Abstract

Background: Artificial intelligence (AI) is increasingly transforming how humans work, communicate, generate knowledge and make decisions. Yet public and scientific debates frequently frame AI in terms of automation, competition and the potential replacement of human capabilities. This perspective raises a fundamental question: rather than asking what AI might replace, should we be examining how human and artificial intelligence can be deliberately combined to expand what humanity can achieve?

Approach: This opinion article critically examines AI within the historical evolution of human technologies, distinguishing computational capabilities from the embodied, biological and relational foundations of human cognition. Drawing on evidence from healthcare digitalisation, implementation science and human-centred evaluation, it explores why technological sophistication does not necessarily translate into meaningful human or societal benefit. The discussion integrates insights from PROLIFERATE, PROLIFERATE_AI, and AWARE_AI, alongside HUMMBER, an emerging human-governed evaluation methodology that examines digital and AI-enabled interventions as they change in use. HUMMBER illustrates how longitudinal observation, human experiences, contextual inquiry, evidence integration, reproducible computational procedures and decision accountability can be brought together to investigate evolving human-technology relationships.

Conclusion: AI's transformative potential lies not simply in improving computational performance or automating existing processes, but in deliberately designing and evaluating the evolving relationships between human and artificial intelligence. This requires approaches that integrate diverse forms of knowledge, preserve human agency, recognise uncertainty and establish accountability for decisions. Rather than treating AI as an independent competitor to human intelligence, we should investigate human-AI interactions as dynamic sociotechnical systems. The central challenge is not whether AI has become part of how we think, but how we can design the most constructive, equitable and accountable symbiosis between human and artificial intelligence.

Technology has always changed what humans can achieve

We created artificial intelligence, yet much of our public conversation sounds as though it has arrived from somewhere beyond us, threatening to replace the very intelligence that made it possible. We worry about losing our jobs, our creativity, and even our capacity to think, while paying far less attention to deciding what we actually want to achieve with this extraordinary technology. I find this contradiction fascinating, particularly because history offers us many examples of technologies that transformed human capabilities rather than simply eliminating them.

Did calculators stop us from thinking mathematically? Did the printing press stop us from writing? Did photography bring painting to an end? Perhaps the more important question is why we keep treating technological transformation primarily as a threat to what we already do, rather than an opportunity to discover what else we could become capable of doing.

When calculators and computers became widely available, mathematics did not disappear. We continued developing mathematical theories, tackling increasingly complex problems and expanding what we could calculate. Similarly, the printing press did not end writing; it transformed how knowledge could be reproduced, shared and accessed. Historical research has linked early adoption of printing technology to substantially greater subsequent urban growth in European cities, although not everyone had equal access to these opportunities [1].

Industrialisation transformed craftsmanship, displaced occupations and created new possibilities for design, production and creativity. Photography changed how we represented the world, yet painting continued to evolve. These technologies did not simply replace human activities; they challenged established practices and opened possibilities that had previously been difficult to imagine.

Then came the internet and social media, fundamentally transforming communication and access to information. Traditional media did not disappear, although their roles and business models changed dramatically. People gained opportunities to reach global audiences without depending entirely on newspapers, television networks or publishing houses. New voices, professions, communities and forms of participation emerged, alongside challenges involving misinformation, inequality and the concentration of digital power. Research on information diffusion on Twitter, for example, found that false news spread faster and farther than true news during the period examined [2].

Throughout history, technological transformations have changed what we can do, how we do it, and who can participate. Some occupations disappeared, others evolved, and entirely new fields of knowledge and human activity emerged. Technological progress has never been without consequences, but it hasn't simply replaced humans.

Artificial intelligence is artificial for a reason

Something important often gets lost in conversations about artificial intelligence: the word artificial. AI is not a mysterious intelligence that emerged independently of human knowledge, mathematics or engineering. It results from decades of scientific development involving mathematical models, computational algorithms, statistical learning, software engineering and increasingly sophisticated technological infrastructures.

From programmed rules to learned representations

Consider the evolution from traditional rule-based systems to machine learning and modern deep neural networks. Earlier AI systems often relied on explicitly programmed rules, while machine learning let computers identify statistical relationships and patterns from data without programming every possible response [3]. Deep learning extended these capabilities through multilayer neural networks that learn complex representations from large amounts of information [4,5].

Even today's most advanced large language models, including systems based on transformer architectures, operate through computational processes grounded in mathematics [6,7]. Their capabilities emerge from learned parameters, training objectives, optimisation algorithms and interactions with data, rather than from programmers writing every sentence or specifying every reasoning step. When these systems generate text, analyse information or solve problems, they perform computations shaped by their training and the context in which they operate.

Autonomy is not independence

Advanced AI systems can generate computer code, use external tools, evaluate outputs and participate in processes that modify or improve software. Automated machine learning, reinforcement learning and AI-assisted programming can enable systems to optimise aspects of their operation or help develop other AI systems [8–11]. However, these processes depend on computational architectures, learning algorithms, available resources, evaluation mechanisms and operational environments that enable such adaptation.

Autonomy in performing a task is not the same as independence from the systems that make it possible. AI can generate solutions its developers did not anticipate, make decisions within defined operational boundaries, and even contribute to the development of other AI systems. Yet none of these capabilities exists independently of the mathematical foundations, algorithms and learning methods developed through human scientific knowledge [3–7], or the physical infrastructure, computational resources and technological ecosystems that humans create, maintain and govern [12,13].

Even highly autonomous AI systems depend on human decisions about their development, objectives, deployment, access to resources and conditions of use. Furthermore, although the possibility of subjective experience or consciousness in AI remains an open scientific and philosophical question [14], neither is required to explain its demonstrated computational capabilities or its dependence on human-created systems. Increasing autonomy therefore reflects an expansion of what our technological systems can accomplish, not evidence that AI has become independent of the human knowledge, resources and decisions that make its existence and operation possible.

The embodied sources of human knowledge

A further dimension of this argument is often overlooked: human knowledge does not arise from logical reasoning alone. It is also grounded in embodied experience. Humans are biological organisms whose cognition is shaped by our bodies, senses, emotions and continuous interaction with our environments [15]. Long before we formulate a thought, our bodies register hunger, fatigue, pain, warmth, the rhythm of a conversation and the presence of another person.

Interoception, sensing our own internal bodily states, is closely connected to awareness, emotion, and decision-making [16]. Much of what we know is shaped by being alive, situated and physically vulnerable, through processes that explicit reasoning alone cannot explain. Our experiences emerge through continuous interactions between biological, psychological, social and environmental systems.

Consider the interactions between the human body and its microbiota. The human body contains tens of trillions of microbial cells, with estimates suggesting that bacterial and human cell numbers are of a similar order of magnitude [17]. Microbial activity, metabolites, immune responses, neural signalling and behavioural processes can influence one another through complex pathways, many of which remain incompletely understood [18,19].

Changes in microbial activity may influence physiological signalling and, under some conditions, contribute to changes in affective states and behaviour. Behavioural and environmental changes can, in turn, modify biological conditions, including those influencing microbial communities [18,19]. These relationships are not universally deterministic, and evidence for specific causal pathways varies considerably, yet they illustrate how human experience can emerge through interactions across multiple levels of living systems.

AI can help us investigate these relationships, but modelling a biological interaction is not equivalent to participating in it as a living organism or acquiring the implicit, non-conscious knowledge that emerges through embodied biological experience. Even when computational systems incorporate sensors, robotics and real-time environmental feedback, their observations remain constrained by what their architectures, instruments and measurement processes can make accessible. A sensor can detect a molecular signal, physiological change or behavioural pattern, but it does not reproduce the biological, emotional and relational processes through which those phenomena acquire significance in a living person.

This is not an argument against AI's intelligence or scientific value. Computational models may reveal relationships that humans could not identify unaided, including relationships that challenge our existing assumptions. Rather, mathematical representation, biological participation and subjective experience are not interchangeable. Recognising their differences may help us develop AI systems that complement different forms of human knowledge instead of assuming that all knowledge can be reduced to the information currently available to a computational model.

For me, recognising AI's artificial nature does not diminish its extraordinary capabilities. On the contrary, it helps us understand how far human knowledge has progressed and how much more we may achieve by developing these technologies responsibly. AI is neither simply a collection of programmed instructions nor an entirely independent entity detached from human technological development. It is a powerful, evolving computational capability embedded within our broader sociotechnical system.

Beyond automation: are we asking the right questions?

Why are we spending so much energy discussing whether artificial intelligence will take our jobs [20], make us stop thinking or replace our creativity? These concerns deserve serious consideration, particularly when they involve employment, equity, accountability and the concentration of power. However, I believe we have an equally important responsibility: to redirect more of our collective energy towards shaping what we want AI to help us achieve.

AI already exists, and it is increasingly embedded in our workplaces, institutions, communication systems and everyday lives. Resisting it won't determine how it develops, nor will uncritically adopting it guarantee better outcomes. What matters now is whether we actively design, implement and govern AI in ways that extend human capabilities and contribute to the world we want to build.

The lesson healthcare digitalisation should have taught us

Healthcare digitalisation offers an important lesson here. For years, substantial investments in electronic health records focused, in part, on reproducing established documentation and administrative processes in digital environments. Although these systems brought important benefits, digitising existing workflows did not automatically transform how healthcare was delivered, experienced or improved. In some settings, digitalisation also introduced administrative burdens, with a time-and-motion study in ambulatory practice reporting nearly two hours of electronic health record and desk work for every hour of direct patient care [21].

Australia has experienced substantial challenges in implementing complex information systems. The Queensland Health payroll implementation became a prominent example of major information-system failure, described as the largest admitted IT project failure in the Southern Hemisphere, with total implementation, stabilisation, and maintenance costs estimated at A$1.25 billion and serious consequences for affected health workers [22]. Later, Queensland's integrated electronic medical record programme faced a projected cost overrun exceeding A$250 million; although the government reported clinical benefits, the Auditor-General questioned whether value for money had been demonstrated [23].

These programmes differed in purpose and outcomes, but they illustrate how technological investment can be undermined by weaknesses in implementation, governance and alignment with the people expected to use the systems. Their lessons extend beyond software performance to the organisational and human conditions required for meaningful transformation. Australia's national My Health Record offers another perspective: an analysis in a Melbourne private hospital emergency department found that clinicians accessed the national record during 19.6% of patient presentations [24]. That finding does not show the system lacked value, but it illustrates why digital infrastructure alone cannot be treated as evidence of consistent clinical use.

We should be careful not to repeat these mistakes with AI. If we limit our ambition to making existing tasks faster, producing the same documents more efficiently or automating the processes we already have, we may miss much of its transformative potential. A more interesting opportunity is to question those processes and imagine different ways of understanding problems, generating knowledge, supporting decisions and improving outcomes.

What could healthcare become if we designed human-AI collaboration around people's needs rather than existing workflows? What could education, science, communication and creativity become if we used AI to expand participation, explore complexity and pursue questions previously beyond our practical capabilities? These are the conversations I believe deserve more of our attention, imagination and investment.

From technological performance to human and AI collaboration

This perspective is closely connected to my research into the implementation and evaluation of innovations within complex adaptive health systems. Through PROLIFERATE, collaboratively developed with colleagues, we explored how different forms of knowledge and evaluation could inform the co-design, measurement and optimisation of innovations in complex healthcare environments [25]. Its subsequent evolution into PROLIFERATE_AI, collaboratively co-developed under my leadership, extended this work by integrating human-centred evaluation with computational modelling to investigate how AI technologies may be understood, experienced and adopted by their intended users [26,27].

Our research in emergency cardiac care has examined the relationship between technological capabilities, human experiences and intended use. In the RAPIDx AI context, PROLIFERATE_AI helped investigate how clinicians' comprehension, preferences, trust and other behavioural factors relate to their intended engagement with AI-supported clinical decision-making [27,28]. These findings do not establish that AI automatically improves clinical outcomes. Rather, they illustrate why technical performance alone cannot tell us whether a technology will be meaningfully integrated into healthcare.

We have also explored computational approaches beyond the original cardiac context. In interprofessional education, PROLIFERATE_AI has integrated quantitative and qualitative evidence to understand learning experiences, barriers and opportunities for improvement [29]. Our AWARE_AI research extends this broader methodological direction towards modelling person-centred wellbeing through the Fundamentals of Care framework, while reinforcing the importance of understanding what computational predictions can and cannot tell us about human experiences [30].

More recently, as a JBI Evidence Ambassador for World Evidence-Based Healthcare Day 2026, themed "Evidence and AI: People at the Centre", I contributed an article titled Making human/AI systems an object of evidence [31]. In that contribution, I argued that the interactions between humans and AI should themselves become objects of systematic investigation. Rather than evaluating technologies, clinical outcomes, organisational processes, and human experiences as isolated components, we need approaches that examine how they interact, change, and influence one another. This requires moving beyond whether AI performs a task successfully to understanding how human-AI relationships generate evidence, shape decisions, and produce consequences within complex healthcare systems.

These experiences have reinforced my view that technological capability alone is insufficient. AI systems must be understood within the human, organisational and social environments in which they operate, recognising that meaningful implementation depends on interactions among people, technologies and contexts. The opportunity is not simply to make existing processes more efficient, but to investigate questions and relationships that our previous methods and practical capabilities may have struggled to address.

From human-AI interaction to evaluating systems that change

If we want to understand human-AI collaboration, where can we observe these relationships as they develop, rather than simply measuring a technology's performance at a particular moment? This raises the challenge of studying human and technological change together, especially when interactions evolve during implementation rather than remaining fixed after a system is introduced.

Digital behaviour change interventions offer a particularly interesting example. These technologies are increasingly becoming part of people's everyday experiences. They can remind us to take medication, encourage physical activity, ask questions about our wellbeing, suggest changes to our routines, provide personalised feedback and engage us in conversations. When AI is incorporated, these interactions can become more responsive to information about the individual, their circumstances, and their behaviour over time.

In these settings, technology does not simply deliver information. It participates in an ongoing interaction with a person whose needs, motivations, opportunities, and circumstances may also change. A reminder that helps today may be unnecessary tomorrow. A recommendation that works for one person may be impractical for another. An intervention may function as intended while becoming less meaningful, accessible, or acceptable to the people using it.

This is precisely where the limitations of evaluating technological performance alone become apparent. We have established approaches for evaluating digital health interventions, implementation processes, behaviour change mechanisms and adaptive interventions. Yet bringing these forms of evidence together remains challenging when the intervention itself is changing during use. A favourable final outcome does not necessarily explain how it was achieved, whether everyone had comparable opportunities to benefit, what changed along the way or which adaptations were justified.

If AI-enabled systems become increasingly capable of adapting their recommendations, interactions, or components, how do we evaluate those changes as they happen? Addressing this question requires attention not only to the evolving technology, but also to the people, contexts and decisions involved in its use.

HUMMBER: evaluating systems that change while we use them

This question motivated our interdisciplinary research team to develop HUMMBER (Human-centred, User-informed, Multimethod Micro-randomised Behaviour Evaluation and Response), a human-governed methodology that brings together complementary approaches to evaluate digital behaviour change interventions, including AI-enabled interventions. Our methodological development is described in a research manuscript submitted to Research Square for preprint publication on 14 October 2026, with public preprint publication pending [32].

HUMMBER is an emerging methodological contribution, not a framework that has already been validated through prospective implementation in real-world services. The underlying challenge is straightforward to describe, even if addressing it is methodologically complex: how can we evaluate what is happening when the technology, the people using it and the circumstances surrounding its use are all changing?

Six questions that make changing systems accountable

HUMMBER approaches this challenge through six connected questions. Together, they provide a structure for examining change, interpreting evidence and clarifying the responsibilities associated with decisions.

1. What changed? Identify changes in outcomes, engagement, intervention delivery or other relevant processes over time.

2. For whom? Examine how experiences, opportunities and outcomes may differ across people and groups.

3. Under what circumstances? Consider the contextual conditions in which those changes occur.

4. What might explain the change? Investigate possible explanations using relevant human, behavioural, contextual and computational evidence.

5. Did a particular component make a difference? Examine component effects when the evaluation design permits valid comparisons, rather than assuming that observed associations establish causality.

6. What can evaluators conclude, and what may authorised decision-makers consider next? Establish the strength and limitations of the available evidence and preserve human responsibility for subsequent decisions.

These questions are operationalised through six interconnected functions: Observe, Detect, Ask, Explain, optional Experiment and Govern, supported by a cross-cutting Update process. Together, these functions provide a structured approach to investigating changes while maintaining visibility of the evidence, uncertainty and decision responsibilities.

Observe and Detect establish what is happening and whether meaningful changes are emerging. Ask brings human experiences and contextual information into the evaluation. Explain examines competing interpretations rather than treating the first detected pattern as a definitive answer. Experiment, when appropriate and feasible, supports valid comparisons of intervention components. Govern makes the evidence status, uncertainty and decision responsibilities explicit. Update connects these functions as new information becomes available and circumstances evolve.

The purpose is not to claim that each of these techniques is new. Longitudinal analysis, process evaluation, realist approaches, human-centred research, Bayesian evidence updating and micro-randomised trials already offer valuable methodological capabilities. HUMMBER's contribution lies in integrating these capabilities into a traceable, human-governed evaluation process, supported by an optional reproducible computational pipeline.

Once the evaluation has been co-designed with the people involved, and its methodological specifications, parameters and decision rules have been established, the software can support the consistent execution of computational procedures, evidence updating and reproducible analyses. This allows computational capabilities to extend what human evaluators can investigate while preserving human responsibility for interpreting the evidence, assessing its limitations and authorising subsequent decisions.

Co-design, changing contexts and human decision authority

Importantly, this process begins with people. Consumers, practitioners, service decision-makers, technical teams and evaluators contribute to determining what matters, what information should be collected, which comparisons are appropriate, what assessment burden is acceptable and who has authority to act on the evidence. The method also incorporates version control so that changes in the intervention, evaluation conditions and decision responsibilities can be documented rather than overlooked.

Computational tools can support this process by organising longitudinal information, detecting patterns, examining possible explanations, updating evidence and conducting appropriate analyses. However, a computational result is not automatically a causal explanation, and a prediction is not automatically a recommendation that should be implemented.

This distinction is central to HUMMBER. The methodology separates what can be described, what may be associated, what might plausibly explain an observation and what can be causally inferred under an appropriate evaluation design. Where evidence is insufficient, that limitation should remain visible.

Our team has explored selected methodological functions through a controlled synthetic computational demonstration, described in the submitted manuscript [32]. This work is intended to investigate methodological feasibility under specified simulated conditions. It does not establish clinical effectiveness or validate the methodology in real-world adaptive AI systems. Those questions require further prospective research.

Designing the conditions for constructive human-AI collaboration

I introduce HUMMBER here because it illustrates something much broader than the evaluation of digital behaviour change interventions. It offers one example of how we might begin investigating the evolving relationships between people, technologies, evidence and decisions rather than treating these as separate concerns.

If AI is becoming part of how we think, decide, communicate and act, we need to investigate not only what AI systems can accomplish, but what happens through their interactions with people. We need to understand how those interactions change, who benefits, whose experiences are overlooked, what evidence supports our interpretations and who remains responsible for decisions.

Computational systems can help us investigate patterns and relationships that would be difficult to examine through unaided human analysis. They can extend our ability to observe complex processes, integrate different forms of information and explore possibilities that might otherwise remain invisible.

Yet humans contribute something different and equally essential: lived and embodied experience, contextual understanding, values, ethical judgement and responsibility for determining what outcomes are worth pursuing. The opportunity is not to treat these contributions as interchangeable, but to design methods and systems that allow them to complement one another while making their respective roles and limitations explicit.

The broader scientific challenge is therefore not simply to evaluate more technologies. It is to reconsider what we recognise as evidence when human and computational capabilities become increasingly interconnected. Evidence should help us understand not only whether an intervention works, but how people and technologies influence one another, how circumstances change, whose experiences are represented, and what conclusions can legitimately support decisions. This is where implementation science, evidence-based healthcare, human-centred research and computational methods have an important opportunity to advance together.

For me, HUMMBER represents one attempt by our interdisciplinary team to address that challenge. It is not the final answer, nor is it an autonomous AI system. It is a proposed methodological structure for making changing human-technology relationships more observable, interpretable and accountable.

This returns us to the article's central question. We should not only develop AI systems capable of learning, predicting and adapting. We should also develop the scientific, ethical, and methodological capabilities that let people understand, question, evaluate, and govern those adaptations as they occur.

If we want a constructive symbiosis between human and artificial intelligence, we must design not only the technologies, but also the relationships, evidence and responsibilities through which they become part of our lives. This requires approaching human-AI collaboration as something we can actively investigate and shape, rather than as an inevitable consequence of technological progress.

Designing the symbiosis we already live in

For me, artificial intelligence is not something separate from humanity that we must inevitably compete against. It is a technology created through human knowledge and ingenuity, and one that is becoming part of our broader human and technological ecosystem. Like many tools before it, AI can extend our abilities, although the direction and consequences of that extension depend on the choices we make.

This does not mean ignoring risks, accepting every application or assuming that technological progress automatically produces social progress. Quite the opposite, it means taking responsibility for how AI is developed, who benefits from it, whose perspectives shape it and what consequences we are willing to accept. Meaningful innovation requires more than technological capability; it requires human purpose, critical thinking, collaboration, evaluation and accountability.

I believe our responsibility now is not simply to prepare for a future shaped by AI, but to actively shape the future we want to create with AI. The greatest opportunity may not lie in using increasingly sophisticated technologies to reproduce what humans already do. It may lie in discovering what becomes possible when embodied human knowledge, lived experience, ethical judgement and computational capabilities are deliberately brought together. To realise that potential, we need to investigate these relationships as evolving systems, generate evidence about what they make possible, and ensure that the people affected can meaningfully shape their development.

Human beings have always thought with tools. We extended our memory into writing, our voices into the telephone, our knowledge into libraries and search engines, and now our reasoning into artificial intelligence. The relationship between human minds and the tools we build is as old as civilisation itself, and AI doesn't break that story; it may be one of its most consequential chapters.

That is why I don't ask whether we will think with AI. We already do, just as we have always thought through what we create.

The question is what we are doing today to design the best symbiosis between AI and human intelligence our future could hold. I have shown you one attempt. Where are the others?

References

1. Dittmar JE. Information technology and economic change: the impact of the printing press. Q J Econ. 2011;126(3):1133–1172. https://doi.org/10.1093/qje/qjr035

2. Vosoughi S, Roy D, Aral S. The spread of true and false news online. Science. 2018;359(6380):1146–1151. https://doi.org/10.1126/science.aap9559

3. Jordan MI, Mitchell TM. Machine learning: trends, perspectives, and prospects. Science. 2015;349(6245):255–260. https://doi.org/10.1126/science.aaa8415

4. LeCun Y, Bengio Y, Hinton G. Deep learning. Nature. 2015;521(7553):436–444. https://doi.org/10.1038/nature14539

5. Rumelhart DE, Hinton GE, Williams RJ. Learning representations by back-propagating errors. Nature. 1986;323(6088):533–536. https://doi.org/10.1038/323533a0

6. Vaswani A, Shazeer N, Parmar N, et al. Attention is all you need. In: Guyon I, von Luxburg U, Bengio S, et al., editors. Advances in Neural Information Processing Systems 30 (NeurIPS 2017). Long Beach, CA: Curran Associates; 2017. p. 5998–6008.

7. Singhal K, Azizi S, Tu T, et al. Large language models encode clinical knowledge. Nature. 2023;620(7972):172–180. https://doi.org/10.1038/s41586-023-06291-2

8. Mnih V, Kavukcuoglu K, Silver D, et al. Human-level control through deep reinforcement learning. Nature. 2015;518(7540):529–533. https://doi.org/10.1038/nature14236

9. Silver D, Schrittwieser J, Simonyan K, et al. Mastering the game of Go without human knowledge. Nature. 2017;550(7676):354–359. https://doi.org/10.1038/nature24270

10. Elsken T, Metzen JH, Hutter F. Neural architecture search: a survey. J Mach Learn Res. 2019;20(55):1–21.

11. Feuerriegel S, Hartmann J, Janiesch C, Zschech P. Generative AI. Bus Inf Syst Eng. 2024;66(1):111–126. https://doi.org/10.1007/s12599-023-00834-7

12. Masanet E, Shehabi A, Lei N, Smith S, Koomey J. Recalibrating global data center energy-use estimates. Science. 2020;367(6481):984–986. https://doi.org/10.1126/science.aba3758

13. Kaack LH, Donti PL, Strubell E, Kamiya G, Creutzig F, Rolnick D. Aligning artificial intelligence with climate change mitigation. Nat Clim Chang. 2022;12(6):518–527. https://doi.org/10.1038/s41558-022-01377-7

14. Seth AK, Bayne T. Theories of consciousness. Nat Rev Neurosci. 2022;23(7):439–452. https://doi.org/10.1038/s41583-022-00587-4

15. Wilson M. Six views of embodied cognition. Psychon Bull Rev. 2002;9(4):625–636. https://doi.org/10.3758/BF03196322

16. Craig AD. How do you feel—now? The anterior insula and human awareness. Nat Rev Neurosci. 2009;10(1):59–70. https://doi.org/10.1038/nrn2555

17. Sender R, Fuchs S, Milo R. Revised estimates for the number of human and bacteria cells in the body. PLoS Biol. 2016;14(8). https://doi.org/10.1371/journal.pbio.1002533

18. Cryan JF, Dinan TG. Mind-altering microorganisms: the impact of the gut microbiota on brain and behaviour. Nat Rev Neurosci. 2012;13(10):701–712. https://doi.org/10.1038/nrn3346

19. Collins SM, Surette M, Bercik P. The interplay between the intestinal microbiota and the brain. Nat Rev Microbiol. 2012;10(11):735–742. https://doi.org/10.1038/nrmicro2876

20. Frey CB, Osborne MA. The future of employment: how susceptible are jobs to computerisation? Technol Forecast Soc Change. 2017;114:254–280. https://doi.org/10.1016/j.techfore.2016.08.019

21. Sinsky CA, Colligan L, Li L, et al. Allocation of physician time in ambulatory practice: a time and motion study in 4 specialties. Ann Intern Med. 2016;165(11):753–760. https://doi.org/10.7326/M16-0961

22. Eden R, Sedera D. The largest admitted IT project failure in the Southern Hemisphere: a teaching case. In: Proceedings of the 35th International Conference on Information Systems (ICIS 2014); Auckland, New Zealand. 2014.

23. Australian Broadcasting Corporation. Queensland digital hospitals program facing $250m cost blowout [Internet]. ABC News; 2018 Dec 3 [cited 2026 Oct 25]. Available from: https://www.abc.net.au/news/2018-12-04/digital-hospitals-blow-out-auditor-general-scathing-report/10581322

24. Mullins AK, Morris H, Enticott J, Ben-Meir M, Rankin D, Mantripragada K, Skouteris H. Use of My Health Record by clinicians in the emergency department: an analysis of log data. Front Digit Health. 2021;3:725300. https://doi.org/10.3389/fdgth.2021.725300

25. Pinero de Plaza MA, Yadav L, Kitson A. Co-designing, measuring, and optimizing innovations and solutions within complex adaptive health systems. Front Health Serv. 2023;3:1154614. https://doi.org/10.3389/frhs.2023.1154614

26. Pinero de Plaza MA, Lambrakis K, Marmolejo-Ramos F, et al. PROLIFERATE_AI: A prediction modeling method to evaluate artificial intelligence in meeting end-user-centric goals around better cardiac care. Heart Lung Circ. 2023;32(Suppl 3)–S365. https://doi.org/10.1016/j.hlc.2023.06.833

27. Pinero de Plaza MA, Lambrakis K, Marmolejo-Ramos F, et al. Human-centred AI for emergency cardiac care: Evaluating RAPIDx AI with PROLIFERATE_AI. Int J Med Inform. 2025;196:105810. https://doi.org/10.1016/j.ijmedinf.2025.105810

28. Pinero de Plaza MA. PROLIFERATE_AI reveals behavioural drivers of RAPIDx AI clinical interface use in emergency care. Heart Lung Circ. 2026;35(Suppl 3)–S422. https://doi.org/10.1016/j.hlc.2026.07.627

29. Pinero de Plaza MA. Advancing digital evaluation for inter-professional education: A PROLIFERATE_AI mixed-methods analysis. Stud Health Technol Inform. 2025;333:82–86. https://doi.org/10.3233/SHTI251580

30. Pinero de Plaza MA, Kitson A, Porcel-Gálvez AM, Visvanathan R, Beleigoli A, Barrera Causil CJ, et al. AWARE_AI, a machine-learning-based reproducible workflow modelling the Fundamentals of Care framework and patient-reported wellbeing: a cross-sectional feasibility study. J Adv Nurs. 2026. Accepted for publication. https://doi.org/10.1111/jan.70800

31. Pinero de Plaza MA. Making human/AI systems an object of evidence. World Evidence-Based Healthcare Day 2026 [Internet]. 2026 Sep 30. JBI-founded global evidence-based healthcare campaign. Available from: https://doi.org/10.70253/FSVN1820

32. Pinero de Plaza MA, Halkett I, Ranasinghe D, Barrera Causil CJ, Dollard J, Visvanathan R. How can digital and AI-enabled behaviour change interventions be evaluated while they are changing? HUMMBER: identifying what changed, for whom, why and with what effect. Research manuscript under publication consideration. 2026. Submitted to Research Square for preprint publication on 14 October 2026. Research Square preprint ID: rs-11369973. Public preprint publication pending.