"AI can democratise diagnostics in India but cannot replace the clinician." Critically examine, with reference to the regulatory framework for AI in healthcare.

India has about 4,000 cardiologists for more than 1.3 billion people [3]. AI-enabled screening can help close gaps like this. But AI recognises patterns; it does not exercise clinical judgement. The statement therefore largely holds, as long as regulation keeps an accountable clinician in the loop.

How AI democratises diagnostics

• Task-shifting: MadhuNetrAI lets non-specialist health workers screen for diabetic retinopathy. It covers 38 facilities in 11 States and about 7,100 patients [2].
• Last-mile reach: AI stethoscopes turn heart and lung sounds into visual data, which cardiologists can review remotely [3].
• Early detection: ICMR-backed AI thermal imaging is being used for early detection of breast cancer [4].

Why the clinician remains indispensable

• Bias: a model may not represent India's diverse population. WHO wants training data "intentionally made representative" [6]. A pilot of about 7,100 patients does not prove the tool works nationally.
• Automation bias: AI can lead professionals and patients to over-trust its output and miss errors [5].
• Equity: the best-performing models may not be affordable or accessible [5].
• Accountability: an algorithm cannot take responsibility for a diagnosis or counsel a patient.

Regulatory framework: strengths and gaps

• Strengths: government AI deployments follow the ICMR Ethical Guidelines (2023) and MeitY's AI Governance Guidelines [1]. The DPDP Act, 2023 governs health data [2]. The Medical Devices Rules, 2017 use risk-based device classes and follow the GHTF framework [7].
• Gap 1, soft law: the ICMR and MeitY documents are guidelines only. They carry no penalties and no power to recall a tool.
• Gap 2, static rules: device rules assume a fixed product. WHO flags "continuous learning" models, which keep changing after approval [6].
• Gap 3, fragmentation: States license Class A/B devices and the Centre licenses Class C/D [7]. Health is a State subject (Entry 6, List II).

Way forward

• External validation on multi-site Indian data before scale-up, modelled on ICMR's thermal-imaging trials across diverse populations [4][6].
• Human oversight matched to risk: light oversight for referral-only screening, a qualified clinician's sign-off for treatment decisions [6].
• Post-market surveillance, plus consent explained in the patient's own language.

AI widens the base of India's diagnostic pyramid, while the clinician remains at its apex. A binding, risk-based framework that covers the whole product life-cycle would let AI expand access without weakening accountability. That would advance the right to health under Article 21 and universal health coverage under SDG 3.