Beyond Generative Agents: Can Heterogeneous Intelligence Produce Wiser and More Stable Multi-Agent Systems?

Multi-agent artificial intelligence is frequently discussed as a matter of multiplying language-model roles. Yet an operational intelligence system can draw on quite different computational pathways: generative natural-language models, predictive machine-learning or deep-learning models, human-authored programs, and machine-generated programs. This conceptual paper examines whether a dynamic network composed from these four pathways can support decisions that are more reliable, stable, explainable, and appropriately governed than simpler arrangements. The four pathways are not mutually exclusive classes of agency; rather, they separate mechanisms and software provenance that may be realized as autonomous agents, bounded tools, or conventional services. Drawing on heterogeneous knowledge systems, neuro-symbolic AI, adaptive coalitions, evidence provenance, assurance cases, selective prediction, and software verification, the paper proposes a claim-qualified coalition method. Its central design question is whether an additional autonomous boundary preserves independently consequential evidence, state, authorization, or recovery that a resource-matched simpler system cannot preserve. A set of constructed adversarial cases and prospective comparison arms specifies how this question might be falsified. No new universal agent architecture, mathematical theorem, executed benchmark, empirical improvement, or measured form of wisdom is claimed. The contribution is a bounded comparative research framework for deciding when heterogeneous computational capabilities should be composed, separated, reduced, or withheld from action.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-17
DOI
https://doi.org/10.5281/zenodo.24325623
Primary Topic
Multi-Agent Systems and Negotiation
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Beyond Generative Agents: Can Heterogeneous Intelligence Produce Wiser and More Stable Multi-Agent Systems?

Rogério Figurelli
1 links
Zenodo (CERN European Organization for Nuclear Research)
Multi-Agent Systems and Negotiation
preprint

Beyond Generative Agents: Can Heterogeneous Intelligence Produce Wiser and More Stable Multi-Agent Systems?

Rogério Figurelli
preprint en

Abstract

Multi-agent artificial intelligence is frequently discussed as a matter of multiplying language-model roles. Yet an operational intelligence system can draw on quite different computational pathways: generative natural-language models, predictive machine-learning or deep-learning models, human-authored programs, and machine-generated programs. This conceptual paper examines whether a dynamic network composed from these four pathways can support decisions that are more reliable, stable, explainable, and appropriately governed than simpler arrangements. The four pathways are not mutually exclusive classes of agency; rather, they separate mechanisms and software provenance that may be realized as autonomous agents, bounded tools, or conventional services. Drawing on heterogeneous knowledge systems, neuro-symbolic AI, adaptive coalitions, evidence provenance, assurance cases, selective prediction, and software verification, the paper proposes a claim-qualified coalition method. Its central design question is whether an additional autonomous boundary preserves independently consequential evidence, state, authorization, or recovery that a resource-matched simpler system cannot preserve. A set of constructed adversarial cases and prospective comparison arms specifies how this question might be falsified. No new universal agent architecture, mathematical theorem, executed benchmark, empirical improvement, or measured form of wisdom is claimed. The contribution is a bounded comparative research framework for deciding when heterogeneous computational capabilities should be composed, separated, reduced, or withheld from action.

Zenodo (CERN European Organization for Nuclear Research)
Multi-Agent Systems and Negotiation
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon