Jason Zagami

Purpose: This conceptual article revisits Heinz von Foerster's Doomsday calculation as a provocation about the authority of formal description and uses it to examine artificial intelligence (AI) as a new condition for second-order cybernetics. Design/methodology/approach: The article develops a theoretical argument through von Foerster's observing systems, non-trivial machines, constructed realities and ethical imperatives. It brings these ideas into dialogue with cybernetic work on feedback, closure, conversation and requisite variety, and with contemporary concerns about algorithmic classification and institutional decision-making. Findings: AI should not be treated as an observer in the full autopoietic or ethical sense. Rather, AI systems computationally recontextualise prior human, institutional and technical observations into new actionable descriptions. This creates a computationally mediated second-order arrangement in which people and organisations encounter reconstructions of prior observations before the phenomena those descriptions claim to represent. Social implications: The argument highlights risks for education, climate governance, health, law and public administration when synthetic descriptions harden into institutional reality without adequate provenance, warrant, contestability or repair. Originality/value: The article contributes three linked concepts to second-order cybernetics: computational recontextualisation as the AI-mediated relocation of prior observations; computationally mediated second-order observation as the condition in which model-produced reconstructions of prior distinctions mediate subsequent human and organisational distinctions; and reality maintenance as an integrative framework for keeping AI-mediated descriptions open to feedback, contestation and repair.

Purpose: The study aims to develop a conceptual account of trust in AI-mediated complex systems by addressing a post-reality problem insufficiently captured by debates on trustworthy AI, misinformation and digital governance: social systems increasingly act on synthetic, predictive or institutionally mediated representations before their provenance, limits and contestability are established. Trust is reconstructed as a second-order cybernetic process of reality maintenance. Design/methodology/approach: Theory is used as the method through an integrative conceptual synthesis of the sociological trust theory, social systems theory, cybernetics, philosophy of technology, AI trust research, organisational systems and education. Systems-theoretical distinctions, including observation, second-order observation, feedback, requisite variety, operational closure and autopoiesis, are used to redescribe trust under AI-mediated post-reality conditions. Findings: Post-reality is not the disappearance of reality, but a condition in which AI-mediated observations become decision premises before they are adequately traced, interpreted, contested or corrected. Trust cannot be reduced to confidence in persons, institutions or technologies. It depends on recursive arrangements through which representations are generated, observed, challenged, repaired and learned from. A reality-maintenance model is developed around provenance, interpretability, participatory contestability, institutional accountability and observer formation. The model identifies post-reality drift as the recursive process through which AI-mediated representations become institutionally consequential, reshape the environments they describe and then reinforce subsequent observations. Originality/value: Trust is reconstructed as a cybernetic achievement of observing systems rather than as a psychological attitude, an institutional property or a technology-adoption variable. The article distinguishes post-reality from post-truth, extends the trust theory to AI-mediated second-order observation and positions education as a formative subsystem for warranted trust and accountable distrust across complex institutional environments today.

Generative artificial intelligence is increasingly framed as a catalyst for deeper learning rather than simple automation. This design-based study investigated how personalised GPT agents were used as cognitive partners by pre-service teachers in an undergraduate Initial Teacher Education course in Digital Technologies within a Bachelor of Education (Secondary) programme. Drawing on distributed cognition, constructivism, cognitive load theory, and critical AI literacy, the intervention embedded personalised GPT agents within a university Canvas Learning Management System. Fourteen students configured and used agents aligned to curriculum goals, teaching contexts, pedagogical preferences, and weekly Prompt–Reflect–Refine tasks. Data included approximately 7,000 AI interactions, reflective journals, assessed lesson and assessment artefacts, and tutorial-group discussions. A hybrid inductive–deductive thematic analysis identified four recurrent functions in students’ use of the agents: conceptual tutor, co-designer, reflective mirror, and cognitive challenger. Across data sources, students reported clearer conceptual support, iterative planning, and reflective prompting, alongside risks of over-reliance, factual drift, and culturally narrow exemplars. The study contributes a context-specific model for LMS-integrated generative AI that foregrounds partnership, personalisation, judgement, and ethical reflexivity. Findings should be interpreted cautiously because the study was small-scale, single-site, and partly reliant on self-reported reflection.

Generative artificial intelligence has unsettled higher education because it has exposed a fragile assumption at the heart of curriculum and assessment: that the submitted artefact can be treated a...

Generative artificial intelligence is becoming part of university learning, but students do not engage with it uniformly. This mixed-methods survey of students at one Australian university examined whether successful-student identity and student-reported academic achievement were associated with different patterns of generative AI engagement. The institution-wide survey yielded 484 responses, including 469 with valid GPA-band data. Higher-achieving students reported lower active AI engagement, positive affect, perceived learning impact, and AI-related disengagement, alongside slightly higher negative affect. Item-level analyses showed less endorsement of AI-supported autonomy, effective learning, and active engagement, and greater agreement that reliance on AI hinders critical thinking and independent problem solving. Open-ended responses indicated that these students used AI selectively for clarification, summarisation, and workflow support while checking outputs and keeping them subordinate to their own judgment. The paper interprets this pattern as guarded adoption: selective, bounded, and verification-intensive use in defence of epistemic control and successful-student identity.

Recent Australian education reform has increasingly foregrounded explicit instruction, curriculum streamlining, and appeals to a singular “science of learning” as foundations for improving school performance, coherence, and teacher practice. While these developments are commonly framed as pragmatic and evidence-informed, this paper argues that their significance cannot be understood at the level of pedagogy alone. Rather, contemporary instructional reform must be examined as a form of curriculum governance that redistributes authority over knowledge, teaching, and professional judgment. Using conceptual and critical policy analysis grounded in curriculum theory, the paper draws primarily on Foucault and Freire to analyse Australian reform discourse, with particular attention to the consolidation of epistemic authority, the narrowing of legitimate evidence, the intensification of implementation fidelity, and the effects of these developments on curriculum interpretation. Recent Australian policy documents and reform artefacts are used as illustrative examples throughout the analysis. The analysis suggests that when explicit instruction and cognitive science are elevated from important pedagogical resources into dominant system logics, they contribute to curriculum closure. This occurs through the reduction of interpretive space, the disciplining of teacher judgment, and the privileging of transmission over dialogue as the model of legitimate curriculum enactment. The paper does not reject explicit instruction or evidence synthesis. Rather, it argues that their value depends on the governance conditions under which they are mobilised. When evidence functions as authority rather than as a resource for professional deliberation, curriculum risks becoming administratively coherent but educationally brittle. The paper concludes that curriculum reform requires forms of governance that preserve professional interpretation, dialogic engagement, and the capacity of systems to learn from practice. Without these conditions, reforms designed to increase certainty may narrow curriculum purpose and weaken long-term educational vitality.

Purpose of the study: To conduct a documentary, ex ante programme evaluation of how a bounded offline AI augmentation could be specified, governed and later tested within an offline OER teacher development programme delivered through Teacher in a Box. Methodology: Documentary programme evaluation using a thesis corpus. Methods: Theory of Change reconstruction, CIPP evaluation framework, directed qualitative content analysis with a scope-aligned codebook, and construct-to-instrument mapping for assessment tools. Offline AI specification is GPT4All class local LLMs plus retrieval-augmented generation (RAG) over the local Teacher in a Box OER and training repository, with governance wrappers. Main Findings: Offline OER effectiveness depends on integrated, in-situ teacher support rather than access alone. Offline AI is most defensible as a governed, retrieval-grounded, teacher-facing layer that improves resource discovery, supports formative assessment routines, and scaffolds assessment literacy. The study outputs measurable CIPP indicators, an evidence-to-implication warrant structure, a baseline-versus-augmentation risk and safeguard matrix, and offline-feasible instruments for formative assessment, summative task quality, character education evidence, and deeper learning portfolios. Novelty/Originality of this study: This study provides an audit-ready, evaluation methodology linking offline OER programmes to offline AI. It advances knowledge by separating evidence, implication, and design claims; specifying a minimum viable governance package; and producing offline-feasible assessment instruments and indicators for formative, summative, character, and deeper learning assessment under low-connectivity conditions, supporting policy and implementation decisions before field trials.