Ethical Deficits of AI in Administration

3.1 Algorithmic Bias

AI systems learn from historical data. If historical administrative decisions were discriminatory (e.g., Dalits received fewer loans, tribal applicants were denied forest rights more often), the AI will learn to replicate those patterns — at scale and with apparent objectivity. This is structural injustice wearing a mathematical mask.

Examples:

• A US study (ProPublica, 2016) found that COMPAS (a recidivism-prediction tool) was biased against Black defendants — predicting higher re-offending risk at nearly twice the rate for Black individuals compared to white individuals, even after controlling for prior criminal history.
• India's facial recognition systems have shown higher error rates for darker-skinned women, per studies on NIST face recognition datasets.

Administrative implication for Rajasthan: If an AI tool for identifying drought-compensation beneficiaries is trained on old data where tribals in Banswara or Dungarpur were systematically undercounted, the AI will continue the exclusion pattern.

3.2 The Black-Box Problem

Many powerful AI models (deep neural networks) cannot explain their decisions even to their creators. This violates two constitutional principles:

1. Article 14 (equality): If similarly situated individuals receive different decisions with no explanation, it denies equal protection.
2. Natural Justice (audi alteram partem): A person denied a benefit must be told why, with an opportunity to contest.

Explainable AI (XAI) is an emerging field that seeks to make model outputs interpretable. Techniques include LIME (Local Interpretable Model-agnostic Explanations) and SHAP (SHapley Additive exPlanations). India's Digital India Act (proposed) is expected to mandate XAI for high-stakes government AI use.

3.3 The Accountability Gap

Traditional accountability chain: Officer → Department → Minister → Legislature → Electorate.

AI accountability chain: Dataset collector → Model trainer → Platform vendor → Procuring ministry → Approving officer. Each player can claim "the algorithm decided" — diffusing responsibility and creating an accountability vacuum.

Philosopher Luciano Floridi calls this the "problem of many hands" in AI ethics: when many parties contribute to a harmful outcome and each contribution was individually innocuous, traditional moral responsibility frameworks break down.

3.4 The Dehumanisation Risk

Administrative ethics is grounded in the idea that citizens are ends in themselves (Kant's Formula of Humanity), not data points to be processed. Over-reliance on AI can reduce an old woman applying for widow pension to a cluster of data attributes — stripping her of dignity and the right to be heard by a fellow human being with moral sensibility.

---