Academic

Rashomon Memory: Towards Argumentation-Driven Retrieval for Multi-Perspective Agent Memory

arXiv:2604.03588v1 Announce Type: new Abstract: AI agents operating over extended time horizons accumulate experiences that serve multiple concurrent goals, and must often maintain conflicting interpretations of the same events. A concession during a client negotiation encodes as a ``trust-building investment'' for one strategic goal and a ``contractual liability'' for another. Current memory architectures assume a single correct encoding, or at best support multiple views over unified storage. We propose Rashomon Memory: an architecture where parallel goal-conditioned agents encode experiences according to their priorities and negotiate at query time through argumentation. Each perspective maintains its own ontology and knowledge graph. At retrieval, perspectives propose interpretations, critique each other's proposals using asymmetric domain knowledge, and Dung's argumentation semantics determines which proposals survive. The resulting attack graph is itself an explanation: it recor

A
Albert Sadowski, Jaros{\l}aw A. Chudziak · · 1 min read · 4 views
#cs.AI

arXiv:2604.03588v1 Announce Type: new Abstract: AI agents operating over extended time horizons accumulate experiences that serve multiple concurrent goals, and must often maintain conflicting interpretations of the same events. A concession during a client negotiation encodes as a ``trust-building investment'' for one strategic goal and a ``contractual liability'' for another. Current memory architectures assume a single correct encoding, or at best support multiple views over unified storage. We propose Rashomon Memory: an architecture where parallel goal-conditioned agents encode experiences according to their priorities and negotiate at query time through argumentation. Each perspective maintains its own ontology and knowledge graph. At retrieval, perspectives propose interpretations, critique each other's proposals using asymmetric domain knowledge, and Dung's argumentation semantics determines which proposals survive. The resulting attack graph is itself an explanation: it records which interpretation was selected, which alternatives were considered, and on what grounds they were rejected. We present a proof-of-concept showing that retrieval modes (selection, composition, conflict surfacing) emerge from attack graph topology, and that the conflict surfacing mode, where the system reports genuine disagreement rather than forcing resolution, lets decision-makers see the underlying interpretive conflict directly.

Executive Summary

The article proposes Rashomon Memory, an AI memory architecture that enables parallel goal-conditioned agents to encode experiences in their own unique ontologies and knowledge graphs. At retrieval, agents negotiate through argumentation, using asymmetric domain knowledge to critique each other's proposals. Dung's argumentation semantics determines which proposals survive, resulting in an attack graph that serves as an explanation. This approach allows for the emergence of retrieval modes, including conflict surfacing, which enables decision-makers to see underlying interpretive conflicts. The proof-of-concept demonstrates the feasibility of Rashomon Memory in capturing nuanced, multi-perspective experiences.

Key Points

  • ▸ Rashomon Memory enables parallel goal-conditioned agents to encode experiences in unique ontologies and knowledge graphs.
  • ▸ Argumentation-based retrieval allows agents to negotiate and critique each other's proposals.
  • ▸ Dung's argumentation semantics determines which proposals survive, resulting in an attack graph explanation.

Merits

Strength in Addressing Multi-Perspective Experiences

Rashomon Memory effectively captures nuanced, multi-perspective experiences that are common in real-world applications, such as client negotiations.

Demerits

Complexity and Scalability

The proposed architecture may introduce additional complexity, potentially affecting scalability and computational efficiency, particularly in large-scale applications.

Expert Commentary

Rashomon Memory represents a significant advancement in AI memory architectures, particularly in addressing the challenges of multi-perspective experiences. The proposed approach aligns with the growing recognition of the importance of nuanced, contextual understanding in AI decision-making. However, as with any innovative architecture, careful consideration of complexity, scalability, and computational efficiency is essential to ensure practical applicability. Future research should focus on evaluating the architecture's performance in various real-world scenarios and exploring strategies to mitigate potential scalability issues.

Recommendations

  • ✓ Future research should investigate the application of Rashomon Memory in diverse domains, including but not limited to, client negotiations, policy-making, and conflict resolution.
  • ✓ Developing strategies to address potential scalability issues and optimize computational efficiency will be crucial for the practical implementation of Rashomon Memory.

Sources

Original: arXiv - cs.AI

Related Articles