Assessing What Matters Most: The Shifting Value Proposition of Human-Centred Work in Assessment Practice

Assessment in higher education has been responding to generative AI with tighter AI Regulation in Education and stronger detection, a solutionist reflex that misreads a wicked problem as a technical one. Elkington argues that assessment should be reconceived as a sociomaterial assemblage of people, tools, spaces and norms, and offers three design vocabularies, assemblage thinking, participatory multimodal assessment and generative constraints, for keeping judgment, meaning making and relational work at the center.

Key Findings

• Regulation is aimed at the wrong object. Surveillance, restrictive invigilation and ever more elaborate exclusion mechanisms treat AI as something to be kept out of assessment, when the epistemic, cognitive and procedural conditions of practice have already changed, and the question is how to guide pedagogy in an environment where AI is an active participant.
• The tensions are structural, not peripheral. Integrity, authenticity and equity are core contradictions of assessment practice, and prohibition-based responses can disadvantage precisely the students who rely on AI for Accessibility, language support or cognitive Scaffolding.
• Thévenot's regimes of engagement explain why consensus is hard to reach. Educators coordinate through justification, planning, familiarity and exploration, and speak from different grammars of commonality, so argument about AI in assessment is often two people defending different kinds of common good.
• Assessment is located in the assemblage, not in the task. A rubric or reflective journal intends judgment, but what a student actually does depends on interfaces, networks, physical space, peer norms and prior history, so designing the task is not the same as designing the conditions for judgment.
• Constraints are the design lever, not the obstacle. Generative constraints, composed deliberately, pace cognitive work, create low-stakes space for exploration, sustain relational routines, and make criteria explicit, rather than being limits a designer should remove to maximize freedom.

Why tighter regulation misreads the problem

The article opens against the sector's dominant response: intensified surveillance, more restrictive invigilation, and mechanisms designed to exclude, bracket out or circumvent AI use. Such measures assume integrity can be preserved by regulatory fortification, and that assumption is what the argument sets out to displace. Elkington draws on work characterising GenAI in assessment not as a discrete technical problem but as a wicked one, resistant to definitive solutions, involving persistent trade-offs, and demanding ongoing professional judgment. Wicked problems are not solved by prescriptive toolkits or uniform institutional policy; they require iterative adaptation, tolerance for ambiguity, and the negotiation of competing values.

Long-standing tensions then become sharper rather than resolved. Validity and integrity are no longer confined to misconduct, because generative technologies mediate, augment or co-produce student work, which raises the question of what counts as legitimate evidence of learning at all. Authenticity, traditionally anchored in individual authorship and demonstrated mastery, is unsettled by students who routinely work within human-AI assemblages. And equity cuts both ways: differential access to tools and uneven institutional support risk widening existing inequalities, while securitised assessment may penalize students who need AI for accessibility or language support. The argument's practical consequence is that a solutionist orientation keeps producing interventions aimed at the wrong object, and that what is required is not tighter regulation but new regimes of engagement.

Evaluation as coordination: regimes of engagement

Elkington's theoretical apparatus is Thévenot's sociology of regimes of engagement, in which people coordinate with their environments through distinct modes of involvement: justification, planning, familiarity and exploration. These are not abstract ideals but practical investments through which people orient themselves and act meaningfully. Thévenot's later grammars of commonality add a second vocabulary for how people articulate shared concerns, ranging from intimate, situated affiliation to public, codified and universalist justification.

This is what makes the AI argument more than a restatement of professional autonomy. Educators find themselves pulled between relational commitments developed with students in everyday teaching and institutional grammars demanding accountability, standardization and demonstrable outcomes, and AI intensifies that conflict because institutions respond with security and procedural standards that may sit uneasily beside professional judgment. Because colleagues are often speaking from different grammars, forming a shared sense of what assessment is for becomes the hard part, and the work required is organizational and relational rather than technical. The article is explicit that this reconceptualisation does not abandon standards or diminish rigor; the claim is that human-AI configurations need new forms of coordination so that the educational purposes of assessment survive the shift.

The assemblage, and what it asks a designer to notice

Assemblage thinking is introduced as the means of making that position operational, and Elkington is careful about what it is not: not a loose metaphor for complexity, and not a descriptive catalogue of interacting elements. An assemblage is a purposeful arrangement of heterogeneous components, people, tools, spaces, discourses and histories, whose relations give rise to practice and which is continually reformed through local action.

Applied to assessment, the consequence is that practice is not located in the task. A reflective journal and a tailored rubric may intend critical judgment, but the student's engagement depends on the interfaces they use, the networks they consult, the physical spaces available, the tacit norms circulating among peers, and the histories they bring. Assemblage thinking therefore shifts the designer's question from what the assessment task is to how the configuration of tools, spaces, norms and relations invites or inhibits the attentional, interpretive, ethical and relational work assessment must support. It also connects back to the regimes: a well-designed rubric invites justification, but if a platform obscures the purpose of the work, if the space discourages dialogue, or if time pressures crowd out relational connection, the assemblage tilts students away from familiarity and into reactive strategic action, displacing the orientation the educator intended.

The article's first set of design nudges follows from this: which material and digital elements currently shape how students interact with the assessment; how reconfiguring the sociomaterial environment would better support the thinking the assessment intends; and which aspects of the physical or virtual environment could unintentionally privilege some students over others. Its equity point is the sharpest of the three, since it treats the environment itself, not student capability, as the thing to redesign.

Modality choice and generative constraints

Two further design vocabularies complete the argument. The first is participatory Multimodal AI assessment. Drawing on work arguing that choosing a modality is never neutral, Elkington treats modality as an educational judgment that shapes what becomes possible for learners, what becomes visible to assessors, and what becomes meaningful in shared academic work. Multimodal artifacts engage students with semiotic resources, image, sound, movement, spatial design, color and narrative structure, that print-based frameworks struggle to recognize, and the article adopts four dimensions proposed for evaluating such work: criticality, cultivating Creativity, holism, and valuing multimodality.

Its strongest claim here is that multimodal artifacts stage the transitions between regimes rather than merely expressing learning: composing across modes requires moving between exploratory involvement, familiar involvement and public justification, which makes the artifact a mediating device that stabilizes shared purposes and shows students not only what is valued but why. That is also where AI can be made pedagogically meaningful rather than merely policed, because students can interrogate generated material, incorporate it into evolving artifacts, reflect on its limitations, and situate their own judgment in relation to machine suggestions within the composition itself.

The second vocabulary is generative constraints, the technologies, resources and activities an educator orchestrates to guide an assessment experience and capture creative demonstrations of learning. The article is emphatic that this is not about removing constraints to maximize freedom but about selecting and composing them: a short visual draft before a final film creates a checkpoint for formative dialogue and paces creative work, and limiting a documentary to three interview clips forces defensible editorial decisions and narrative coherence. Controlled choice among modalities is itself a generative constraint, sustaining creativity while keeping structure and comparability. Mapped onto the regimes, constraints stabilize planning through structured pathways, support exploration through low-stakes spaces such as storyboards and audio sketches, enable familiarity through predictable routines and relational encounters, and support justification by making criteria and public standards explicit.

What this means for practice

• Instructors. Audit the assemblage before redesigning the task: name the platforms, spaces, peer norms and time pressures that currently shape how students do this assessment, and ask which of them, not which students, need changing.
• Assessment designers. Justify the modality pedagogically rather than by habit, and treat oral, written, digital and hybrid options as different conditions for expressing judgment instead of interchangeable containers.
• Assessment designers. Compose constraints deliberately, draft checkpoints, clip or format limits, controlled choice among modes, so that limits pace cognitive work and create space for Feedback rather than simply restricting output.
• Faculty developers. Read prohibition and heavy securitisation as an equity decision, since it can disadvantage students relying on AI for accessibility or language support, and put that trade-off in front of the committee that approves assessment policy.
• Administrators. Fund professional judgment and room to iterate rather than imposing uniform policy, because a wicked problem is not solved by compliance and the flexibility to adapt is the capability being asked for.

Limitations

• The article is conceptual: it advances a framework and design nudges, and reports no empirical study, no student outcomes and no evaluation of the designs it proposes.
• Its central check on use is professional judgment, so a reader looking for evidence that the nudges improve learning will not find it here, and the argument should be treated as a design vocabulary to test rather than a validated intervention.
• The account is grounded in higher education, and the transfer of the assemblage framing to schooling, vocational or professional settings is not examined.
• It argues from Thévenot's sociology of engagement and does not weigh rival explanations, such as institutional-theory or policy-implementation accounts, that could explain the same regulatory response.
• Effective detection and its limits are discussed as a policy posture rather than analyzed on their own terms, so the case against surveillance is made on conceptual rather than measurement grounds.

Citation

Elkington, S. (2026). Assessing what matters most – the shifting value proposition of human-centred work in assessment practice. Teaching in Higher Education, 31(8), 1748-1765.