University of Southampton
From discretion to dependence: trust calibration and AI adoption by UK defence analysts
Abstract
dc:description.abstractThis thesis examines how expert intelligence analysts in UK Defence calibrate and revise their trust in artificial-intelligence (AI) tools, from their first encounter through early operational use, and in a contrasting setting of enforced operational dependence. Trust has been studied in controlled environments, often with non-expert participants, at a single point in time. Longitudinal, naturalistic evidence of how domain experts come to rely on AI in authentic, high-stakes work is largely absent, and to the author’s knowledge no prior study has compared AI adoption by Defence analysts across qualitatively distinct domains. The thesis asks what situated factors shape trust and adoption when expert analysts first use AI tools in complex operating environments (R1), and how trust changes with experience and across situational contexts (R2).<br/><br/>Two contrasting field studies provide our evidence. The first examined nine analysts using a machine-vision Object Detection and Recognition Overlay under discretionary adoption, in a paired pre- and post-deployment design. The second examined five Space Domain Awareness practitioners in UK Space Command’s Joint Commercial Operations Cell, operating under enforced dependence on commercial providers, through a three-phase qualitative design. Our studies contrast discretionary with enforced adoption, dyadic with triadic trust, and accuracy- with intent-graded feedback.<br/><br/>Both studies show that analyst–AI trust is multi-dimensional, not monolithic: analysts recalibrate the tool’s confidence against their boundary-condition knowledge, outcome-level trust dissociates from in-the-moment confidence, and under pressure, analysts re-ground delegation away from the tool’s self-reported competence. Where they diverge, the differences appear more consistent with the structural features of the domain, than with analyst explanations.<br/><br/>The thesis develops three cross-domain constructs – Calculative Employment, Enforced Interdependence paired with Situated Proficiency, and the Fulfilment Belief dissociation, – and generalises three propositions to cross-domain scope. Together they support a programme of Calibrated Resilience: designing, procuring, and organising for trust that tracks a tool’s actual competence. The findings and recommendations are analytically rather than statistically generalisable, and have implications for the design, adoption, and procurement of AI in Defence intelligence analysis.
Degree
thesis:*- Name dc:type.qualificationname
- Ph.D.
- Level dc:type.qualificationlevel
- doctoral
- Grantor dc:publisher.institution
- University of Southampton
- Year dc:date.issued
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Lamb, Shaun Christopher
- Advisors dc:contributor.advisor
-
- Ramchurn, Gopal
- Norman, Tim
- ORCID person:identifier.orcid
Rights
- Language dc:language
- English, eng
Identifiers
dc:identifier.*- DOI dc:identifier.doi
- https://doi.org/10.5258/SOTON/01/T166