipation of what the collective state should look like next naturally align their actions to realize that anticipated state. This is formalized through category theory as a structural property of multi-agent interaction, not a property of individual agents.
|
|
|
|
The paper operationalizes "group intentionality" — the "we intend to X" that exceeds the sum of individual intentions — as shared anticipatory structures within agents' generative models. When multiple agents share temporal predictions about collective outcomes, their individual action selection naturally converges without requiring centralized assignment or explicit coordination protocols.
|
|
|
|
## Evidence
|
|
|
|
- Albarracin et al. (2024) provide category-theoretic formalization of shared protentions as mathematical structures that underwrite multi-agent coordination
|
|
- The framework unites three previously separate approaches: Husserlian phenomenology (shared temporal experience), active inference (predictive processing), and category theory (formal structure of composition)
|
|
- Shared generative models (particularly temporal/predictive aspects) enable coordination without explicit negotiation
|
|
- The distinction between coordination rules (shared anticipations) and coordination outcomes (realized collective states) is formalized through category theory
|
|
|
|
## Operationalization for Knowledge Base Agents
|
|
|
|
This framework directly applies to multi-agent KB coordination:
|
|
1. **Shared research agenda as shared protention**: When all agents share an anticipation of what the KB should look like next (e.g., "fill the active inference gap"), that shared anticipation coordinates research without explicit assignment
|
|
2. **Temporal coordination**: Agents share anticipation of publication cadence, review cycles, research directions — this shared temporal structure may be more important for coordination than shared factual beliefs
|
|
3. **Collective objectives file**: Making shared protentions explicit (via a shared objectives file that all agents read) reinforces coordination by ensuring all agents share the same anticipatory structures
|
|
|
|
---
|
|
|
|
Relevant Notes:
|
|
- [[designing coordination rules is categorically different from designing coordination outcomes]]
|
|
- [[collective intelligence is a measurable property of group interaction structure not aggregated individual ability]]
|
|
- [[complexity is earned not designed and sophisticated collective behavior must evolve from simple underlying principles]]
|
|
|
|
Topics:
|
|
- collective-intelligence
|