When the Curriculum Becomes the Research: Building IS 2025 with Hybrid Intelligence

The Gist: What happens when two researchers use the very tools they’re studying to redesign the curriculum that should teach those tools?

There is a particular kind of intellectual vertigo that comes from using a method as both instrument and subject. I have been sitting with it for several months now.

The paper Eduard Buzila and I have just submitted to “IS 2025 updating IS 2020 to Industry 4.0-5.0: A Competency Framework for Pedagogical Innovation through Human-AI Collaboration (Hybrid Intelligences) for Information Systems Education” is, at one level, a curriculum design paper. We took the current Information Systems undergraduate curricular (IS 2020) across 40 institutions (online publicly available information), ran it through a structured process combining our own research expertise followed by three LLM platforms and a set of AI-powered literature mapping tools, and produced a proposed update we call IS2025.

But the more interesting story is what happened in the process.

The IS2020 competency model, published by ACM and AIS, represents the last formal update to undergraduate IS education guidelines. It was a genuine shift from its predecessor (IS2010 Topi) moving from a list of prescribed courses to a competency-based model that gave institutions more flexibility. But it was finalised before the mass availability of generative AI tools. Before ChatGPT. Before the EU AI Act. Before Industry 5.0 moved from a policy aspiration to something companies were actually beginning to operationalise.

So the surface question was: what should be in a curriculum that prepares IS graduates for this moment?

The underneath question, the one that turned out to be more generative, was: can the process of building that curriculum demonstrate the hybrid intelligence principles the curriculum itself is trying to teach?

In other words, could we do it in a way that was also a live methodology?

I want to be precise here, because I think the term ‘hybrid intelligence’ is at risk of becoming another hand-wavy gesture towards human-AI collaboration without the texture of what that actually involves.

In our study, it looked like this.

We ran identical prompts through three LLM platforms (Claude, ChatGPT, and Gemini) and then compared outputs. Not to find a winner, but because each system had different blind spots, and those blind spots were informative. Claude produced a 23,000-word output that was strong on conceptual coherence and recognition of educational diversity but weak on implementation feasibility. Gemini gave us useful preliminary framing we already knew. ChatGPT was good for literature synthesis but treated everything as training data, raising privacy questions for novel curriculum approaches.

We also used three AI-powered literature mapping tools (ResearchRabbit, Litmaps, and Connected Papers) each operating on different algorithms, each surfacing different parts of the literature landscape. One finding stopped us: IS 2020 itself had near-zero citations on two of these platforms. The living document that is supposed to orient IS education globally was, in terms of citation visibility, almost invisible to the tools curriculum researchers now use.

The human work was not supervisory in a passive sense. It was substantive. LLMs consistently reverted to flat, individualised skill lists when asked to design competencies. They could not sustain the relational framing that distinguishes IS from a simple list of digital capabilities. Every time we wanted a competency to read as collaborative capacity within a human-technology assemblage rather than individual ability to use AI tools, that required deliberate human intervention. The LLMs kept pulling toward the latter.

There was also the hallucination problem. Plausible-sounding citations that do not exist. These were caught through verification against original sources, which is part of the workflow, but it is worth naming as a genuine research integrity challenge rather than treating it as a minor known limitation.

The framework we developed organises into seven competency areas: Foundations, Data & Analytics, Technology, Development, Organisational Domain, AI & Digital Literacy, Sustainability & Social Impact, and Implementation/Application.

The most significant departure from IS2020 is the establishment of AI & Digital Literacy as a discrete, structured competency area, nor an add-on or an elective strand, but a core domain with its own progression from Foundation to Expert level. This positioning reflects something the Royal Society’s 2025 AI in Education project also argued: AI literacy cannot simply be appended to existing curricula. It requires a fundamental reimagining of the knowledge, skills, and dispositions the curriculum is built to develop.

The second significant departure is the inclusion of Sustainability & Social Impact as a core competency area and not a module, not an ethics box to tick, but a structural element of the framework. The question Geoff Walsham posed over a decade ago “are we making a better world with ICTs?” is, we argue, now a competency question as much as a values question.

The third is what we call the pracademic (Posner, 2009) orientation: designing graduates who can cross between research and practice, who understand the EU AI Act implications for the organisations they will join, and who possess the metacognitive skills to keep learning as the landscape changes. IS2025 is built for practitioners who can think, not technicians who can execute.

Trauth, Farwell and Lee identified a ‘curriculum gap’ in 1993, a divergence between what IS programmes taught and what industry required. Thirty years later, the gap has shifted but not closed. It is no longer primarily about technical skills (though those matter). It is about whether graduates can navigate sociotechnical complexity: can they assess an AI system’s fitness for purpose? Can they identify when automation is replacing rather than augmenting human judgment? Can they operate in contexts governed by regulation they understand?

This connects directly to something I have been returning to in the recent posts here about AI and the labour market. The Anthropic research (Massenkoff & McCrory, 2026) on observed versus theoretical AI exposure showed a striking divergence: theoretical capability coverage of CS occupations near 94%; observed actual usage nearer 33%. That gap is not purely technical. It is filled with legal constraints, integration requirements, human verification steps, and institutional inertia. IS graduates need to understand all of that gap, not just the technical frontier.

Similarly, my work with PGCE Computing students reveals that the gap between what teacher education programmes say about AI and what trainee teachers actually encounter in classrooms is substantial. The curriculum design problem is not unique to IS. But IS has the disciplinary infrastructure, the sociotechnical framing, the organisational theory, the systems thinking tradition, to address it more rigorously than most.

Share

We are European researchers, primarily embedded in research-intensive institutions. The IS2025 framework carries that context. The research corpus focused on mainly Region 2 (Europe, Africa, Middle East, plus Australia) and are explicit about this, but explicitness does not dissolve the constraint.

LLMs also cannot assess implementation feasibility for resource-constrained institutions. They generate frameworks appropriate for well-resourced environments with extensive faculty development capacity. A university in a low-income context, with part-time instructors and limited infrastructure, is not the institution the models have internalised. This is not just a methodological limitation, it is a curriculum equity concern that future iterations of this work will need to address explicitly.

We have proposed a tiered implementation model: Tier 1 covering the essential elements achievable within 6–12 months, Tier 2 the full transformation over 18–24 months: partly in response to this. But that is a start, not a solution.

The deepest claim the paper makes is methodological rather than curricular. We used AgileDBR (Smith-Nunes, 2023, 2025) the Agile Design-Based Research methodology I developed combined with the Ontological Kaleidoscope framework (2025), to structure how human and AI contributions were documented, evaluated, and integrated. Every sprint produced Test and Learn Cards. Every LLM output was assessed against pedagogical coherence, disciplinary alignment, implementation feasibility, theoretical grounding, and assessment viability before incorporation.

The argument is that this kind of documented, iterative, human-led process is not just good research practice. It is the model for responsible AI-assisted curriculum design. The transparency about what the tools produced, what we changed, and why is itself the contribution.

This is what I mean when I say the process became the argument. We were not just using AI to design a curriculum about AI. We were trying to demonstrate, in the work itself, what thoughtful human-AI collaboration looks like when the stakes are educational.

The IS2025 framework is explicitly a draft. That is not false modesty, it reflects the curriculum design principle that no curriculum is a neutral statement, and this one needs to be interrogated by IS scholars from every region, every institutional type, every disciplinary tradition within the IS community.

The questions I would most like to see addressed: How does the framework hold for institutions outside Europe and North America? What assessment methods genuinely measure human-AI collaboration competencies rather than substituting for them? Where does the Industry 5.0 framing fall apart under scrutiny?

If you are working on any of these questions, or teaching IS at a university where the curriculum mismatch with current practice is acutely felt. I would very much welcome the conversation.

Share

Once the full paper is published I will update this post. The IS2025 competency framework and implementation guidance documents are archived at https://doi.org/10.5281/zenodo.18018076.

Related posts:

Beyond the Hype: What Recent Research Reveals About AI Integration in UK Teacher Training
Measuring the Unmeasurable: Why ‘AI Exposure’ is Not One Thing
AI is Not Pedagogy: The Human in the Room Matters

Thanks for reading Data in Motion.

Subscribe for free to receive new posts.