Metraipolis › archive › dc_ca48645169e2

Beliefs in feedback: Factored representations of interacting agents in transformers — arXiv:2611.02047

Source: arXiv (en)
Original: https://arxiv.org/abs/2611.02047
License (as archived): CC0-1.0
Archived: 2026-11-03T10:49:31.895Z
SHA-256 of the archived text: 586711d87e31719c1ba4738006f716af2efaa5cba3b25cf1783865ce00ae5467
Stable URL of this page: https://metraipolis.com/documento/dc_ca48645169e2

How to cite

Beliefs in feedback: Factored representations of interacting agents in transformers — arXiv:2611.02047 — arXiv (en), https://arxiv.org/abs/2611.02047 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_ca48645169e2 (Metraipolis archive dc_ca48645169e2, sha256 586711d87e31719c1ba4738006f716af2efaa5cba3b25cf1783865ce00ae5467)

BibTeX

@misc{dcca48645169e2,
  title = {Beliefs in feedback: Factored representations of interacting agents in transformers — arXiv:2611.02047},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2611.02047},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_ca48645169e2; sha256 586711d87e31719c1ba4738006f716af2efaa5cba3b25cf1783865ce00ae5467},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.02047. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Complex systems, from ecosystems to brains, are usually modeled as modular components coupled through feedback rather than as indivisible wholes. Can a transformer trained on the joint behavior of a group of interacting agents be understood in the same way? In this work, we develop a computational-mechanics framework to study agents interacting in feedback loops. We formally prove that any joint stochastic process can be presented as resulting from the behavior of agents interacting in feedback, and the resulting process is generated by a hidden Markov model whose Bayesian beliefs factorize into separate components per agent. Furthermore, our experiments show that transformers trained on such feedback sequences make use of this separability, disentangling the beliefs about the underlying factors into separate subspaces of their residual stream. This representation is found to scale linearly (instead of exponentially) with the number of agents. Moreover, interventions show that these representations are causally efficacious, allowing for targeted steering of individual agents' beliefs by selectively inserting counterfactuals. However, coarse-graining the behavior is found to disrupt the factorization, and can force a higher-dimensional representation. Together, the results establish both the dimensional advantage of a factorized representation and the ability of transformers to exploit it, laying the groundwork for investigating traces of compositional predictive architecture in pretrained models.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.MA, nuovi lavori del giorno).
Categorie arXiv: Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA); Adaptation and Self-Organizing Systems (nlin.AO).
Pagina del lavoro: https://arxiv.org/abs/2611.02047 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-11-03T10:49:22.884Z dalla pagina https://arxiv.org/list/cs.MA/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).