Patient-first, not gatekeeper-first
Last week in the tearoom in Cairns Hospital, where I’m doing a locum, we debated an “AI boycott.” Too risky, too opaque, too dehumanising. Fair points. Then we finished up, pulled out our phones, and ordered Ubers back to our clinic.
My view is simple: in medicine, tools that expand access and improve quality usually win. Our role isn’t to ban them—it’s to ensure they’re safe, equitable, and effective. If we keep patients at the centre—especially the billions who can’t access timely, quality care—the focus shifts from “boycott” to “building it right.”
We’ve Been Here Before
Every new tool has faced accusations of disrupting medicine. The stethoscope, X-rays, calculators, telehealth, and electronic records—each initially sparked fear, then prompted workflow changes, and eventually became the ‘standard of care.” These innovations shifted us from rote tasks to judgment, triage, communication, and empathy. AI represents another step in that ongoing journey.
What AI Actually Does in a Clinic
Forget the hype. In practice, AI:
- listens and writes (notes, summaries),
- translates and explains (consent, discharge instructions),
- suggests and checks (differentials, risks, guideline nudges),
- coordinates (referrals, inbox triage, prior auths),
- spots deterioration, supports adherence, and helps monitor at home.
Different apps, different risks—but most are “clinician-assist,” not “auto-doctor.”
The Calculator Analogy
We didn’t ban calculators because long division “builds character.” We teach people to verify and think. AI is the clinical calculator: it drafts; you decide. It offers options; you weigh trade-offs. It recalls guidelines; you tailor to the person in front of you. Workload shifts; responsibility doesn’t.
The Patient Case (the only case that matters)
Most of the world cannot access quality healthcare. If AI can open the front door—triage, self-care advice, translation, remote follow-up—that’s a moral win, not a professional threat. Decision support reduces unsafe variability. Language assistance and simple monitoring are crucial in rural and remote areas. Spending less time on documentation allows more focus on what patients truly value: undivided human attention.
The Serious Concerns (name them, fix them)
Privacy and consent. Hallucinations and automation bias. Distribution shift. Fairness. Accountability. Security. Workforce disruption. Environmental cost. All of these are real concerns. They justify implementing guardrails—not a complete freeze.
The Less Noble Concerns (say the quiet part)
Some objections are based on principles; keep those voices close. Others serve as gatekeepers under the guise of ethics—related to status, income, or regulatory barriers. We embrace disruption when it suits us (hello, ride-share) but oppose it when it challenges us directly. If patients remain the guiding light, the double standards become evident.
What Good Looks Like (a practical playbook)
- Consent & transparency: patients choose if and how AI is used.
- Clear scope: indications, contraindications; clinicians accept or reject suggestions.
- Provenance: record model, version, prompt; attach AI artefacts to the chart.
- Validation: external tests, subgroup performance, and real-world monitoring with drift alarms.
- Safety nets: warnings that require human review; incident reporting, like medical-error systems.
- Security & minimal data: lock it down; red-team regularly.
- Efficiency with equivalence: if two models tie on safety/quality, pick the leaner one.
None of this is exotic—it’s routine quality and safety applied to a new tool.
Clinician Upskilling (the new literacy)
Habits, not heroics. Build a routine where you ask the model—and yourself—what it doesn’t know, what the alternatives are, and which diagnoses would disconfirm your favourite idea. Cross-check plans with structured tools and run a quick premortem: if this goes wrong, how and where? Keep bias front of mind by asking who this approach is most likely to fail. Speak plainly with patients about when AI helped and make it clear you still own the decision. And know when to leave the tool on the shelf—novel presentations, unusual populations, and high-stakes ambiguity still demand careful human judgement.
Real World Examples
In general practice, an AI-scribe drafts the note, and a quick guideline prompt identifies missed vaccines, allowing the GP to look up and actually see the person in front of them. In imaging, smart queues prioritise concerning studies, while humans interpret the edges and call out the nuance. In chronic care, quiet remote monitoring and timely nudges detect the gradual decline before it results in an admission. In mental health, stepped-care triage and clear psychoeducation are delivered consistently, with clinicians setting the direction and pace. And in remote communities, asynchronous consultations are connected through succinct summaries and respectful translations that honour local expertise. In every situation, a human makes the final decision.
Measure What Matters
Don’t just trust it—measure it. Track diagnostic accuracy, time-to-treatment, safety events, equity gaps, patient-reported outcomes, and costs. Use practical rollout methods, such as shadow mode, stepped-wedge, and post-market surveillance. Share both successes and failures. Keep iterating.
Policy That Helps, Not Hinders
Pay for AI-assisted care when it genuinely improves access or quality. Make vendors provide clear “labels” for their tools—detailing the data they used, how the model behaves, and proof that the content is authentic. Require results to be demonstrated across different patient groups and address gaps when identified. Develop open medical datasets with proper consent and equitable benefit-sharing, including respect for Indigenous data ownership. Establish clear rules about responsibility when issues arise, and offer clinicians protection when they follow approved guidelines. Safeguard patients without hindering practical innovation.
Boycott or Leadership?
Boycotts widen the divide: wealthy systems forge ahead while others wait. Professional leadership involves setting standards, sharing results, demanding ethical supply chains, and guiding the next generation in using tools effectively. First, do no harm. Second, don’t hoard capacity.
A Small Manifesto for Clinicians
- Patient-first: if it safely expands access and quality, we back it.
- Assistive, not autonomous—until autonomy is proven, audited, and governed.
- Transparent and fair: consent, explainability, and equity are features, not footnotes.
- Learn, measure, improve.
AI will transform medicine. Good. Our task is to ensure that this change is disciplined, ethical, and focused on those who need care most. We can maintain high standards and keep an open mind. Take the Uber to the clinic if it helps—but you set the destination, watch the route, and step out if it’s unsafe. Tools assist; judgment leads.