What if “Ethical AI” isn’t ethical at all… because it quietly leaves whole communities out?

By Dr Alwin Tan, MBBS, FRACS, Executive MBA (Melbourne Business School)

Surgeon | Governance Leader | HealthTech Co-founder | Harvard Medical School — AI in Healthcare Cohort

🔥 Summary: Why this matters

• Some patients are almost invisible in the datasets used to train medical AI.
• When they’re missing, the algorithm cannot protect them — and that becomes silent, systemic harm.
• The danger isn’t just “bias.” It’s invisibility.
• Ethical AI requires representation, not just regulation.

But who forgets to learn from?**

Every week in hospitals, I meet patients with rich, complex stories — stories that rarely make it into the datasets used to build AI tools:

👵🏽 A Vietnamese grandmother who explains pain through metaphors.

👨🏻🦱 A Lebanese father who downplays symptoms to be “strong.”

👩🏽🎓 A young Indian student who has no long-term GP records.

👩🏿🦱 A Somali mother whose trauma lives between the lines, not in the checkbox.

These people exist in waiting rooms, emergency departments, and GP clinics.

But to many algorithms… they don’t.

Not because anyone intended to exclude them.

But because their experiences were never captured in the first place.

Across major research studies, the same pattern keeps appearing:

A highly accurate dementia model trained mostly on wealthier, healthier, White British participants. Great numbers — but limited diversity. So its predictions weaken the moment we apply it to multicultural communities.

A sophisticated tool built mainly on older male veterans. But women, younger patients, migrants, and rural survivors have different risks. The model simply doesn’t know their stories.

A tool predicting 30-day hospitalisation found homelessness, race and nursing-home residence to be major predictors. Important, yes — but also reflections of structural disadvantage. AI can encode the inequity it observes.

Polygenic risk scores are several times more accurate in Europeans than in other ancestries. That means many high-risk multicultural patients may be missed — quietly and consistently.

Across all these examples, one truth stands out:

When someone is missing from the dataset:

• Their symptoms may be misinterpreted,
• their risk underestimated,
• their diagnosis delayed,
• and their outcomes worsened.

This is not a technical glitch. It’s a moral failure.

If we want AI that truly serves everyone, we need more than good intentions. We need deliberate design.

Pain, breathlessness, dizziness — they sound different in different cultures.

Representation must be planned, not assumed.

Otherwise, our models reflect city life only.

If they can’t engage, their data won’t exist — and neither will their protection.

Not “overall accuracy,” but fairness accuracy.

This is healthcare safety, not a research luxury.

The question is no longer: “Is our AI accurate?”

The better question is: “Accurate for whom?”

If AI is going to shape the future of healthcare, then the first ethical act is to make sure everyone is seen.

Ethical AI is not just bias-free AI. It is inclusive AI — built on the dignity and diversity of real human lives.

Dr Alwin Tan, MBBS, FRACS, EMBA (MBS) is a senior surgeon, governance leader, co-founder of health-tech innovation and Faculty Advisor to AIHE. A dual Bastas Academy Health Leadership Scholar and AUSCEP 2025–26 Fellow, he focuses on building safe, ethical and inclusive digital health systems. He is also Mamuk, the adopted son of Murrundindi, Head of the Wurundjeri people — a connection that grounds his commitment to cultural respect and drives his quiet determination to stand against prejudice in all its forms.

• Ren et al. Dementia risk prediction (UK Biobank)
• Vitzthum et al. Opioid use, abuse, and toxicity in cancer survivors
• Aboumrad et al. COVID-19 30-day hospitalisation score
• Aragam et al. Polygenic risk and coronary artery disease
• Martin AR et al. Polygenic risk scores and cross-ancestry disparities

DrAlwinTan #EthicalAI #AIinHealthcare #HealthEquity #InclusiveDatasets #ClinicalGovernance #DigitalHealth #AIHE #HarvardMedicalSchool #MelbourneBusinessSchool #AUSCEP #BastasAcademyForHealthLeadership #IndigenousHealth #CALDCommunities #HumanCentredAI #FutureOfHealthcare #HarvardBusinessReview #MedTechInnovation #PatientSafety #DataBias #PrecisionMedicine #HumanRightsCommisioner#VMC #AHRC #FWC #FWO #MJA

Originally published at https://www.linkedin.com.