Research Article
The human in the loop: a conceptual framework for understanding humanness in AI-integrated higher education
Synthesis: Cheeseman, Ndorley and Cai ask what happens to educators' sense of being human in their work when AI takes over parts of it, and answer with a four-dynamic framework and eight propositions rather than findings. Drawing on Haslam's dual-dimension model of humanness, sociomateriality and professional identity theory, they propose that AI pressure pushes educators to articulate an indispensable human core, to navigate identity disruption, to reclaim humanness through intensified relationality, and to live with "productive ambivalence" — holding positive and negative appraisals at once rather than resolving them. The framework's useful move is analytical: the teacher role is renegotiated through identity work, not through task substitution, and the cost of relational work lands on institutions rather than on individual educators' coping.
Key Findings
- Identity, not tasks, is the site of the disruption. Educators whose identity rests on capacities AI can simulate are expected to experience more disruption than those whose identity rests on relational, judgment-based and accountability work.
- An articulated human core is a buffer. Proposition 1 treats a professional identity already framed in terms of care, judgment and responsibility as a resource available at the first encounter with AI capability.
- Disruption is existential, not task-level. The framework expects AI integration to trigger questioning of professional meaning and value beyond adjustments to particular duties.
- Disrupted educators are expected to intensify relational work. Mentorship, discussion and modeling of intellectual struggle are proposed as the route back to recognized grounds of value.
- Relational intensification is a cost institutions carry. Proposition 5 expects positive outcomes where staffing, workload and promotion criteria support relational work, and burnout where they do not.
- Productive ambivalence is defined precisely. It requires simultaneity of positive and negative appraisal, reflexive engagement, and sustained task-specific decisions, including reasoned decisions not to use AI.
- The propositions come with falsification conditions. Each is paired with an operationalization and with what evidence would weaken it, which is unusual for a conceptual framework.
A framework built to be tested
The paper's eight propositions specify a sequence: articulating a human core buffers identity disruption, disruption triggers intensified relationality, and productive ambivalence is associated with more favorable integration outcomes after accounting for disruption. The authors call this one theoretically plausible ordering among several and specify at least three competing causal orders that empirical work would need to adjudicate.
Each proposition carries an operationalization. Relational practice could be observed through time-use diaries, supervision and office-hour records, student-reported mentoring or coded teaching documents, compared before and after AI integration and between educators reporting high and low disruption. Ambivalence would be assessed through within-person designs that measure coexisting positive and negative appraisals in the same educator, rather than assuming a single attitude per person.
The empirical section is handled with similar care. The authors review faculty perception studies, technology-anxiety evidence, qualitative work on complementarity and meta-analytic acceptance findings, then state plainly what that literature does not show: it documents attitudes and correlates, not the identity processes the framework proposes.
The institutional argument
The framework's practical weight sits in Proposition 5 and its implications. If relational intensification is how educators reclaim professional value, and relational work is invisible in workload models and promotion criteria, then the framework predicts burnout as a structural outcome rather than a personal failing. The authors are explicit that this should not be attributed to individual educators' response to AI.
For faculty development the implication is that AI training aimed at tool competence addresses the smallest part of the problem. What the framework says educators need is a language for what remains distinctively theirs, and organizational recognition when they act on it.
For administrators, the claim is that procedural measures alone are insufficient: supporting professional identity requires institutional investment in staffing, class size and the recognition of relational work. That is a budgeting argument derived from an identity theory, and the paper is honest that it is a derivation rather than a finding.
What this means for practice
- Faculty developers: open AI professional development with the identity question — what in your work is not substitutable — before tool training. The framework treats that articulation as the buffer against disruption.
- Instructors: treat ambivalence as a professional posture rather than a stage to get past. Holding both appraisals while making task-specific decisions is the model's definition of a considered position.
- Administrators: count relational work. Office hours, mentorship and modeling of intellectual struggle are the activities the framework expects educators to intensify, and they are the ones workload models usually leave out.
- Department chairs: watch for workload-driven burnout after AI adoption rather than treating resistance as the problem; the framework locates the risk in unsupported relational intensity.
- Researchers: use the operationalizations as written. Each proposition names the design and the evidence that would weaken it, which makes the framework unusually easy to test and easy to falsify.
Limitations
- The paper is conceptual: eight propositions, no data, and no evidence that the proposed sequence holds in any institution.
- The propositions require operationalization across institutional, disciplinary, career-stage and national contexts, and the authors state that generalization is an open boundary condition rather than an assumption.
- "AI integration" is treated as an undifferentiated pressure; the authors call for finer-grained typologies of AI applications in future work.
- The literature engaged is not exhaustive, and the treatment of emotional labor and academic identity is partial.
- Competing causal orderings among the four dynamics cannot be adjudicated without longitudinal or qualitative designs, none of which is reported here.
Citation
Cheeseman, V. H., Ndorley, J. S., & Cai, B. (2026). The human in the loop: A conceptual framework for understanding humanness in AI-integrated higher education. Frontiers in Education, 11.