Blog

Google DeepMind's recent "AI co‑clinician" research represents a shift from prediction engines to reasoning partners. This post unpacks the planning‑and‑execution architecture, the NOHARM evaluation framework, and why evidence synthesis and arguable systems matter for real‑world clinical deployment in diverse settings like Bangladesh.

Drawing on recent research and a real case from rural Bangladesh, this post argues that prompt engineering alone cannot guarantee LLM safety in clinical settings. Hallucinations, bias, and shallow reasoning remain even with advanced prompting.

Aggressive differential privacy does not just protect data. It systematically hides rare diseases, marginalized communities, and emerging outbreak signals in low-resource settings like Bangladesh.

Moving beyond explainability as a technical patch, this post introduces "arguable systems" as a new design principle for clinical AI, with a five-step pipeline for building systems that invite clinical scrutiny rather than demanding blind trust.

Drawing on Miranda Fricker's framework of epistemic injustice, this post examines how clinical AI systems commit testimonial and hermeneutical wrongs at scale, and proposes five directions for building epistemically just AI.

This piece argues for designing systems that cultivate epistemic virtues — intellectual humility, courage, and responsibility — rather than treating privacy as a simple trade-off.

Examining the philosophical clash between clinicians' experiential wisdom and AI's opaque reasoning, this post proposes designing inherently arguable systems using hybrid rule-augmented architectures.

This post argues that real trust in health data systems requires moving beyond technical compliance to embrace epistemic responsibility — auditing not just data but knowledge itself.

Exploring how the very noise that protects patient privacy can erode clinicians' ability to form justified beliefs about medical reality, and how to bridge the gap.

Examining the fundamental tension between Differential Privacy's mathematical guarantees and the need for transparent explanations in clinical AI, proposing a co-design framework.

Exploring why federated learning's privacy guarantees are insufficient without systematic ethical audits, with a practical framework combining rule-augmented networks and continuous fairness monitoring.

How traditional approaches can erase rare disease signals and marginalized communities, and adaptive methods that protect privacy while preserving clinical truth and equity.

A personal reflection on Sutskever's insights about AI's limitations and the transition from scaling to true understanding, connecting his observations to real challenges in global health AI.

How hybrid symbolic-neural models solve AI's black box problem by combining neural network pattern recognition with symbolic AI's logical reasoning, and the ethical implications for healthcare.

A journey from pure coding to philosophically-informed AI development, and how building a dengue symptom triage chatbot revealed the ethical dimensions of technical decisions.

Examining the critical gap between AI ethical training and real-world vulnerabilities, proposing a framework for healthcare AI that is both powerful and trustworthy.

Exploring how cultural contexts and economic realities create divergent AI adoption paths in healthcare, from Scandinavia's privacy-focused systems to Bangladesh's access-driven triage AI.

A practical deep dive into SHAP and LIME based on real research in intersectional fairness, with code examples, critical analysis of limitations, and how these tools revealed fairness violations in image classifiers.

Two critical case studies of algorithmic bias in healthcare AI and an interdisciplinary framework combining technical fairness methods with ethical principles to build more equitable clinical systems.

Learning Norwegian one Duolingo streak at a time, alongside my research career. Reached Diamond League with 30,000+ XP and a 175-day streak.

An in-depth analysis of Western healthcare economies through recent research on Norwegian systems, examining circular economy implementation, strategic hospital design, and the ethical integration of AI.

AI's potential to save lives is revolutionary — but in this high-stakes domain, potential is meaningless without a foundation of trust. An exploration of what genuine trust in clinical AI actually requires.

Two patients with the same symptoms — one flagged for a life-saving screening, one not. The difference is their postal code. An examination of algorithmic bias as the central ethical crisis of AI in medicine.

Exploring the subtle epistemic and social costs of privacy-preserving techniques in healthcare AI and how they change the way medical knowledge is produced and shared in clinical settings.

High accuracy numbers do not always translate to real-world clinical trust. A reflection on the gap between model metrics and clinician understanding in healthcare AI systems.

Moving beyond technical explanations to philosophical frameworks for building trustworthy AI systems, exploring epistemic opacity and designing for different stakeholder needs.

An overview of my latest research contributions covering dengue symptom triage chatbots, health indicator analysis, air quality prediction, and fairness in ECG models.

A deep dive into Dawn Brown's exploration of hidden wisdom and its connections to cognitive science and AI research.