Academic papers — especially ones that you’ve written — look different depending on when you read them. Papers written in 2019, before COVID, before generative AI, before agentic AI systems became operationally real, carry assumptions about the world that the following six years tested in ways nobody might have anticipated. That retrospective effect cuts both ways: the same paper can look clear-sighted in one paragraph and overtaken in the next. The honest response is not to try to resolve that tension but to sit with it — and to be clear about how much the value of this kind of work depends on timing as much as content.
I’ve been part of three collaborative research efforts on AI that, taken together, trace a thread from 2019 to 2025: a paper on AI challenges, opportunities and an agenda for research, practice and policy, first published in the International Journal of Information Management in 2019; a paper responding to the arrival of publicly available generative AI tools, especially ChatGPT, also in IJIM, in early 2023; and a paper on AI agents and agentic systems, published in the Journal of Computer Information Systems in 2025. I’ve never written about any of them here in any great detail, partly because the years since have been full ones (more time doing research and policy work in this space than blogging about it) and partly because this kind of reflection benefits from some distance. This feels like the right moment to do so, not as an announcement, but as a way of tracing how my own thinking about AI as a complex socio-technical system has had to keep pace with how fast the tools, systems and infrastructure have changed.
The 2019 paper was a broad multidisciplinary synthesis: more than fifty international contributors offering perspectives on AI across business, education, government, healthcare and social science. It appeared into a specific moment worth recalling: national AI strategies were emerging but thin, regulatory frameworks were almost non-existent, and what governance attention existed was largely absorbed by GDPR, which had recently come into force. The dominant public anxiety about AI was automation and labour market disruption (what AI would do to workers) rather than the harder systems-level questions of how institutions, infrastructure and people would need to adapt together. My own contribution drew as much on a systems engineering perspective as a policy one: treating AI adoption not as a single technology decision but as the integration of a new, evolving component into existing socio-technical systems, with all the questions of skills, workforce transition, institutional capability and governance that integration implies.
What the paper did well was establish, with some breadth and authority, that AI was not just a technical problem with a technical solution, nor a policy problem that technology would simply resolve; it was, and remains, a complex socio-technical systems challenge, in which technology, institutions, people and governance structures evolve together and cannot be understood in isolation. That framing has held. But before the research agenda it sketched could be properly tested, the COVID-19 pandemic arrived: an uncontrolled, real-world stress test of exactly these systems. Digital technologies were deployed across public services, healthcare and education at extraordinary speed across the world, reconfiguring how these systems operated almost overnight. The assurance, governance and oversight structures around them did not move at anything like the same pace. That gap — between how fast systems can change and how slowly the structures around them adapt — is the kind of question I find genuinely interesting, and one that sits behind a good deal of what I work on now, in both research and in supporting policy.
What we did not fully anticipate was the speed with which AI would move from specialist systems into everyday institutional workflows, nor the extent to which publicly accessible generative AI would collapse the distance between technical capability, public use, regulatory concern and institutional response. The 2019 paper treated AI as a socio-technical systems challenge, but the period after 2022 demonstrated how quickly those systems could change once powerful capabilities became widely available. Many institutions found themselves adapting in months rather than years.
The serendipity of the 2023 generative AI paper is, if anything, more legible in retrospect. ChatGPT (GPT‑3.5) launched publicly in late November 2022; our paper appeared in March 2023. In systems terms, what had just happened was that a new, highly capable, probabilistic component had been integrated, almost instantly, and at global scale, into millions of existing workflows, institutions and systems, with none of the integration testing, risk assessment or assurance processes that would normally precede the deployment of a new (non-deterministic) technical component into any system. Universities, government departments, regulators, and millions of people simply trying to work out how to use the tool responsibly, were all asking some version of the same question: what do we do about this? There was almost nothing substantive to point to. A rapid, multidisciplinary response, drawing on perspectives from business, education, law, healthcare and public policy, provided a shared reference point at exactly the moment one was needed.
This is worth naming explicitly, because it describes something real about what this kind of research does and does not do. Complex systems engineering has always depended on bringing together technical, human-factors and governance perspectives at the point of greatest uncertainty, and multidisciplinary synthesis papers perform a version of that same function for research and policy communities. They do not produce original empirical findings or resolve contested questions; they convene expertise across disciplines at moments when no shared framework yet exists, giving researchers, practitioners and policymakers something to build on, test and refine. My focus across these papers, on skills, institutions, public services and governance, was always about ensuring that dimension was present in the synthesis from the outset. But papers written at speed, with many international contributors convened by my former Swansea University colleague Yogesh Dwivedi, identify questions more reliably than they answer them. Understanding what that kind of contribution is, and isn’t, matters when these papers get picked up by processes that want clearer answers than research can yet provide.
The 2025 paper extends the thread into different, and in some respects more demanding, territory: AI agents and agentic systems (autonomous systems capable of operating, learning and making decisions with minimal human intervention). The shift matters because it changes the nature of the engineering and governance problem together. With generative AI, the central questions were about how humans use a capable new component, and what that meant for knowledge, trust and accountability. With agentic systems, the questions become architectural: where system boundaries sit when an AI component can initiate actions across multiple systems; how human oversight is designed into a system operating faster than any human-in-the-loop process can realistically function; how failure modes are identified and contained when behaviour is adaptive rather than fixed. My contribution focused on how these questions play out in public sector and education contexts, where the consequences of getting system design, assurance and oversight wrong are not abstract. More speculative possibilities, including systems that can modify or optimise aspects of their own operation, extend these questions further still, raising the assurance and oversight challenge to a qualitatively different level, and have moved from a largely theoretical concern to an active question in both research and policy circles.
These questions are not unique to any one country. The international AI summits starting at Bletchley Park in November 2023 have brought together governments, industry leaders and researchers from across the US, China, India, the EU and many others, around a common underlying engineering challenge: how to verify, assure and govern systems whose behaviour is probabilistic, adaptive, and increasingly embedded in critical infrastructure and public institutions, at a pace that outstrips traditional regulatory and assurance cycles. That international breadth is also reflected, in a smaller way, in the authorship of the 2019 and 2023 papers themselves, drawing contributors from across Europe, North America, Asia and the Middle East, which is part of why the convening function described earlier matters: these are genuinely global questions, and no single national or disciplinary perspective can answer them alone. The EU AI Act (much of which becomes applicable from 2 August 2026, with a staggered implementation timetable and recent amendments delaying some specific obligations) remains one of the most significant legislative attempts at a regulatory framework for this challenge. Other jurisdictions are taking different approaches, each navigating the same underlying tension: how to build or maintain competitive advantage in AI while putting in place the assurance, regulatory and institutional structures needed to protect citizens, society, culture and the economy from its risks. It remains genuinely unclear which combinations will prove durable, or whether the two goals are as compatible as most national strategies assume. That is not a judgement on any one approach; these are fast-moving, live questions. It is a description of a problem that engineering, governance and policy communities everywhere are still working through, together and separately.
The honest version of what I think these three papers contributed is this: they named, early enough and with enough breadth to matter, that AI was an engineering and systems challenge as much as a policy one; that skills, institutions and public services were central rather than peripheral to how AI would be integrated into how societies function; and that questions of assurance, oversight and accountability needed to be designed in from the outset, not added afterwards. The progression from AI, to generative AI, to agentic systems represents an escalating version of the same underlying challenge; and the questions raised in 2019 have not been answered so much as they have become more urgent, more technically demanding, and more international. Those are the questions — about assurance, governance and the design of complex AI-enabled systems at scale — that this blog, and the research and policy work that sits alongside it, will keep returning to. They are also, I think, where the field as a whole still has the most work to do.
- The 2019 paper: Dwivedi, Y. K., Hughes, L., Ismagilova, E., Aarts, G., Coombs, C., Crick, T., Duan, Y., Dwivedi, R., Edwards, J., Eirug, A., Galanos, V., Ilavarasan, P. V., Janssen, M., Jones, P., Kar, A. K., Kizgin, H., Kronemann, B., Lal, B., Lucini, B., … Williams, M. D. (2021). Artificial Intelligence (AI): Multidisciplinary perspectives on emerging challenges, opportunities, and agenda for research, practice and policy. International Journal of Information Management, 57, 101994. https://doi.org/10.1016/j.ijinfomgt.2019.08.002
- The 2023 paper: Dwivedi, Y. K., Kshetri, N., Hughes, L., Slade, E. L., Jeyaraj, A., Kar, A. K., Baabdullah, A. M., Koohang, A., Raghavan, V., Ahuja, M., Albanna, H., Albashrawi, M. A., Al-Busaidi, A. S., Balakrishnan, J., Barlette, Y., Basu, S., Bose, I., Brooks, L., Buhalis, D., … Wright, R. (2023). Opinion Paper: “So what if ChatGPT wrote it?” Multidisciplinary perspectives on opportunities, challenges and implications of generative conversational AI for research, practice and policy. International Journal of Information Management, 71, 102642. https://doi.org/10.1016/j.ijinfomgt.2023.102642
- The 2025 paper: Hughes, L., Dwivedi, Y. K., Malik, T., Shawosh, M., Albashrawi, M. A., Jeon, I., Dutot, V., Appanderanda, M., Crick, T., De’, R., Fenwick, M., Gunaratnege, S. M., Jurcys, P., Kar, A. K., Kshetri, N., Li, K., Mutasa, S., Samothrakis, S., Wade, M., & Walton, P. (2025). AI Agents and Agentic Systems: A Multi-Expert Analysis. Journal of Computer Information Systems, 65(4), 489–517. https://doi.org/10.1080/08874417.2025.2483832
Digital Society & Policy 