The study surveyed 935 students, faculty members, academic administrators, and digital administration staff across public and private universities in Kazakhstan. Researchers collected the data through a multi-channel online questionnaire written in English and carefully translated into Russian and Kazakh using a back-translation procedure to preserve meaning across languages. After filtering out non-consent and incomplete responses, the team analyzed the remaining 935 valid cases using partial least squares structural equation modeling, or PLS-SEM, a statistical technique well suited to testing networks of direct and indirect relationships in complex organizational data.

The theoretical scaffolding draws on two influential frameworks. Dynamic Capability Theory describes how organizations sense change, seize opportunities, and reconfigure their resources to stay competitive; Sociotechnical Systems Theory emphasizes that technology, organizational processes, and human competencies are interdependent and must evolve together. Combining these lenses, the researchers built a model linking five constructs: AI Governance Readiness, Administrative Efficiency through AI, AI Fluency Orientation, Equity Perception, and Student Retention Intentions. The idea was to capture AI not as a pedagogical gadget but as integrated institutional infrastructure — the digital plumbing of modern academia.

The statistical results were striking. AI Governance Readiness strongly predicted Administrative Efficiency through AI, with a standardized path coefficient of 0.593, and predicted AI Fluency Orientation even more strongly at 0.698, both significant at well below the conventional 0.001 threshold. In plain terms, universities whose stakeholders perceive clear AI policies, ethical oversight, and coordinated decision structures also report smoother, faster, AI-supported administration and a stronger institutional orientation toward developing AI-related skills. The effect sizes ranged from medium to large, indicating relationships of genuine practical significance rather than statistical noise.

Both downstream pathways mattered for students. Administrative Efficiency through AI predicted Student Retention Intentions with a coefficient of 0.640, while AI Fluency Orientation did so at 0.626. The model explained roughly 49 percent of the variance in retention intentions, alongside 35 percent for administrative efficiency and nearly 49 percent for AI fluency. Mediation analysis confirmed that governance readiness influences retention indirectly: efficient AI-enabled administration carried an indirect effect of 0.378, and institutional AI fluency an indirect effect of 0.437. Together, these transmission mechanisms show how abstract governance arrangements translate into something students actually feel — responsive services, timely academic support, and a campus that prepares them for an AI-saturated labor market.

The most surprising finding, however, was what failed to matter. The researchers hypothesized that Equity Perception — stakeholders’ sense that AI systems are fair, transparent, and accessible — would strengthen or weaken the link between AI capabilities and retention. It did neither. The moderation coefficients were essentially zero, with p-values of 0.582 and 0.645. This non-result challenges a common assumption in the technology-adoption literature that fairness perceptions act as a boundary condition on institutional AI effectiveness, and it invites future research into why equity concerns appear not to condition these particular relationships in this context.

The measurement behind these conclusions was rigorous. Internal consistency indicators comfortably exceeded accepted thresholds, with Cronbach’s alpha at 0.874 and composite reliability at 0.867. Convergent validity was supported by an average variance extracted of 0.614, while discriminant validity held across both the Fornell-Larcker criterion and the more demanding HTMT ratio, whose maximum value of 0.881 stayed below the 0.90 cutoff. Hypothesis tests used bootstrapping with 5,000 resamples, and common method bias was checked twice — through procedural safeguards such as anonymity and construct separation, and statistically through full-collinearity diagnostics, all of which fell below the conservative 3.3 threshold.

Kazakhstan makes a compelling laboratory for this work. The country has aggressively pursued digitalization, innovation, and internationalization in its universities, yet AI implementation remains uneven, with gaps in governance readiness, infrastructure, faculty preparedness, and digital access. Earlier Kazakhstan-specific studies have documented varying digital literacy among students, evolving digital competence among master’s students, and growing behavioral engagement with tools such as ChatGPT. The new study adds a missing institutional layer, showing that the connection between digital competence initiatives and student outcomes may hinge on how coherently universities govern their AI resources in the first place.

For university leaders, the implications are concrete. The authors recommend institutionalizing AI governance frameworks with clear roles, accountability structures, and decision rights; investing in administrative digital infrastructure because efficient services demonstrably support retention; prioritizing structured AI fluency training for academic and non-academic staff; and embedding AI literacy and governance awareness into institution-wide programs to prevent fragmented, department-by-department adoption. In dynamic-capability language, AI governance is the coordinating capability through which universities sense, seize, and reconfigure — turning scattered tools into organizational intelligence.

The authors are careful about scope. The cross-sectional, self-reported design supports associations rather than causal claims, and the sample — 77.9 percent from public universities, 99 percent concentrated in Almaty, and 85.5 percent undergraduates — cannot be treated as nationally representative of Kazakhstan’s diverse higher education landscape. Still, the study marks a shift in how AI in academia is studied: away from individual tool acceptance and toward AI as institutional infrastructure, governed, felt, and ultimately consequential for whether students choose to stay. Whether the pattern holds in other digitally transforming education systems is now an open and urgent question for comparative research.

Subject of Research: AI governance readiness and its effects on administrative efficiency, AI fluency, and student retention in Kazakhstan's higher education sector

Article Title: Artificial intelligence as academic infrastructure: governance, administrative intelligence, and AI fluency in Kazakhstan’s higher education sector

Article References: Potluri, R. M., & Polozova, V. (2026). Artificial intelligence as academic infrastructure: governance, administrative intelligence, and AI fluency in Kazakhstan’s higher education sector. Frontiers in Education, 11, Article 2005440. https://doi.org/10.3389/feduc.2026.2005440

Image Credits: AI Generated

DOI: 10.3389/feduc.2026.2005440

Keywords: artificial intelligence, AI governance, higher education, Kazakhstan, student retention, administrative efficiency, AI fluency, digital transformation, PLS-SEM, sociotechnical systems, dynamic capability theory, equity perception

The research team, led by Sarah C. Keithly of the VA Puget Sound Healthcare System’s Center of Innovation for Veteran-Centered and Value-Driven Care, set out to fill a gap in the scientific record. Early studies of the VA transition documented profound disruption to care during the initial implementation phase, including clinician burnout, frustrated patients, and confusion over redefined roles. Far less was known about what happened after the dust settled. To find out, the investigators used a mix of purposive and snowball sampling to recruit direct care clinicians across a range of professions at the first five sites to adopt the new federal system. Of 104 clinicians invited, 30 — a participation rate of 28.8 percent — agreed to semi-structured interviews. The team then applied a hybrid inductive-deductive content analysis, combining templated summaries with matrix analysis, a method designed to surface both expected and unexpected patterns in qualitative data.

What emerged was a portrait of a workforce that has learned to survive a difficult technology by routing around it. The analysis generated four interlocking themes. First, ongoing EHR-related challenges continued to impede clinicians’ job performance long after go-live. Second, workarounds emerged in direct response to those persistent challenges. Third, clinicians perceived these workarounds as necessary but risky — to patients, to users themselves, and to the health care system as a whole. Fourth, and perhaps most troubling, as the underlying EHR problems persisted, the workarounds stopped feeling like temporary fixes and became routine practice. In other words, the improvisations that clinicians adopted to keep care moving were not fading away as the system matured; they were hardening into standard operating procedure.

The concept of the workaround occupies a well-studied niche in health informatics research. Prior work, including Koppel and colleagues’ influential analysis of barcode medication administration systems and subsequent scoping reviews of EHR workarounds, has shown that clinicians frequently develop alternate pathways when technology conflicts with clinical reality. These can range from harmless efficiency tricks to genuinely dangerous practices, such as documenting medication administration before it occurs or sharing login credentials to save time. The sociotechnical perspective, formalized in frameworks like Sittig and Singh’s model for studying health information technology in complex adaptive systems, holds that such behavior is rarely a simple failure of individual discipline. Instead, workarounds are signals — diagnostic clues that the interaction between technology, workflow, and organization has broken down somewhere. The new VA study treats them exactly that way, as a window into where the federal EHR is still failing its users.

The technical stakes of an EHR-to-EHR transition are considerable. Unlike a first-time implementation, a transition requires migrating years of clinical data, rebuilding decision-support rules, recreating order sets and documentation templates, and retraining a workforce that had achieved fluency in the old system. Systematic reviews of EHR transitions, including a 2023 review in the same journal, have catalogued recurring pitfalls: data loss during migration, mismatched workflows, degraded usability, and communication breakdowns. The VA’s homegrown system, known as VistA, had decades of local customization and a devoted user base, which made the switch to a commercial vendor product especially jarring. Previous mixed-methods evaluations of the transition documented clinicians fighting burnout for the first time, struggling with role-specific workflows, and fielding frustrated patients who bore the brunt of scheduling and documentation errors. The new study extends that timeline, showing that the problems did not resolve once the initial crisis passed.

The persistence of workarounds years into implementation carries specific safety implications. When a clinician devises an informal path around a system limitation, that path typically bypasses the safeguards built into the official workflow — the hard stops, the required fields, the verification steps. A workaround that saves thirty seconds on an order entry screen may also skip a drug-interaction check. One that speeds up documentation may leave critical information in a place where the next clinician cannot find it. The study’s participants were acutely aware of this tension. They described their workarounds as necessary to deliver care at all, yet simultaneously risky to patients and to the integrity of the medical record. That dual perception — beneficial and dangerous at once — is precisely what makes workarounds so difficult to manage through policy alone. Banning them without fixing the underlying problems simply pushes the improvisation further underground, where it becomes invisible to safety monitoring.

The normalization of workarounds also has implications for how health systems evaluate technology. If informal practices become routine, then the data flowing into the EHR increasingly reflects not actual clinical processes but the distorted output of patched-together workflows. Quality metrics, utilization data, and clinical decision support all depend on the assumption that documentation mirrors reality. When that assumption erodes, the organization loses visibility into its own operations. The VA researchers argue that this is why evaluating the role and impact of workarounds should be a formal part of EHR-to-EHR transitions, not an afterthought. Systematic workaround analysis, such as the SEWA framework for sociotechnical analysis of EHR workarounds, can identify which workarounds indicate design flaws that need engineering fixes, which reflect training gaps, and which reveal workflow mismatches that require organizational redesign rather than software changes.

The timing of these findings matters for the VA’s broader modernization program. The department has paused and restarted rollouts in recent years, and officials have announced early-stage planning for the next phase of federal EHR deployment while continuing improvement efforts at existing sites. The study’s conclusions suggest that the lessons from the first five sites should inform those decisions. If persistent EHR challenges continue to generate workaround reliance years after go-live, then simply extending the same implementation model to new sites risks replicating the same pattern. The authors emphasize that many of the transition’s challenges have persisted beyond the initial implementation phase, and that this persistence — not the transition itself — is what sustains the workaround culture. That distinction reframes the problem from a temporary disruption to be endured into a chronic condition to be treated.

For the wider health informatics community, the study adds an important longitudinal dimension to a literature dominated by snapshots taken during or immediately after implementation. Most organizations that swap one EHR for another — hospitals consolidating, health systems merging, national programs modernizing — face the same technical and human factors: interface usability, documentation burden, workflow fragmentation, and the sheer cognitive cost of relearning. The VA experience, documented with methodological care through information-power-guided sampling and rigorous qualitative analysis, offers a preview of what the years after a big-bang implementation can look like when problems are not fully resolved. The message from clinicians is not that the new system is unusable, but that using it as designed has become the exception rather than the rule. Closing that gap, the researchers argue, requires treating workarounds not as clinician misbehavior but as actionable intelligence — a map, drawn by the people closest to the work, of exactly where the system still needs to be fixed.

Subject of Research: Clinician workarounds and persistent challenges following the VA's transition to a federal electronic health record system

Article Title: “Most of what I do is a workaround now”: Clinician Use of Workarounds Following an EHR Transition

Article References: Keithly, S. C., Sulc, C., Cole, E., Sterling, R., Lin, Y.-L., Obara, E., Shirley, S., Wong, E. S., Rinne, S. T., & Young, J. (2026). “Most of what I do is a workaround now”: Clinician Use of Workarounds Following an EHR Transition. Journal of General Internal Medicine. https://doi.org/10.1007/s11606-026-10840-6

Image Credits: AI Generated

DOI: 10.1007/s11606-026-10840-6

Keywords: electronic health record, EHR transition, workarounds, Veterans Affairs, health informatics, clinician burnout, patient safety, workflow, qualitative research, health IT, documentation burden, sociotechnical systems

The timing of the study reflects a broader shift in the engineering profession. As industry moves from the automation-focused vision of Industry 4.0 toward the human-centered ideals of Industry 5.0, employers increasingly expect graduates who can weigh well-being, sustainability, and resilience alongside performance metrics. Education 5.0, the parallel transformation in universities, asks curricula to pair advanced tools such as artificial intelligence and machine learning with an explicit concern for quality of life and long-term social consequences. The authors argue that these competencies cannot be left to develop by accident during a student’s first job. They must be deliberately cultivated in coursework, and the capstone project, where teams tackle real, ill-defined problems over an entire term, is an ideal place to start.

The BPSCm organizes that cultivation around four interrelated elements: artifacts, beliefs and culture, environment, and habits. When students analyze an engineering problem through the model, they are prompted to ask not only whether a technical solution will function, but who will use it, how it fits the social and cultural context, and what conditions might determine whether it is adopted at all. The framework draws on the biopsychosocial tradition, in which biological factors such as human physical characteristics, psychological factors such as cognition and motivation, and sociocultural dynamics such as communication patterns are treated as interacting dimensions rather than separate checklists. Crucially, the model does not claim causal relationships between these dimensions. Instead, it provides a qualitative structure for reflection, helping students notice how a change in one area, say a community’s habits, might ripple through the others and alter the fate of a project.

Methodologically, the study took an unusual and refreshingly honest approach. Rather than measuring student outcomes with tests or surveys, the researchers used an Educational Design Research framework combined with a Self-Study of Teaching Practice methodology. The two course instructors, who are also the study’s authors, kept separate reflective journals, recording one entry after each of roughly sixteen weekly class sessions across two successive academic terms. The first cohort included all 77 enrolled students and the second all 44, but the authors are careful to stress that these numbers describe the classroom context only. Students were never recruited as research participants, and no student assignments, grades, or presentations were collected or analyzed as data. The unit of analysis was the instructors’ own documented observations and interpretations, a design that trades generalizability for a detailed window into pedagogical reasoning.

In the first iteration, the instructors wove the BPSCm into the Technology Transfer Cycle, a phased process running from problem identification and formulation through referential framework building, technological solution development, and three forms of validation: academic, static, and dynamic. Students submitted written reflections analyzing an engineering solution through the model’s four components. The instructors’ journals suggested that students did begin connecting technical decisions with social, cultural, and biological considerations, and a semantic analysis of the journal entries revealed three interconnected domains underpinning engineering competence: education, comprehensive analysis, and skill development. But a persistent problem emerged. Students struggled to apply the framework consistently across project stages and had difficulty translating abstract biopsychosocial dimensions into concrete engineering decisions, particularly when trying to link social and cultural factors with technical variables.

That observation directly shaped the second iteration, where the instructors introduced a graphical representation of the BPSCm as a scaffolding device. Students first performed a context analysis, drawing the actors of a system as boxes, describing the artifacts and tools currently in use, and mapping the relationships among them through the model’s components. They were encouraged to begin from any element of the model to avoid value judgments and reach a more objective interpretation. Then came a solution analysis, in which students reflected on how the system would change once their proposed artifact was implemented, defining general and specific objectives and key variables to evaluate how the target context might transform. The visual structure gave teams a shared language: instructors noted that when students explained their diagrams during presentations, previously hidden interconnections between actors, technical variables, and sociocultural factors became explicit and discussable.

The thematic analysis of the second iteration’s journals surfaced four emerging categories: engineering skills development, analytical skills, formative research, and professional values. Compared with the first cohort, the instructors observed a shift from a broad, experience-based approach toward a more structured and methodical one. Students appeared to break complex problems into manageable components, question their own assumptions, and revise designs iteratively after considering social, cultural, and organizational implications. The instructors documented students moving away from treating solutions as fixed outcomes and toward viewing them as evolving processes open to refinement. They also reported growth in teamwork and research practices, with teams dividing tasks more deliberately, documenting findings systematically, and validating assumptions with data rather than intuition. One instructor reflected that without the model’s analysis, students would not have had the opportunity to address their problems holistically and systemically.

The study is candid about its limits, and those limits matter. Some teams never fully managed to connect the biopsychosocial and cultural dimensions with technical design variables, and a few focused on the visual polish of their diagrams without genuinely engaging with the underlying relationships. Students encountered a learning curve with the graphical aspects of the model, requiring structured workshops and continuous mentoring to apply it coherently. The instructors also noted that inconsistent exposure to the framework across earlier courses created confusion when students reached the capstone. Because the analysis rests entirely on instructor reflections, with the researchers simultaneously serving as teachers, interpretive bias is an acknowledged risk, mitigated only partially by immediate journaling, collaborative discussion, and an external collaborator’s detached perspective on the analysis.

Even with those caveats, the findings carry real weight for anyone thinking about the future of technical education. The instructors’ reflections indicate that a structured conceptual model, especially one made visible through diagrams, can help students move beyond purely technical reasoning and integrate human, social, and environmental considerations into engineering judgment. The approach aligns with Ausubel’s Meaningful Learning theory, which emphasizes active, hands-on engagement that produces durable understanding, and it complements existing tools such as Causal Loop Diagrams and Unified Modeling Language diagrams that already help engineers visualize system behavior. The authors propose extending the BPSCm with an additional analytical layer inspired by System Request, incorporating business need, requirements, value, and constraints to make contextual analysis even more explicit.

What happens next could determine whether this idea spreads beyond a single South American capstone course. The authors call for longitudinal studies that follow students into their careers to see whether biopsychosocial and cultural perspectives persist once the scaffolding is removed, and for implementations in different cultural and educational settings to test the model’s adaptability. They are also explicit that their conclusions are exploratory interpretations rather than direct evidence of changed student learning. Still, the core message resonates far beyond one classroom: engineering solutions are built for specific communities with particular beliefs, habits, and environments, and the engineers of the Industry 5.0 era will need more than code and calculus to serve them. Teaching students to see the whole system, the researchers suggest, may be the most important systems lesson of all.

Subject of Research: Integrating a biopsychosocial and cultural model into engineering capstone education to foster systems thinking

Article Title: Examining systems thinking through a biopsychosocial and cultural perspective in engineering education

Article References: Sabogal-Alfaro, G. P., Feijoo-Garcia, M. A., & Magana, A. J. (2026). Examining systems thinking through a biopsychosocial and cultural perspective in engineering education. Discover Education, 5(1), Article 1138. https://doi.org/10.1007/s44217-026-02257-4

Image Credits: AI Generated

DOI: 10.1007/s44217-026-02257-4

Keywords: systems thinking, engineering education, biopsychosocial model, capstone projects, project-based learning, sociotechnical systems, sociocultural aspects, Educational Design Research, self-study of teaching practice, Industry 5.0, conceptual modeling, learner-centered learning

The paper introduces what its author calls the 3×2A Framework, a profile-based interpretive approach designed to coordinate the ethical, functional, and institutional questions raised by AI-mediated organizations. The name encodes its structure: three diagnostic dyads, each containing two interdependent axes. The first dyad pairs automation with augmentation, capturing how work is redistributed between machines and people. The second pairs alliance with alignment, describing the relational quality of human–AI cooperation and the degree to which system behavior tracks human values and intentions. The third pairs adaptation with accountability, addressing how institutions learn and change over time and who answers for outcomes when things go wrong.

What distinguishes the framework from existing checklists and compliance tools is its insistence on situatedness. Rather than asking whether an AI system is safe or fair in the abstract, the 3×2A approach asks how a specific configuration of people, models, tasks, and institutions behaves in a specific context. A radiology department using AI-supported screen reading, for example, occupies a very different position on the automation–augmentation axis than a law office deploying automated contract analysis, even if both rely on similar underlying technology. The framework’s output is a six-axis profile, an ordered qualitative description that reveals functional, relational, adaptive, and accountability-related tensions within that configuration.

The epistemology of these profiles receives careful attention, and it is here that the article makes one of its most consequential moves. Ungureanu is explicit that the qualitative profile bands are ordered, evidence-informed interpretive categories, not numerical or cardinal scores. They convey direction and relative position, but they do not support arithmetic. Where numerical scoring is methodologically justified, it requires explicit rubrics, evidence standards, scoring rationales, and clearly stated interpretive limits. The framework also permits a composite synthesis, the 3×2A quotient, but only as a secondary and conditional construct built on an established profile, subject to documented activity–axis judgments, evidential bases, weighting assumptions, aggregation rules, and interpretive limitations.

This epistemic modesty is a deliberate corrective to a familiar failure mode in AI governance: the seduction of the single number. Risk scores, maturity indices, and compliance dashboards promise comparability and ease of communication, but they can obscure the qualitative judgments buried inside them and create false confidence in precision that the underlying evidence cannot support. By insisting that profiles come first and any quotient comes second, the framework attempts to preserve the richness of situated analysis while still allowing structured synthesis when the evidential ground is firm enough to bear it.

To demonstrate the framework in use, the article offers a worked benchmark re-analysis of a publicly documented case: the environmental-monitoring pilot conducted with the Province of Fryslân in the Netherlands, part of a responsible-use-of-AI project carried out with the Rijks ICT Gilde and the Z-Inspection initiative. That project examined a trustworthy AI and fundamental rights assessment of an AI system used in environmental oversight. Ungureanu re-reads the publicly available documentation through the six axes, producing an ordinal qualitative profile of the configuration without numerical scoring or a composite quotient. The exercise illustrates procedure rather than proof; the author is careful to state that the re-analysis does not claim empirical validation of the framework itself.

The Fryslân benchmark is instructive precisely because it shows the framework’s diagnostic texture. On the automation–augmentation dyad, the relevant question is not whether the AI replaced human monitors but how the division of labor shifted and whether the resulting configuration preserved meaningful human judgment. On alliance–alignment, the analysis probes whether the human–AI relationship functioned as genuine cooperation and whether the system’s outputs remained aligned with the values and legal obligations embedded in the assessment. On adaptation–accountability, it asks how the institution absorbed lessons from the pilot and how responsibility for outcomes was distributed among developers, public officials, and external assessors. Each axis generates tensions that a simple compliance verdict would flatten into invisibility.

The broader intellectual context matters here. The article draws on a rich literature spanning human-centered AI, machine behavior, algorithmic legitimacy, and the sociology of sociotechnical systems. It engages the well-documented observation that fairness and ethics principles often fail when abstracted from the concrete systems in which they must operate, and it echoes calls to shift from asking whether AI is good or fair to asking how it shifts power. It also responds to a growing consensus that generative AI in particular demands adaptive governance: rules and review processes that evolve as systems, users, and institutions co-evolve. The 3×2A Framework can be read as an attempt to give that consensus an operational instrument, one that treats governance as an ongoing interpretive practice rather than a one-time certification.

The stakes are considerable. As AI systems become embedded in knowledge production, labor organization, and public decision-making, the question of who is accountable when automated processes fail grows sharper. Research on moral crumple zones has shown how humans positioned near automated systems can absorb legal and moral responsibility for failures they could not realistically prevent. A profile-based approach makes such misalignments visible by mapping, for a given configuration, where control actually resides, where adaptation happens, and where accountability gaps open. In that sense, the framework functions less as a gatekeeping device and more as a mirror, forcing institutions to confront the structure of their own AI-mediated arrangements.

Critics may reasonably ask whether an interpretive, qualitative instrument can carry the weight of governance, particularly in regulatory environments moving toward harmonized rules such as the European Union’s Artificial Intelligence Act. Ungureanu’s answer, embedded in the framework’s design, is that interpretive rigor and regulatory utility are not opposites. The framework does not replace risk management standards or legal compliance; it complements them by providing a structured way to reason about configurations before and after formal requirements are applied. Its profiles are meant to be evidence-sensitive and revisable, updated as systems and institutions change. Whether the approach gains traction among practitioners will depend on whether organizations find that a six-axis portrait of their human–AI arrangements yields insights that dashboards and checklists miss. The article’s wager is that in a world where humans and machines are coevolving faster than governance can adapt, the most valuable tool is not another score but a clearer way of seeing.

Subject of Research: A profile-based interpretive framework for human–AI coevolution and adaptive AI governance

Article Title: The 3×2A framework: a profile-based interpretive approach to human–AI coevolution and adaptive governance

Article References: Ungureanu, V. (2026). The 3×2A framework: a profile-based interpretive approach to human–AI coevolution and adaptive governance. AI & SOCIETY. https://doi.org/10.1007/s00146-026-03352-8

Image Credits: AI Generated

DOI: 10.1007/s00146-026-03352-8

Keywords: generative AI, large language models, human–AI coevolution, adaptive governance, AI accountability, automation, augmentation, alignment, sociotechnical systems, AI ethics, AI & Society, 3×2A framework

The study, authored by Joshua King Obeng-Nyarko of Essex Business School at the University of Essex, introduces what its author calls the AI-Assessment Ecosystem Model. Rather than treating AI in assessment as a discrete technical intervention to be evaluated on accuracy and efficiency, the framework reframes it as a sociotechnical and institutional arrangement whose legitimacy depends on the interaction of five interdependent domains: technological innovation, pedagogical alignment and learner needs, institutional strategy and governance, faculty and professional readiness, and ethical and inclusive implementation. The model’s central claim is deliberately provocative: a system can be technically brilliant and still fail, because legitimacy is not produced by any single element but by how the elements relate to one another.

To build the framework, Obeng-Nyarko conducted a conceptual synthesis rather than a systematic review, searching Scopus, Web of Science, ERIC, IEEE Xplore and Google Scholar for literature published primarily between 2019 and 2024, the period in which scholarship on AI in higher education assessment expanded dramatically. From an initial pool of 107 potentially relevant publications, 58 sources were retained for close analysis, selected according to interpretive criteria that included whether a source addressed AI or digital assessment in higher education, contributed to one of the emerging model’s domains, explained relationships or tensions between domains, or provided foundational theory in areas such as assessment validity, feedback, institutional legitimacy, organisational change, inclusive design or computer-supported collaborative learning. The result is a theory-building contribution intended to organise a fast-moving field, not an empirical validation of a single tool.

The fragmentation the model responds to is real and consequential. One strand of the literature celebrates technical capability: automated scoring, adaptive question sequencing, predictive analytics, multimodal analysis and explainable AI. A second strand foregrounds ethics and politics, warning that algorithmic systems embed bias, opacity, surveillance and data extraction into decisions that shape learners’ futures. A third examines pedagogy, asking whether automated feedback genuinely supports self-regulated learning or merely accelerates surface-level correction. A fourth addresses governance, procurement, quality assurance and organisational change. Each strand offers insight, but the study argues that when they are examined in isolation, AI adoption risks being driven by technical feasibility, institutional positioning or policy pressure rather than educational purpose.

Assessment is a uniquely high-stakes site for this tension. It is the mechanism through which institutions translate educational values into judgments about performance, progression and achievement. AI may bring speed, scale and responsiveness, but it can also intensify standardisation, obscure how judgments are made and displace human interpretation in ways that are difficult to defend on educational grounds. The study draws on assessment scholarship to stress that feedback only becomes educationally valuable when learners can interpret it, act on it and incorporate it into future work — a standard that apparent immediacy and personalisation from AI systems can easily mask. If students use an AI tool to polish text without ever understanding criteria or disciplinary standards, the technology may increase efficiency while quietly eroding assessment literacy.

The framework’s organising concept is legitimacy, theorised not as a property of a technology but as a relational and contested process of justification. Drawing on institutional theory, the study distinguishes pragmatic legitimacy, in which stakeholders see an arrangement as serving their interests; moral legitimacy, in which it is judged fair and aligned with educational values; and cognitive legitimacy, in which it becomes taken for granted as normal practice. The distinctions matter because they can pull in opposite directions. An AI feedback system may have pragmatic legitimacy for institutional leaders because it delivers timely feedback at scale, yet lack moral legitimacy among students if its judgments are opaque or inequitable. It may be normatively defensible to faculty who see formative value in it, yet culturally resisted where it conflicts with disciplinary expectations of human judgment. A system can even become cognitively normalised across a university while remaining ethically problematic — normalisation, the study warns, should never be mistaken for educational defensibility.

Each of the five domains carries its own technical and practical demands. Technological innovation is treated as enabling but insufficient: AI operates through data models, training assumptions and probabilistic outputs that may not align with the interpretive, context-sensitive nature of educational judgment, making explainability a condition of educational intelligibility rather than a mere engineering feature. Pedagogical alignment demands that AI use be evaluated against validity, formative value, student agency and disciplinary difference, since the standards governing fine arts, clinical education, laboratory science and the humanities are not reducible to a single model of performance. Institutional governance serves as the connective infrastructure linking innovation to accountability, covering procurement, data management, academic integrity policy and the conditions under which AI can be authorised, monitored, challenged and revised. The study also draws on organisational change literature to note that universities may adopt AI partly to secure legitimacy by conforming to sectoral expectations — a form of isomorphic pressure that can generate pragmatic approval while leaving moral and pedagogical questions unresolved.

The remaining two domains address people and principles. Faculty and professional readiness is described as the model’s interpretive core, because AI does not remove the need for expertise but reconfigures it: educators must now interpret automated outputs, evaluate the quality of AI-generated feedback, recognise overreliance, identify bias and decide when human judgment should take precedence. Recent AI literacy research suggests this requires identifiable competencies — understanding data, models, ethics and limitations — rather than generic digital confidence, meaning professional development must go beyond introductory workshops to support assessment redesign and routes for student dialogue. Ethical and inclusive implementation, meanwhile, is framed as constitutive rather than supplementary. Students differ in technology access, linguistic repertoire, disability status, neurodiversity and confidence in contesting institutional decisions, so AI-enabled assessment may redistribute advantage and disadvantage even when systems appear technically neutral. Inclusion, the study argues, must be built into design from the outset through accessible formats, transparent criteria, clear appeal routes and genuine opportunities for students to shape AI-mediated practices.

To illustrate the model in action, the paper offers a hypothetical case: a university introduces a generative AI tool providing draft feedback on written assignments in a large first-year module. Viewed through the technological domain alone, the system looks like a clear win — fast, adaptive, detailed. But the ecosystem lens surfaces harder questions. Does the feedback develop students’ evaluative judgment or merely polish prose? Do staff understand the system’s limits well enough to explain appropriate use and catch misleading advice? Who is accountable when the tool produces harmful or biased guidance, and is student work being used to train commercial models? Can multilingual and disabled students access the tool on equal terms, and can they contest its advice? The tool may enjoy pragmatic legitimacy for leaders burdened by feedback workloads and for students craving immediate support, yet still lack moral legitimacy if it weakens agency or obscures accountability.

The study is explicit about its limits: the model derives from conceptual synthesis rather than empirical testing, its domains reflect interpretive judgment, and its usefulness will depend on how well it travels across institutions, disciplines and regulatory environments. But its diagnostic ambition is clear. Misalignment between domains is not always a coordination failure to be engineered away; sometimes it reveals genuine conflicts of value — between efficiency and human judgment, standardisation and disciplinary diversity, innovation and accessibility. The framework’s purpose is not to resolve those tensions into consensus but to make them visible, to show whose interests proposed alignments serve, and to insist that AI-enabled assessment be justified through transparent, participatory and revisable institutional processes. As universities race to embed AI in grading and feedback, the message is stark: the technology that works is not necessarily the technology that can be defended.

Subject of Research: A conceptual ecosystem framework for evaluating artificial intelligence in higher education assessment across pedagogical, governance, professional and ethical domains

Article Title: Artificial intelligence in higher education assessment as an ecosystem of pedagogy governance and legitimacy

Article References: Obeng-Nyarko, J. K. (2026). Artificial intelligence in higher education assessment as an ecosystem of pedagogy governance and legitimacy. Discover Education, 5(1), Article 1074. https://doi.org/10.1007/s44217-026-02225-y

Image Credits: AI Generated

DOI: 10.1007/s44217-026-02225-y

Keywords: artificial intelligence, higher education, assessment, automated feedback, educational legitimacy, institutional governance, faculty readiness, inclusive assessment, learning analytics, generative AI, sociotechnical systems, academic integrity

The method is called the Functional Resonance Analysis Method, or FRAM, and it was created by safety researcher Erik Hollnagel to study complex sociotechnical systems in which humans, organizations, and technologies interact in ways that no flowchart can fully capture. Rather than describing a process as a fixed sequence of steps, FRAM represents it as a network of interdependent functions that can activate simultaneously and influence one another. Its central insight is that small, everyday variations in timing, resources, or behavior are normal and usually harmless, but when several of these variations align, they can resonate and amplify into major failures, much like structural resonance in engineering. In surgery, that resonance looks like a canceled operation caused by a sick nurse, a late pre-operative test, and an overrunning morning list all landing on the same day.

Traditional surgical process models, the formal or semi-formal representations used in computer-assisted surgery research, have struggled to capture this reality. Most focus narrowly on what happens inside the operating room, describe work as it is imagined in official protocols rather than work as it is actually done, and assume that workflows proceed sequentially. Real operating room management is nothing of the sort. It is a parallel, adaptive, multi-agent endeavor in which patients, residents, attending surgeons, and central patient management continuously negotiate and renegotiate the plan. The Munich team, led by Sidra Rashid of the MITI Research Group, argued that any scheduling model realistic enough to support AI-assisted planning must first capture this genuine complexity, and that FRAM offered a way to do it.

To build their model, the researchers spent months embedded in the visceral surgery department of University Hospital rechts der Isar. They conducted semi-structured interviews with twelve staff members, including central patient management staff, surgeons, residents, schedulers, coordinators, assistants, anesthetists, and nursing management, and complemented the interviews with direct shadowing during morning meetings, in operating rooms, and in informal staff interactions. This combination proved essential: the observations revealed informal practices and operational challenges that never surfaced in the formal interviews, such as surgeon preferences and meeting-based coordination, exactly the work-as-done details that conventional models omit.

From this evidence, the team constructed a FRAM model containing 98 unique interconnected functions spanning four central human agents, the patient, the medical resident, the attending surgeon, and central patient management, along with a fifth systemic agent representing the scheduling software infrastructure. Each function is described by six aspects: the input that triggers it, the output it produces, the preconditions that must be met, the resources it requires, the controls or rules that govern it, and the timing constraints that bound it. Functions are linked through these aspects, so that one function’s output becomes another’s input, resource, or precondition. This web of couplings is precisely how variability propagates through the system. The model was organized into four temporal phases, from the initial patient visit through the day of surgery, and validated iteratively with the interviewed staff.

The genuinely novel step came next. Using the FRAM Model Interpreter, a simulation tool that performs cycle-by-cycle analysis of which functions can activate, the team ran the model through ten simulation cycles, mapping each cycle onto the temporal phases of the scheduling process. They also introduced a structured interpretation protocol that assigns each function a profile specifying which inputs, preconditions, resources, and timing requirements must be satisfied for activation, using logical rules such as all, any, or none. Clinical functions like examining and consulting a patient demanded complete information before activating, while coordination functions like planning the surgical schedule used flexible profiles, reflecting the improvisational nature of real scheduling work. The researchers say this cycle-mapping interpretation approach had not been used in prior qualitative applications of FRAM.

The simulation results were revealing. The very first cycle activated 58 functions simultaneously, fourteen entry functions and forty-four background functions, underscoring how much of surgical scheduling begins in parallel rather than in sequence. Coordination functions such as planning the schedule and rescheduling surgeries activated repeatedly across multiple cycles, demonstrating that scheduling is iterative and adaptive rather than a one-time decision. Notably, a four-cycle gap emerged between the initial planning of the schedule in cycle four and the final provision of the daily schedule in cycle eight, a window filled with repeated team allocation, discussion, and adaptation that the analysis identified as the primary bottleneck in finalizing the schedule.

The simulation also exposed how disruptions compound. A disruptive entry function representing a staff member calling in sick recurred in cycles two, four, six, and eight, and each time it coincided with rescheduling and cancellation functions, tracing a direct causal pathway from staff unavailability to schedule upheaval. Three separate delay functions, originating with surgeons, patients, and residents, co-activated in multiple cycles, indicating that delays from different sources stack and amplify one another rather than occurring independently. This is functional resonance made visible: ordinary variability in several places at once producing an outsized systemic effect.

To move beyond qualitative description, the team applied FRAMalyse, an open-source analysis tool, to compute quantitative measures of variability across the coupling network. The Overall Functional Coupling Variability metric integrates network centrality measures, including Katz, in-closeness, out-closeness, and betweenness centrality, to rank functions by their potential to contribute to resonance. The results were unambiguous: allocating the operating room team showed the highest variability of any function in the entire model, burdened by a dense web of upstream dependencies including staff availability, surgery characteristics, team qualifications, documentation, and meeting outcomes, combined with constraints on training, skills, preferences, and shift composition. Attending surgeons performed the most functions overall, while central patient management showed the highest coupling density, reflecting its role in implementing decisions made by clinicians and patients. The patient visit phase emerged as the most functionally dense stage, where key surgical decisions are made through direct patient-surgeon interaction.

The authors are careful to note the limits of this first quantitative application. The model covers a single department in one hospital, the simulation represents just one possible instantiation of system behavior, and the weighting factors within the variability metric are not yet empirically validated, so the bottleneck ranking should be read as indicative rather than definitive. No external validation against observed disruption data has yet been performed. But the roadmap is ambitious: the team plans to validate the candidate bottlenecks against historical scheduling data, compare FRAM against alternative modeling formalisms such as BPMN and Petri nets, extend the model across the full perioperative pathway, integrate AI-based scheduling systems, and ultimately develop dynamic long-term simulations that could evolve into a digital twin of hospital operations. If that vision is realized, a safety method born in industrial risk analysis could become a predictive decision-support tool, helping hospitals design schedules that bend with reality instead of breaking under it.

Subject of Research: Quantitative application of the Functional Resonance Analysis Method to model surgical scheduling complexity and identify process bottlenecks

Article Title: Introducing FRAM as a tool for surgical process modeling: a case study for understanding surgical scheduling complexity

Article References: Rashid, S., Weber, R., Geiger, A., Wagner, L., Bernhard, L., Spicker, E., Jell, A., Fottner, J., Wilhelm, D., & Grabbe, N. (2026). Introducing FRAM as a tool for surgical process modeling: a case study for understanding surgical scheduling complexity. International Journal of Computer Assisted Radiology and Surgery. https://doi.org/10.1007/s11548-026-03800-2

Image Credits: AI Generated

DOI: 10.1007/s11548-026-03800-2

Keywords: FRAM, surgical scheduling, operating room management, surgical process modeling, functional resonance, hospital operations, variability propagation, bottleneck identification, sociotechnical systems, digital twin, computer-assisted surgery, patient safety

The stakes could hardly be higher. The world’s population stood at 8.2 billion in 2024 and is projected to peak at roughly 10.3 billion around 2080, making food security a defining challenge of the century, enshrined in the United Nations’ Sustainable Development Goal 2 for zero hunger. Agricultural land use has already expanded from 3.7 billion hectares in 1950 to 4.83 billion hectares in 2023, covering about 32 percent of Earth’s land surface, while the share of the global labor force employed in agriculture has fallen from 44 percent in 1991 to 27 percent in 2019. The result is a sector asked to grow more food with fewer farmers, on finite land, under mounting environmental constraints. In developed nations, workforces are aging and potential entrants are declining; the shortage of agricultural labor in OECD countries pushed U.S. labor costs up by 17 percent in 2023, and the COVID-19 pandemic exposed how fragile the reliance on migrant seasonal workers truly is when borders close.

Paradoxically, productivity keeps rising. Between 2001 and 2015, global gross agricultural output per worker grew at an average annual rate of 3.77 percent, driven by technological innovation and improved practices spanning fertilizers, seeds, irrigation, and machinery. The OECD-FAO outlook projects that world agricultural production will expand by roughly 14 percent between 2025 and 2034, with middle-income countries leading the charge. This trajectory has historically tracked the industrial revolutions. Agriculture 1.0 relied on manual labor with rudimentary tools; the machinery of Industry 1.0 and the oil-powered mechanization of Industry 2.0 brought engine-driven equipment across the entire production process; and Industry 3.0’s embedded systems and software enabled precision agriculture. The current era, Agriculture 4.0, converges artificial intelligence, machine learning, digital twins, edge computing, the Internet of Things, robotics, and smart and nanosensors into systems that collect, transmit, and analyze data at a speed and scale no human can match, generating actionable instructions for irrigation, fertilization, sowing schedules, and crop management.

On paper, the promise is extraordinary. Drones both sense the field and intervene in it, spraying water and pesticides with precision guided by real-time environmental data. Wearable sensors track animal health and reproductive cycles. Cyber-physical systems enhanced with AI can perform targeted interventions autonomously, and agricultural research has historically delivered average social returns exceeding 40 percent annually in developing countries. Yet the deployment record tells a different story. Of the world’s 608 million farms, 84 percent are smaller than two hectares, while the largest 1 percent control more than 70 percent of global agricultural land. Advanced applications such as variable-rate technologies, drones, and robotic systems show adoption rates generally ranging from just 4 to 22 percent, varying widely by region and crop. In low- and middle-income countries, drones and robots are rarely adopted at all—although, tellingly, farmers in both rich and poor countries consistently express strong interest and positive attitudes toward these technologies.

The barriers are well documented: uncertainty about cost-effectiveness, substantial upfront infrastructure and maintenance investments, misalignment with existing workflows and equipment, knowledge gaps tied to an aging workforce—the average farm manager worldwide is over 55 years old—and concerns about data security and usability. Robotic harvesting illustrates the problem acutely. Agricultural robots must operate in unstructured environments with variable light, weather, and terrain, interacting with irregular plants, perishable produce, and unpredictable livestock. No robotic harvesting solution has yet been commercially adopted for tree fruit crops, which researchers attribute to inadequate performance compared with human workers, limited adaptability to diverse orchards, and high financial risk from uncertain returns. While farmers remain unconvinced, agri-tech companies themselves often lack adequate knowledge of farm business models, leaving a two-sided information vacuum that neither marketing nor engineering has filled.

This is where the paper’s central argument enters. The European Commission’s Industry 5.0 framework—built on human-centricity, sustainability, and resilience—offers a policy vision, but it does not specify how to achieve it. Salzer contends that Human Factors and Ergonomics, or HF/E, provides precisely the methods needed to operationalize that vision and, more urgently, to close the adoption gap. The discipline has simply never turned its attention to farming. An analysis of the Human Factors and Ergonomics Society’s annual meetings from 2015 to 2025 found that of 5,047 individual presentation titles, only nine—roughly 0.17 percent—were agriculture-related. The society’s flagship journal, Human Factors, published 1,067 papers over the past decade with only nine addressing agricultural contexts. Conversely, of 2,240 articles in the leading journal Biosystems Engineering between 2015 and 2025, just 32 incorporated HF/E concepts, and most of those addressed narrow physical safety issues like machinery rollover rather than cognitive ergonomics or sociotechnical design.

The paper maps a practical toolkit across the technology lifecycle. To understand the work domain, developers can use knowledge elicitation, direct observation, and Hierarchical Task Analysis—decomposing tasks such as pesticide application into subtasks to reveal what farmers perceive, decide, and know tacitly, as demonstrated in vineyard safety research. Anthropometric review ensures tools fit diverse user populations, adapting hand tools to local body-measurement data to reduce strain. In the design phase, participatory methods—contextual inquiry, focus groups, co-design workshops with sketches and low-fidelity prototypes—involve farmers as collaborators before resources are committed, an approach used successfully in developing an E. coli risk decision-support system with regulators, industry, academics, and farmers at the table. For AI integration, the paper stresses that agricultural expertise is fundamentally tacit, built on seasons of observation, and that explainable AI must make sense to farmers in a way that fits how they naturally think, not merely to the researchers who built the models. Sociotechnical frameworks examining people, tasks, tools, environments, and organizations help anticipate implementation conflicts before deployment.

Evaluation and adoption round out the framework. Usability testing and heuristic evaluation, drawing on methods proven in aerospace and healthcare, assess whether systems fit real workflows, while simulation with digital human models can expose usability problems before full-scale deployment. Cognitive Work Analysis, applied early in design, informed a pioneering robotic Medjool date thinning system, where abstraction hierarchies and event sequence diagrams refined human-robot coordination requirements. Time and motion studies combined with economic modeling have helped apple growers decide whether mechanical harvest platforms are worth buying and how to deploy them. On the adoption side, training needs analyses, knowledge-sharing platforms, and the UTAUT framework—which identifies performance expectancy, effort expectancy, social influence, and facilitating conditions as drivers of acceptance—address the human dimension of uptake, particularly for older farmers who dominate the workforce. A newly proposed Technology Acceptance Level metric aims to measure whether deployed systems achieve routine, trusted use.

The author is careful about limits. HF/E methods cannot guarantee that a technology is worth the investment, resolve credit access, build infrastructure, or fix unclear regulations; they are one contributing factor among several in a broader adoption challenge, and the proposition that they narrow the Agriculture 4.0 gap has yet to be empirically substantiated. Nor is scalability trivial: farming is intensely heterogeneous, and technologies demanding heavy customization pose economic risks for developers targeting smallholders and niche crops. As one cited analysis cautions, technology shaping better futures will not have a future if it stays concentrated in the northern hemisphere. Yet the concluding message is unambiguous: transitioning from technology-driven to human-centered innovation aligns with the experiential nature of agricultural work, and the systematic application of human factors methods—while not sufficient—is necessary to finally realize what Agriculture 4.0 promised, through the human-centric lens of Agriculture 5.0. Whether the next generation of farm machines is built with farmers, rather than merely for them, may determine the sustainability and equity of the food systems on which billions depend.

Subject of Research: Applying human factors and ergonomics methods to overcome low adoption of Industry 4.0 technologies in agriculture and advance toward human-centric Agriculture 5.0

Article Title: Can industry 5.0’s human-centric approach fulfill industry 4.0’s unrealized promise to agriculture?

Article References: Salzer, Y. (2026). Can industry 5.0’s human-centric approach fulfill industry 4.0’s unrealized promise to agriculture?. Smart Agricultural Technology, 15, Article 102551. https://doi.org/10.1016/j.atech.2026.102551

Image Credits: AI Generated

DOI: 10.1016/j.atech.2026.102551

Keywords: Agriculture 5.0, Industry 4.0, human factors and ergonomics, smart farming, agricultural technology adoption, agricultural robotics, explainable AI, precision agriculture, human-centered design, food security, farm labor shortage, sociotechnical systems