Opinion Before AI Summit, an ethics checklist

AI systems must also be grounded in India’s lived realities . This includes caste dynamics, gendered labour, linguistic diversity, rural urban divides and digital precarity. These realities demand intersectional fairness and representational justice from technological systems . One way to advance this is through audits that test AI systems for overlapping harms — harms faced by Dalit women, migrant workers, Adivasi youth, persons with disabilities and linguistic minorities. An intersectional audit does not examine gender or caste in isolation but looks at how gender, caste and class interact in practice. This is essential because Indian datasets often invisibilise intersectional groups. Ethical principles must emerge from the ground rather than being parachuted in from Western contexts with very different power structures.

Transparency needs a clear operational meaning. AI systems should be accompanied by publicly accessible model cards that function like nutrition labels for algorithms. These should document training data sources, strengths and limitations, known biases, appropriate and inappropriate uses, and contact points for grievance. This also helps counter the inflated claims and hype that often accompany AI deployments in public systems.

Any ethical framework must guarantee consent, community control over data, fair value sharing and protection against extractive practices. One promising mechanism is the idea of community data trusts — legally recognised bodies that hold and manage data on behalf of communities, much like forest or natural resource trusts. Benefits generated from community data must flow back to those communities so that India does not become a data colony where lived experiences are harvested without consent or benefit.

Ethics without remedial mechanisms are largely decorative. Clear liability rules are needed to establish responsibility when AI systems cause harm. If an elderly or disabled person is denied food rations due to facial recognition failure, the government department deploying the system should bear primary liability. Vendors may be secondarily liable if the system was flawed, biased or misrepresented. This prevents responsibility from being shifted on to opaque systems .

Independent grievance redress systems are essential to give teeth to ethical commitments. For high-risk applications in policing, welfare or medicine, mandated human oversight is necessary to override algorithmic decisions when required.

Finally, people must be able to understand why an AI system made a decision about them nd the steps available to contest it. Although conversations on ethics often find it hard to reach a consensus on account, anchoring AI ethics in these human rights principles could provide a common ground for much of humanity. By taking an active lead in implementing such principles, India could be the vishwaguru it deserves to be.

The writer, a former fellow of Harvard Kennedy School, teaches at Jindal School of Government and Public Policy