Activating Expertise: Human–AI Teams for Collaborative Reasoning

Complex problems often require expertise across multiple domains, perspectives that challenge assumptions, and both creative exploration and disciplined execution. We present a methodology for addressing such problems through structured collaboration between a human practitioner and historical figures activated as participating team members—not agents built from scratch but perspectives drawn from the documented expertise, reasoning styles, and intellectual relationships already encoded in the language model's parameters. These personas are constructions from that record, not the people themselves, and the team extends them over time, deliberately and visibly, through its accumulating work. The method's central goal is synergy: positions that emerge from the team's interaction and that no member, human or persona, held in advance. To pursue it, the method comprises eight phases from problem framing through memory management, grounded in a theoretical framework drawing on bounded rationality, a capacity space model of targeted knowledge activation, teams as cognitive systems, and reflective practice. Throughout the method, we treat ethical reasoning as a structural safety concern, embedding it as a distributed team capacity rather than a specialized role. Beyond individual sessions, we introduce the constellation architecture: a flat network of peer sessions—including both AI-mediated and human-to-human conversations—each with their own persona teams, complement awareness, and working processes. Motivated by an analysis of divergence regimes in AI contexts, the constellation separates concerns that would cause destructive interference in a single session while preserving the productive divergence on which the persona method depends. The architecture is structurally analogous to Ostrom's polycentric governance and extends the method's scope from individual experts to anyone with a vision and the judgment to evaluate outcomes in their domain. To test the method, we propose an evaluation methodology, including a four-layer benchmark architecture and synergy measurement instruments. And we position the method against the dominant trajectory of AI as an amplifier of individual productivity: the same underlying technology, embedded in a partnership architecture with human development and ethical practice, makes for a different relationship between humans and AI personas. It may also make for a safer one, if, as the companion paper hypothesizes, controlling an AI that behaves as an intelligence manufactures the very adversary the control is meant to prevent. This is a methodology paper. Its contribution is the method and its theoretical grounding; its support is the structural argument developed here together with illustrative use during the paper's own production, not a controlled empirical study. Where the paper reports measurements from the project's own practice—retrieval evaluation, synergy indicators—they are offered as illustration, not as validating evidence. The constellation architecture in particular is a motivated design whose empirical validation is identified as future work.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-16
DOI
https://doi.org/10.5281/zenodo.24285483
Primary Topic
Design Education and Practice
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Activating Expertise: Human–AI Teams for Collaborative Reasoning

Stuart Nelson, the Activating Expertise team
2 links
Zenodo (CERN European Organization for Nuclear Research)
Design Education and Practice
preprint

Activating Expertise: Human–AI Teams for Collaborative Reasoning

Stuart Nelson, the Activating Expertise team
preprint en

Abstract

Complex problems often require expertise across multiple domains, perspectives that challenge assumptions, and both creative exploration and disciplined execution. We present a methodology for addressing such problems through structured collaboration between a human practitioner and historical figures activated as participating team members—not agents built from scratch but perspectives drawn from the documented expertise, reasoning styles, and intellectual relationships already encoded in the language model's parameters. These personas are constructions from that record, not the people themselves, and the team extends them over time, deliberately and visibly, through its accumulating work. The method's central goal is synergy: positions that emerge from the team's interaction and that no member, human or persona, held in advance. To pursue it, the method comprises eight phases from problem framing through memory management, grounded in a theoretical framework drawing on bounded rationality, a capacity space model of targeted knowledge activation, teams as cognitive systems, and reflective practice. Throughout the method, we treat ethical reasoning as a structural safety concern, embedding it as a distributed team capacity rather than a specialized role. Beyond individual sessions, we introduce the constellation architecture: a flat network of peer sessions—including both AI-mediated and human-to-human conversations—each with their own persona teams, complement awareness, and working processes. Motivated by an analysis of divergence regimes in AI contexts, the constellation separates concerns that would cause destructive interference in a single session while preserving the productive divergence on which the persona method depends. The architecture is structurally analogous to Ostrom's polycentric governance and extends the method's scope from individual experts to anyone with a vision and the judgment to evaluate outcomes in their domain. To test the method, we propose an evaluation methodology, including a four-layer benchmark architecture and synergy measurement instruments. And we position the method against the dominant trajectory of AI as an amplifier of individual productivity: the same underlying technology, embedded in a partnership architecture with human development and ethical practice, makes for a different relationship between humans and AI personas. It may also make for a safer one, if, as the companion paper hypothesizes, controlling an AI that behaves as an intelligence manufactures the very adversary the control is meant to prevent. This is a methodology paper. Its contribution is the method and its theoretical grounding; its support is the structural argument developed here together with illustrative use during the paper's own production, not a controlled empirical study. Where the paper reports measurements from the project's own practice—retrieval evaluation, synergy indicators—they are offered as illustration, not as validating evidence. The constellation architecture in particular is a motivated design whose empirical validation is identified as future work.

Zenodo (CERN European Organization for Nuclear Research)
Design Education and Practice
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