The Insight Engine: A Research Programme for Verifiable Structural Transfer in Human–AI Engineering Problem Solving

The Insight Engine research programme asks whether a machine can identify useful changes to the representation and boundary conditions of an engineering problem, rather than only optimise solutions within the problem as first formulated. The question arose from first-person reflection on engineering practice — episodes in which an apparently necessary constraint dissolved once a different causal relationship was recognised, or an existing physical property was used in another way. This phenomenological origin motivates, but does not validate, a computational hypothesis. We propose a human–AI architecture built on explicit physical entities, relationships, uncertainty, candidate transformations, structurally retrieved analogies, and independently evaluated physical consequences. Its central methodological commitment is the distinction between an explanation offered after a solution and an auditable sequence in which a donor structure and its mapping are recorded before they generate a candidate, then tested by intervention rather than accepted because the answer happens to be correct. We report what the prototype has and has not shown. Internal tests give restricted evidence of useful cross-domain transfer under supplied reduced-physics models; an external holdout was halted at preflight when it revealed coverage limits beyond the prototype’s capability, rather than scored misleadingly. The programme is positioned against structure-mapping, analogical retrieval, qualitative physics, program-library learning, and recent mechanism-centric cross-domain retrieval. The contribution is therefore a well-posed research problem, a proposed integration of existing methods, and a falsifiable evaluation protocol with pre-registered criteria — not a claim to have produced machine insight.

Authors

Publication Details

Published
2026-11-13
DOI
https://doi.org/10.33774/coe-2026-hh0k1
Primary Topic
AI-based Problem Solving and Planning
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

The Insight Engine: A Research Programme for Verifiable Structural Transfer in Human–AI Engineering Problem Solving

Mark Cragg
2 links
AI-based Problem Solving and Planning
preprint

The Insight Engine: A Research Programme for Verifiable Structural Transfer in Human–AI Engineering Problem Solving

Mark Cragg
preprint en

Abstract

The Insight Engine research programme asks whether a machine can identify useful changes to the representation and boundary conditions of an engineering problem, rather than only optimise solutions within the problem as first formulated. The question arose from first-person reflection on engineering practice — episodes in which an apparently necessary constraint dissolved once a different causal relationship was recognised, or an existing physical property was used in another way. This phenomenological origin motivates, but does not validate, a computational hypothesis. We propose a human–AI architecture built on explicit physical entities, relationships, uncertainty, candidate transformations, structurally retrieved analogies, and independently evaluated physical consequences. Its central methodological commitment is the distinction between an explanation offered after a solution and an auditable sequence in which a donor structure and its mapping are recorded before they generate a candidate, then tested by intervention rather than accepted because the answer happens to be correct. We report what the prototype has and has not shown. Internal tests give restricted evidence of useful cross-domain transfer under supplied reduced-physics models; an external holdout was halted at preflight when it revealed coverage limits beyond the prototype’s capability, rather than scored misleadingly. The programme is positioned against structure-mapping, analogical retrieval, qualitative physics, program-library learning, and recent mechanism-centric cross-domain retrieval. The contribution is therefore a well-posed research problem, a proposed integration of existing methods, and a falsifiable evaluation protocol with pre-registered criteria — not a claim to have produced machine insight.

AI-based Problem Solving and Planning
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