What the study contributes to academic research & business practice
The thesis combines accounts of organisational practice with theory to examine two connected questions: who shapes the fairness of a decision, and what those people understand fairness to mean. It develops contributions to agency theory and boundary object theory, alongside proposals that still need development and testing.
Explaining the organisational reality
Organisations can make sincere commitments while their everyday arrangements pull in another direction. Different functions face different expectations, measures of success and pressures; a policy may be adopted without the resources or connections needed to change practice.
The thesis draws on work about institutional logics, decoupling and gaps between knowing and doing to explain these conditions. These are established theoretical resources, not theories invented by this research.
They help move the discussion beyond a choice between blaming individuals and assuming another formal process will solve the problem. The question becomes how the organisational conditions work, and what happens when AI formalises choices within them.
Bringing upstream influence into the relational analysis of AI control
Agency theory examines relationships in which one party acts on another’s behalf, including how their actions can be governed. The thesis develops that analysis to include upstream actors whose choices affect a decision without passing through human judgement within the decision process.
I call the diagnostic question the “unmediated influence test”: does a fairness-relevant choice affect the decision without being subject to human judgement within that process? This concerns review of the consequential choice, not simply whether a human originally designed the system.
Consider an illustrative situation, rather than a reported interview case. A manager reviews a shortlist, but earlier choices determine how candidates are ranked and which applications receive attention. Approving the final selection does not necessarily mean the manager has examined those earlier choices.
The contribution asks how to bring that influence into governance. It develops the use of agency-theory controls through requirements, disclosure, inspection, monitoring and routes for correction, including arrangements established when systems are acquired.
Monitoring alone is not enough if someone can identify a problem but cannot get the relevant choices reconsidered. The analysis therefore connects what people learn downstream with the authority and means to change things upstream.
This is a theoretical development informed by downstream accounts. It does not establish that these arrangements have been implemented successfully or directly verify the practices of the upstream actors concerned.
Making visible consequential differences in the meaning of fairness
Boundary object theory helps explain how people can coordinate through a shared document, tool or idea while interpreting it differently. Complete agreement is not always necessary for people to work together.
The thesis examines a condition in which that flexibility can become a problem: differences in interpretation matter to the outcome, but the shared language or artefact allows them to remain unnoticed. I describe this as the “consequential difference condition”.
Fairness is a useful setting in which to examine it. People may use the same term while referring to contextual judgement, consistent rules, organisational obligations or technical measures. Those differences are not automatically mistakes, but they may become important when the interpretations are expected to guide the same decision.
The thesis proposes the Fairness Consistency and Consensus Framework, or FCCF, to make such differences visible. It uses a four-part analytical structure to compare understandings across philosophical, societal, business and technical perspectives, without claiming these are the only possible ways of understanding fairness.
Despite the framework’s name, its purpose is not to force everyone to agree. Making the differences visible allows an organisation to decide whether to align, negotiate or accept them.
The FCCF is a design specification, not a completed or validated tool. Whether a deliberately designed artefact can support the intended coordination is an empirical question that remains open.
What AI Changes
The wider argument concerns more than the reproduction of historical bias. Organisations also make deliberate choices about priorities, evidence and acceptable trade-offs. Turning those choices into parameters and thresholds changes how they operate and who can see or question them.
Likewise, turning a contextual assessment into a digital representation can make comparison possible while leaving some information and meaning behind. The thesis examines these transformations alongside the distinction between carrying out a task and exercising judgement about what is appropriate.
This does not mean people always judge well or that computation has no place in the decision. It asks what is being delegated, what is changed by the delegation, and what meaningful control would require.
Where the research goes next
The two contributions connect. Identifying who shapes a decision is incomplete without understanding what those actors are trying to achieve; exposing differences in meaning is insufficient if nobody can act on them.
Further inquiry includes:
Investigating upstream practices directly:
How developers, suppliers and procurers make and negotiate the relevant choices.
Following attempts to change a decision process:
Whether problems recognised downstream can reach people with the authority and means to respond.
Developing and testing the proposed framework:
Whether it makes consequential differences visible and supports useful discussion without requiring artificial consensus.
Exploring other settings:
How the arguments apply beyond the recruitment and promotion contexts studied, and what would need to change.
I welcome discussion of these arguments and opportunities to investigate them with others. Get in touch, read the findings, or explore the study’s methods and limits.