Discuss the ethical concerns of using large multi-modal models in clinical decision-making. How do the ICMR guidelines and the WHO guidance address them?

Large multi-modal models (LMMs) are a type of generative AI. They take in text, images and other data, and they can produce outputs of a different type from the input [1]. They can help with diagnosis and clinical notes. But they raise ethical risks, and India's ICMR norms and WHO's guidance deal with these risks only in part.

Ethical concerns in clinical decision-making

• Accuracy and safety: LMMs can give "false, inaccurate, biased, or incomplete" answers that still sound confident [1].
• Automation bias: clinicians and patients may miss AI errors or hand over decisions the AI should not make [1]. This weakens the doctor's own judgement.
• Algorithmic bias: models trained on data skewed by race, sex or age work less well for groups that are under-represented in that data [1]. This risk is high in a country as diverse as India.
• Privacy and cybersecurity: pooled patient data can be breached, and models can be tampered with [1].
• Equity: high-performing models may not be accessible or affordable for everyone, which can widen the digital divide [1].
• Accountability gap: it is unclear whether the doctor, the hospital or the developer answers for a wrong diagnosis led by AI.

How the ICMR Guidelines (2023) respond

• They set 10 patient-centric ethical principles for developers, researchers and clinicians [2].
• They lay down an ethics review process and informed-consent norms [2].
• AI deployed under the IndiaAI Mission must follow the ICMR and MeitY AI Governance Guidelines, with privacy-by-design [3]. An IndiaAI–ICMR MoU supports responsible adoption [4].
• Limitation: the guidelines are advisory. They name no regulator to approve tools and require no checks after launch.

How the WHO guidance (2024) responds

• It makes 40+ recommendations and says governments have primary responsibility for setting standards [1].
• A regulatory agency should assess and approve LMMs used in health care [1].
• Independent audits and impact assessments after release should be mandatory, with the results made public [1].
• Governments should fund public or not-for-profit infrastructure that developers can use only if they follow ethical principles [1].
• Developers should involve patients and health workers early and design each tool for a well-defined task with proven accuracy [1].

The ICMR norms set ethical rules for designing and testing these tools, while WHO adds regulatory approval and checks after launch. India should back the ICMR norms with law and name a single regulator for health AI. It could also use part of the ₹10,371.92 crore IndiaAI outlay [3] for public infrastructure that is independently audited. That would keep LMMs as aids to doctors rather than replacements, and would advance the right to health under Article 21 and SDG 3.