Making human/AI systems an object of evidence

Author: Dr Maria Alejandra Piñero de Plaza¹

1. Caring Futures Institute, Flinders University

Artificial intelligence (AI) is usually introduced through the language of disruption, as we ask what it will automate, which tasks it will replace and which work may no longer be needed. These questions capture only the most immediate layer of the change now underway; the more consequential opportunity lies in understanding what humans can investigate as searching, calculating, modelling and detecting patterns become increasingly routine. AI should therefore be approached not primarily through fear, but as a change in human capability similar to that when calculators transformed mathematics. Calculators did not diminish mathematical thinking; they reduced the burden of routine calculation and let mathematicians and ordinary people work beyond it.

In the same way, AI may let us direct more human attention towards interpretation, relationships, systems and questions whose complexity once exceeded our practical analytical capacity. The challenge for evidence-based healthcare is therefore to understand how human and artificial capabilities interact, and how that interaction can expand what we can understand and implement. This framing matters because healthcare’s hardest problems are not linear or isolated technical problems with single solutions; they emerge from exchanges among people, relationships, organisations, behaviours, technologies and policies, and they change as we intervene. AI could help us escape linear thinking, provided our methods remain capable of representing the complexity, uncertainty and lived experience involved in receiving and providing care.

The opportunity, then, is not simply greater computational power, but the ability to examine these interacting layers in ways that ask what is changing, for whom, through which mechanisms, under what conditions and with what consequences. AI may help us move towards a higher-order evidence position in which clinical outcomes, patient experience, organisational factors and relational dynamics can be examined together rather than as separate domains. This would allow us to investigate not only whether an intervention works but also how different parts of a system influence one another as change occurs. In that sense, AI can extend the questions that evidence-based healthcare can ask, while human judgement remains essential to deciding what those answers mean.

What we learned by measuring care

Our work in the Fundamentals of Care and HUMANCUIDA program shows what this shift can make possible, because it lets us examine caring interactions and therapeutic relationships alongside organisational context and care integration. In our patient-centred predictive model of caring interactions, and later in validating the Fundamentals of Care Intelligence Modelling Tool with 1,053 hospitalised patients, we used predictive approaches to examine how contextual and relational components connect rather than treating them as isolated outcomes. 

Our pathway analysis showed that organisational context influences the nurse–patient relationship, which in turn influences the integration of fundamental care; computational approaches can therefore help us see how systems support or constrain the human relationships through which care is delivered. Their value is not that machine learning replaces caring knowledge, but that it makes complex patterns visible so clinicians, managers and researchers can focus their judgement and attention where they have the greatest human and organisational value: person-centred care. Efficiency is therefore not the goal, because using AI merely to make professionals do the same work faster would be a limited ambition. 

If computational systems can support routine information processing, documentation and pattern recognition, clinicians may have more opportunity to concentrate on communication, interpretation, relationships and the person in front of them, while managers can focus more on understanding and improving care systems. However, these benefits are not guaranteed; poorly designed AI can create new burdens, require staff to correct errors or navigate poor interfaces and make patients feel less seen when technology becomes a barrier rather than a support. 

From my perspective, the more important evidence questions are what humans do with the capacity technology releases, how context shapes that response and whether human and technological capabilities together can create better evidence and better healthcare systems.

Lessons from evaluating AI in emergency care

To co-design and evaluate change within complex adaptive health systems, we created the PROLIFERATE framework and, later, PROLIFERATE_AI, in which evaluation focuses on the interactions surrounding an innovation rather than treating technical performance as sufficient evidence of success. 

PROLIFERATE examines comprehension, emotional responses, barriers, motivation and opportunities for optimisation because implementation occurs through people interacting with technologies and systems, while PROLIFERATE_AI extends this reasoning to settings where AI may also help us predict and co-design ways to influence behaviour, expectations and decision-making. In evaluating RAPIDx AI, a tool assessing the likelihood of myocardial infarction, across 12 South Australian emergency departments, we examined not only whether the technology functioned but also how clinicians experienced it, whether they intended to use it, how they understood it and which behavioural responses accompanied its use. 

Three lessons stand out, because they show why evidence about human/AI interaction must be generated alongside evidence about technical performance.

• What worked: Measuring comprehension, enjoyment and preference revealed important behavioural drivers of intended use; these relationships would not have been visible through technical performance measures alone
   
• What did not work as well: Evaluation arrived alongside deployment rather than genuinely before it, after consequential design decisions had already been made and the opportunity for communities to shape the system had narrowed
   
• What we would do differently: Involve patients, carers and communities from the start, helping define the questions, outcomes and boundaries of acceptable use, then continue that partnership as the technology, people and the organisation adapt

Together, these lessons show that technical accuracy matters, but it is insufficient without evidence of how people understand, experience and respond to the system. Human-centred evaluation should therefore begin before deployment, because once humans and AI start responding to one another, the interaction itself becomes an object of evidence that we need to co-design, observe, explain and optimise. The purpose should not be maximum trust, maximum adoption or unrestricted technological capability, but determining which combinations of human judgement and computational capacity produce appropriate, intelligible and beneficial action.

An AI system may excel at pattern recognition while remaining poorly suited to deciding what matters to a particular person, whereas a clinician may understand context, values and relational nuance but cannot process thousands of simultaneous statistical relationships without computational support. Appropriate design therefore requires complementary capabilities, explicit limits and points at which human authority remains decisive, while avoiding unnecessary restrictions that prevent AI from contributing to exploration, modelling, simulation and discovery.

A higher-order evidence position

We cannot always understand healthcare outcomes through a single endpoint measured after an intervention ends, because every solution changes the system it enters; systems change, people adapt, technologies evolve and evidence and methodologies must follow. AI may increase our capacity to detect these changes, connect patterns that would otherwise remain separated and test explanations across larger and more diverse information, yet identifying a pattern does not determine whether it matters, whether acting on it is desirable or whose interests should define success. 

Human reasoning therefore does not disappear as computational capability increases; its role becomes more important in interpretation, values, governance and deciding which problems are worth solving. This is the meta-level opportunity I see for evidence-based healthcare: a higher-order position from which technical, behavioural, relational, contextual and outcome evidence can be examined together to ask whether a combined human/AI system produces a better understanding, designs, decisions and care.

The next frontier for evidence-based healthcare is not better AI alone, but better evidence and learnings on how human/AI methodologies work across time, user types, demographics and technologies, bringing together behavioural, relational, contextual and technical evidence across health systems. Patients, communities, clinicians and leaders must help decide which questions matter and what counts as improvement, while evaluation supports decisions to continue, adapt, limit or stop an AI-supported process as evidence grows. 

The real value of the AI revolution will not be measured by how much human work it replaces, but by how much further it lets us see, understand and act. By making human/AI interaction an object of evidence, we can learn not only what each can do but also what becomes possible when they work together within real systems. The opportunity is to use computation to extend human curiosity and judgement, while keeping relationships, care and values at the centre of healthcare improvement.

Take-home messages

• Look beyond automation: the larger opportunity is to understand how human and AI capabilities can interact to investigate problems that were previously difficult to analyse or even think of
   
• Make human/AI interaction an object of evidence: measure how people, technologies, relationships and contexts influence one another before, during and after implementation
   
• Co-design computational capacity to strengthen human capacity: AI should create more space for relationships, judgement, interpretation and system improvement, rather than simply making existing work faster

References

Allande-Cussó, R., Piñero De Plaza, M. A., Gómez-Salgado, J., & Porcel-Gálvez, A. M. (2025). Measuring the science of caring: A patient-centred predictive model for the caring interaction. Scandinavian Journal of Caring Sciences, 39(3), e70110. https://doi.org/10.1111/scs.70110   

Allande-Cussó, R., Piñero de Plaza, M. A., Kitson, A., Feo, R., Conroy, T., & Porcel-Gálvez, A. M. (2026). How context shapes person-centred fundamental care through nurse-patient relationships: Validation of the FoC Intelligence Modelling Tool and predictive pathway analysis. Journal of Advanced Nursing, 82(7), 7200–7210. https://doi.org/10.1111/jan.70335   

Chew, D., & Lambrakis, K. (2023). Re-engineering the clinical approach to suspected cardiac chest pain assessment in the emergency department by expediting research evidence to practice using artificial intelligence (RAPIDx AI). Health Translation SA. https://healthtranslationsa.org.au/wp-content/uploads/2023/09/RAPIDx-AI-Executive-Summary_FOR-WEBSITE.pdf

Conroy, T., Pinero de Plaza, M. A., Mudd, A., Mitchell, M., & Kitson, A. (2023). Measuring fundamental care using complexity science: A descriptive case study of a methodological innovation. Journal of Clinical Nursing, 32(11–12), 2903–2912. https://doi.org/10.1111/jocn.15905

Piñero de Plaza, M. A. (2025). Advancing digital evaluation for inter-professional education: A PROLIFERATE_AI mixed-methods analysis. Studies in Health Technology and Informatics, 333, 82–86. https://doi.org/10.3233/SHTI251580   

Piñero de Plaza, M. A. (2026). Make innovation work in the real world. https://pinero-de-plaza-research.alejandra-pinero.chatgpt.site/#about

Piñero de Plaza, M. A. (2026). PROLIFERATE_AI reveals behavioural drivers of RAPIDx AI clinical interface use in emergency care. Heart, Lung and Circulation, 35(Suppl 3), S421–S422. https://doi.org/10.1016/j.hlc.2026.07.627   

Piñero de Plaza, M. A. (2026). The 2026 challenge: turning wicked problems into living evidence for care, health innovation, and evaluation in a post-truth political economy. Springer Nature Research Communities. https://go.nature.com/3OzMCEJ 

Piñero de Plaza, M. A. (2026). Why and how health research must escape linear thinking. Springer Nature Research Communities. https://go.nature.com/4eSHOE2   

Pinero de Plaza, M. A., Beleigoli, A., Brown, S., Bulto, L. N., Gebremichael, L. G., Nesbitt, K., Tieu, M., Pearson, V., Noonan, S., McMillan, P., Clark, R. A., Hines, S., Kitson, A., Champion, S., Dafny, H., & Hendriks, J. M. (2022). Effectiveness of telehealth versus standard care on health care utilization, health-related quality of life, and well-being in homebound populations: A systematic review protocol. JBI Evidence Synthesis, 20(11), 2734–2742. https://doi.org/10.11124/JBIES-21-00410

Pinero De Plaza, M. A., Beleigoli, A., Kitson, A., McMillan, P., & Barrera Causil, C. J. (2024). Piloting a big data epidemiology approach to support frail, homebound, and bedridden people. Studies in Health Technology and Informatics, 310, 1292–1296. https://doi.org/10.3233/SHTI231173

Pinero de Plaza, M. A., Gebremichael, L., Brown, S., Wu, C.-J., Clark, R. A., McBride, K., Hines, S., Pearson, O., & Morey, K. (2023). Health system enablers and barriers to continuity of care for First Nations Peoples living with chronic disease. International Journal of Integrated Care, 23(4), 17. https://ijic.org/articles/10.5334/ijic.7643

Pinero de Plaza, M. A., Gulyani, A., Bulto, L. N., Allande-Cussó, R., Pearson, V., Lange, B., Marin, T., Gebremichael, L., Brown, S., Dafny, H., Sajeev, S., Bulamu, N., Beleigoli, A., Nesbitt, K., McMillan, P., Clark, R., Tieu, M., Kitson, A., Champion, S., Hines, S., & Hendriks, J. M. (2025). Telehealth as a care solution for homebound people: Systematic review and meta-analysis of healthcare utilization, quality of life, and well-being outcomes. Health & Social Care in the Community, 7224151. https://doi.org/10.1155/hsc/7224151

Pinero de Plaza, M. A., Hutchinson, C., Beleigoli, A., Tieu, M., Lawless, M., Conroy, T., Feo, R., Clark, R. A., Dafny, H., McMillan, P., Allande-Cussó, R., & Kitson, A. A. (2025). The Caring Life Course Theory: Opening new frontiers in care – A cardiac rehabilitation example. Journal of Advanced Nursing, 81(8), 5163–5180. https://doi.org/10.1111/jan.16312 

Piñero de Plaza, M. A., Lambrakis, K., Marmolejo-Ramos, F., Beleigoli, A., Archibald, M., Yadav, L., McMillan, P., Clark, R., Lawless, M., Morton, E., Hendriks, J., Kitson, A., Visvanathan, R., Chew, D. P., & Causil, C. J. B. (2025). Human-centred AI for emergency cardiac care: Evaluating RAPIDx AI with PROLIFERATE_AI. International Journal of Medical Informatics, 196, 105810. https://doi.org/10.1016/j.ijmedinf.2025.105810   

Piñero de Plaza, M. A., Lambrakis, K., Marmolejo-Ramos, F., Beleigoli, A., Clark, R., McMillan, P., Morton, E., Khan, E., Visvanathan, R., Chew, D., Kitson, A., Hendriks, J., & Barrera Causil, C. (2023). PROLIFERATE_AI: A prediction modelling method to evaluate artificial intelligence in meeting end-user-centric goals around better cardiac care. Heart, Lung and Circulation, 32(Suppl 3), S364–S365. https://doi.org/10.1016/j.hlc.2023.06.833   

Piñero de Plaza, M. A., Taghian, M., Marmolejo-Ramos, F., Barrera-Causil, C. J., & Hall, J. (2021). Investigating salience strategies to counteract obesity. Health Promotion International, 36(6), 1539–1553. https://academic.oup.com/heapro/article/36/6/1539/6143529

Piñero de Plaza, M. A., Yadav, L., & Kitson, A. (2023). Co-designing, measuring, and optimizing innovations and solutions within complex adaptive health systems. Frontiers in Health Services, 3, 1154614. https://doi.org/10.3389/frhs.2023.1154614   

Wiles, L. K., Kay, D., Luker, J. A., Worley, A., Austin, J., Ball, A., Bevan, A., Cousins, M., Dalton, S., Hodges, E., Horvat, L., Kerrins, E., Marker, J., McKinnon, M., McMillan, P., Pinero de Plaza, M. A., Smith, J., Yeung, D., & Hillier, S. L. (2022). Consumer engagement in health care policy, research and services: A systematic review and meta-analysis of methods and effects. PloS one, 17(1), e0261808. https://doi.org/10.1371/journal.pone.0261808

To link to this article - DOI: https://doi.org/10.70253/FSVN1820 

Disclaimer

The views expressed in this World EBHC Day Blog, as well as any errors or omissions, are the sole responsibility of the author and do not represent the views of the World EBHC Day Steering Committee, Official Partners or Sponsors; nor does it imply endorsement by the aforementioned parties.