{"id":17356,"date":"2026-09-21T20:54:00","date_gmt":"2026-09-21T20:54:00","guid":{"rendered":"https:\/\/cjni.net\/journal\/?p=17356"},"modified":"2026-09-30T09:41:02","modified_gmt":"2026-09-30T09:41:02","slug":"designing-what-ai-makes-possible-in-healthcare","status":"publish","type":"post","link":"https:\/\/cjni.net\/journal\/?p=17356","title":{"rendered":"Designing What AI Makes Possible in Healthcare"},"content":{"rendered":"
\n

by Gabriel Chow, RN, MN<\/em><\/p>\n\n\n\n

Clinical Applications Column<\/em>ist
<\/strong>Patient Care Manager,<\/em>
Mount Sinai Hospital, Toronto, Ontario<\/em><\/p>\n\n\n\n

Bio<\/a><\/p>\n\n\n\n

Citation:<\/strong> Chow, G. (2026). Designing What AI Makes Possible in Healthcare. Clinical Applications Column. Canadian Journal of Nursing Informatics, 21<\/em>(3). https:\/\/cjni.net\/journal\/?p=17356<\/p>\n\n\n

\n
\"Designing<\/figure><\/div>\n\n\n

Dedicated to a mentor whose example continues to shape how I think about leadership, care, and what we leave behind.<\/em><\/p>\n\n\n\n

What Should AI Make Possible?<\/strong><\/h2>\n\n\n\n

It is exciting to imagine what artificial intelligence might make possible in healthcare. Safer decisions. Less administrative burden. Faster access to information. More efficient use of scarce resources. Perhaps even more time. As nurse leaders and informaticians, we are rightly focused on whether these technologies are safe, ethical, equitable, and useful in practice.<\/p>\n\n\n\n

In my previous column, I argued that nursing needs a seat at the AI design table. This time, I want to ask what we should be bringing to that table. Because the more interesting question may not only be what AI can do.<\/p>\n\n\n\n

It is what we want healthcare to become because of it.<\/h4>\n\n\n\n

AI promises efficiencies of many kinds: less administrative work, faster processes, greater capacity, and potentially lower costs. In systems facing rising demand and finite resources, those gains matter. But an efficiency gain is an intermediate outcome, not the final one. Shand and Utley (2026<\/a>) argued that organizations should be deliberate about how AI-generated efficiencies become meaningful outcomes.<\/p>\n\n\n\n

Imagine that a task that once took a nurse ten minutes can now be completed safely in five. What once allowed six tasks in an hour could now allow twelve. Sometimes that is exactly the benefit we need: greater capacity, shorter waits, less overtime, or lower cost. But removing work and reducing workload are not the same thing. If every efficiency becomes additional activity, technology may remove tasks while increasing the density of work around them\u2014creating capacity without creating relief.<\/p>\n\n\n\n

But what happened to the five minutes?<\/p>\n\n\n\n

Perhaps it allowed a nurse to notice that a patient had not understood the plan, answer a family\u2019s question, teach instead of rush, or remain present for a conversation that should not be compressed any further.<\/p>\n\n\n\n

The value of an efficiency dividend depends on what we choose to do with it.<\/strong><\/h5>\n\n\n\n

Financial sustainability, access, workforce capacity, and responsible stewardship matter. But productivity should not become our only definition of value.<\/p>\n\n\n\n

Designing the Conditions for Compassion<\/strong><\/h2>\n\n\n\n

I have been thinking about this through work that initially seemed like a different conversation: compassionate care. In developing a compassionate-care initiative with colleagues, we encountered a familiar problem. Asking healthcare professionals simply to \u201cbe more compassionate\u201d is inadequate. Compassion is also shaped by workload, workflow, communication, leadership, and the environment in which care takes place.<\/p>\n\n\n\n

We began by mapping what was already happening: where patient priorities surfaced, concerns were followed through, teams reflected and learned, and compassionate conditions existed but were invisible in our usual metrics. That raised another question: whose experience were our measures capturing? Survey results matter, but whose voices are missing? What might routine data fail to reveal?<\/p>\n\n\n\n

Our approach became multidimensional, combining quantitative and qualitative measures with patient, family, and staff perspectives, process measures, and balancing measures. No single measure can tell us whether compassionate care is occurring reliably or equitably. Together, they can help us understand whether the system is working\u2014and for whom. Models optimized around average efficiency can unintentionally disadvantage people whose care appropriately requires more time: patients who need language interpretation, live with cognitive impairment or disability, have limited health literacy, require caregiver involvement, or face complex social circumstances. Their care may not be inefficient; it may simply resist compression.<\/p>\n\n\n\n

If something matters enough to expect from clinicians, it should matter enough to design for\u2014and measure deliberately.<\/strong><\/h5>\n\n\n\n

Who Defines Value?<\/strong><\/h2>\n\n\n\n

That experience has shaped how I think about AI implementation. Our compassionate-care work is not a technology intervention; it is a bundled approach designed to strengthen the conditions in which compassionate care can occur. But the underlying lesson is relevant to AI: outcomes do not emerge from an intervention in isolation. They are shaped by the people, workflows, expectations, measures, and organizational conditions around it.<\/p>\n\n\n\n

This is why a sociotechnical view of AI matters. McCradden et al. (2023<\/a>) argued that AI should not be understood as an isolated technical tool, but as part of a broader system shaped by the people, workflows, responsibilities, practices, and organizational conditions around it. Strong technical performance alone does not guarantee better care. What ultimately matters are how the technology is implemented, what it changes in practice, and whether the system around it produces the outcomes we intended.<\/p>\n\n\n\n

Defining value is not straightforward. A health system may need greater capacity or lower costs. Clinicians may value less cognitive burden and more usable workflows. Patients may value timely access, understanding, or enough time to ask questions. For some, safe and equitable care simply requires more time. These perspectives are not inherently wrong, but they may not carry equal influence.<\/p>\n\n\n\n

Who gets to define value matters, because those definitions eventually become workflows, measures, and expectations.<\/strong><\/h5>\n\n\n\n

This is where compassion and efficiency meet. They are not opposites. Reducing unnecessary work can create capacity while making it easier for clinicians to communicate, exercise judgment, and remain present.<\/p>\n\n\n\n

If ambient documentation gives a clinician twenty minutes back, is success another appointment, less after-hours charting, more attention during the encounter\u2014or some combination? If discharge instructions can be generated almost instantly, have we improved care because they were produced faster, or because the patient understands what happens next?<\/p>\n\n\n\n

Efficiency tells us something about the resources we use. Saving time tells us what changed in the process. Value requires us to ask what became possible because of it.<\/p>\n\n\n\n

Technical success can tell us whether a tool works. Implementation success should tell us whether those capabilities translate into meaningful outcomes in practice. Rising demand, limited resources, workforce constraints, funding expectations, and performance pressures all influence what appears on dashboards and receives attention.<\/p>\n\n\n\n

What we measure determines what becomes visible, and what becomes visible influences what systems optimize. What we choose to measure also reflects what the system values.<\/p>\n\n\n\n

That is why value belongs at the beginning of implementation, not only at the end of evaluation. Before deploying an AI tool, we should articulate what we hope it will make possible and measure accordingly. Greater capacity, lower cost, safer care, less administrative burden, improved staff experience, and more time with patients are all legitimate outcomes. They should be named deliberately\u2014and examined for whether their benefits are experienced equitably.<\/p>\n\n\n\n

Financial savings and increased productivity may be necessary. But if we cannot articulate what we hope to gain\u2014and what we are unwilling to lose\u2014we may notice the consequences only after they are embedded in the system. Some losses may become so normalized that we struggle to name them.<\/p>\n\n\n\n

Compassion can erode this way. Not because anyone decides it no longer matters, but because other outcomes are easier to see, measure, and optimize.<\/p>\n\n\n\n

What is lost through optimization is difficult to protect if we have never made it visible in the first place.<\/strong><\/h5>\n\n\n\n

What We Choose to Leave Behind<\/strong><\/h2>\n\n\n\n

There is also a more personal reason I return to this question. Recently, I lost a colleague and mentor who had dedicated decades to our hospital and mentored countless people, including me. His leadership was evident in service, responsiveness, generosity, and a willingness to step forward when someone needed help.<\/p>\n\n\n\n

His influence has made me reflect more deeply on stewardship\u2014about what we inherit, what we shape while it is entrusted to us, and what we leave for those who follow. What we pass from one generation to the next is not only knowledge, policy, or measures. Much of it is learned through what people observe: what leaders pay attention to, what organizations reward, which outcomes dashboards prioritize, and what systems communicate about what matters. Trust, generosity, presence, judgment, and care are carried forward this way too. They may be difficult to reduce to a metric, but they are no less real.<\/p>\n\n\n\n

Technology will become part of that inheritance too.<\/p>\n\n\n\n

Nursing therefore needs more than a seat at the AI design table. We need to bring a clear vision of what technology should make possible. Our profession understands that care is both technical and relational, measurable and deeply personal. We see what policies, workflows, technologies, and organizational decisions look like when they meet a person at the bedside.<\/p>\n\n\n\n

The challenge is not to choose between efficiency and compassion. It is to design systems capable of achieving both.<\/p>\n\n\n\n

AI may reduce burden, strengthen safety, improve decisions, expand access, lower costs, and help us use scarce resources more wisely. These gains may be essential to sustaining healthcare.<\/p>\n\n\n\n

But as AI creates new possibilities, the question is not only what we can gain. It is what we want to preserve, strengthen, and make easier to sustain.<\/p>\n\n\n\n

Every generation inherits a healthcare system and leaves one behind. If artificial intelligence becomes one of the defining technologies of ours, our responsibility is not only to make it more capable or efficient. It is to be deliberate about what those capabilities are for, what we hope to gain, and what we are unwilling to lose.<\/p>\n\n\n\n

Because the systems we leave behind reveal what we ultimately chose to value.<\/strong><\/h5>\n\n\n\n

References<\/strong><\/h2>\n\n\n\n

McCradden, M. D., Joshi, S., Anderson, J. A., & London, A. J. (2023). A normative framework for artificial intelligence as a sociotechnical system in healthcare. Patterns, 4<\/em>(11), 100864. https:\/\/doi.org\/10.1016\/j.patter.2023.100864<\/a><\/p>\n\n\n\n

Shand, J., & Utley, M. (2026). Time is not money: Rethinking AI\u2019s efficiency dividend in health care. NEJM Catalyst Innovations in Care Delivery, 7<\/em>(9), CAT.25.0474. https:\/\/doi.org\/10.1056\/CAT.25.0474<\/a><\/p>\n<\/div>","protected":false},"excerpt":{"rendered":"

Gabriel Chow
\nClinical Applications Column<\/p>\n

Volume 21 No 3 2026<\/p>\n","protected":false},"author":1,"featured_media":17389,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[1,2211,2301],"tags":[2308,2304,2310,1098,2307,2305,2265,2306,2309],"_links":{"self":[{"href":"https:\/\/cjni.net\/journal\/index.php?rest_route=\/wp\/v2\/posts\/17356"}],"collection":[{"href":"https:\/\/cjni.net\/journal\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/cjni.net\/journal\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/cjni.net\/journal\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/cjni.net\/journal\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=17356"}],"version-history":[{"count":19,"href":"https:\/\/cjni.net\/journal\/index.php?rest_route=\/wp\/v2\/posts\/17356\/revisions"}],"predecessor-version":[{"id":17449,"href":"https:\/\/cjni.net\/journal\/index.php?rest_route=\/wp\/v2\/posts\/17356\/revisions\/17449"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/cjni.net\/journal\/index.php?rest_route=\/wp\/v2\/media\/17389"}],"wp:attachment":[{"href":"https:\/\/cjni.net\/journal\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=17356"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/cjni.net\/journal\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=17356"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/cjni.net\/journal\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=17356"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}