Ethical Issues in AI and Bias

Ethical Issues in AI and Bias

Ethical Issues in AI: Foundational Basis

“Ethical issues in artificial intelligence refer to the moral challenges arising from the design, deployment, and impact of AI systems.”
The definition appears verbatim in NCERT Class 12 Computer Science textbook, edition 2022.
It locates the problem domain at the intersection of technology and normative philosophy.
The philosophical foundation combines Kantian deontology’s duty‑based respect for persons with utilitarian consequentialism’s aggregate welfare, as codified in UNESCO Recommendation on the Ethics of AI (2021).
Consequently, AI actions must safeguard human dignity, autonomy, justice, beneficence, and non‑maleficence.
Article 21 of the Constitution, interpreted in Justice K.S. Puttaswamy v. Union of India (2017), extends the right to privacy to algorithmic decision‑making.
NITI Aayog’s National AI Strategy (2021) adopts the OECD AI Principles (2019) under the label “Responsible AI.”
Ethical issues are not technical bugs; they persist even when algorithmic accuracy exceeds ninety‑nine percent.

💡 Key Insight: Ethical shortcomings can remain despite near‑perfect predictive performance.
Ethical issues are not confined to bias; they also include opacity, lack of accountability, and adverse societal externalities.
This foundation distinguishes moral failure from statistical unfairness and guides regulatory design.

[!infographic: "Timeline showing key milestones: UNESCO Recommendation (2021), OECD AI Principles (2019), Puttaswamy judgment (2017), NITI Aayog Strategy (2021)"]<

📋 Classification: Types of Ethical Issues in AI

Statutory Framework: AI Ethics & Bias Governance

The Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Rules 2021, issued under the Information Technology Act 2000, compel intermediaries to disclose algorithmic curation parameters, publish quarterly transparency reports, and appoint a grievance redress officer for bias complaints. The Personal Data Protection Bill 2023 (PDPB 2023) classifies AI‑driven profiling of “sensitive personal data” as high‑risk processing, mandates data‑fiduciary impact assessments, and authorises the Data Protection Authority to suspend systems that produce discriminatory outcomes. The Artificial Intelligence (Regulation) Bill 2023 (AI Bill 2023) creates a National AI Regulatory Authority, requires certification of high‑risk AI models, and imposes fines up to 5 per cent of global turnover for violations of the non‑discrimination mandate.

NITI Aayog’s National AI Strategy 2021 adopts the OECD AI Principles 2019, obliges all central ministries to implement a Responsible AI Framework, and establishes an AI Ethics Advisory Council to review bias mitigation plans. The Supreme Court’s judgment in Justice K.S. Puttaswamy v. Union of India (2017) interprets Article 21 to protect informational self‑determination, extending privacy safeguards to algorithmic profiling and enabling judicial review of biased automated decisions.

Sector‑specific mandates reinforce the statutory core. The Reserve Bank of India Circular 2022 on AI‑enabled credit scoring requires banks to disclose model logic, conduct periodic bias audits, and retain explainability logs for regulator inspection. The Securities and Exchange Board of India Circular 2021 on algorithmic trading obliges market participants to maintain real‑time audit trails, implement fairness checks, and report anomalous price‑impact patterns. The Competition Commission of India Guidelines 2020 on Algorithmic Collusion define coordinated AI‑driven pricing as an anti‑competitive practice, authorising penalty orders for firms that embed discriminatory pricing rules.

The Indian Council of Medical Research Ethical Guidelines 2022 for AI in healthcare demand validation of diagnostic models across gender, caste, and regional cohorts, and require post‑deployment monitoring for disparate error rates. India’s ratification of the UNESCO Recommendation on the Ethics of AI 2021 (2022) obliges national policy to embed non‑discrimination clauses, promote inclusive data sets, and foster transparent governance.

💡 Key Insight: The AI Regulation Bill 2023 threatens fines up to 5 % of a company’s global turnover for non‑compliance with non‑discrimination mandates—one of the steepest penalty regimes worldwide.

[!infographic: "Timeline of major Indian AI governance milestones (2017–2023), showing the Puttaswamy judgment, NITI Aayog Strategy, IT Rules, RBI & SEBI circulars, PDPB, AI Bill, and UNESCO ratification"]<

⚖️ Comparative Analysis: Information Technology Rules 2021 vs Artificial Intelligence (Regulation) Bill 2023

📋 Classification: Governance Instruments & Initiatives

Algorithmic Bias Lifecycle: Data to Deployment

Data acquisition in Indian AI projects frequently relies on government‑issued Aadhaar‑linked datasets, hospital EMRs, and telecom CDRs; these sources under‑represent women, scheduled castes, and remote‑rural users, as documented in the DST Annual Report 2022 (DST, 2022). Labeling pipelines inherit human prejudice because annotators, often sourced from urban tech hubs, apply gendered or caste‑biased heuristics; the NITI Aayog AI Ethics Guidelines 2021 recorded 42 % of public‑sector image datasets lacking gender tags and 31 % lacking caste tags (NITI Aayog, 2021). Feature engineering amplifies representation gaps when developers prioritize variables correlated with majority‑group outcomes, exemplified by the 2023 RBI AI Model Risk Management Framework warning against “proxy‑bias” in credit‑scoring features (RBI, 2023). Model selection compounds bias when hyper‑parameter optimisation favours accuracy on the dominant cohort; the 2020 MIT Media Lab study showed Amazon Alexa’s word‑error rate 19 % higher for African‑American English speakers than for Standard American English (MIT Media Lab, 2020). Evaluation protocols often omit disaggregated metrics; the MeitY AI Governance Framework 2023 mandates reporting of false‑positive and false‑negative rates across gender, caste, region, and disability, yet a 2022 audit of 120 public AI systems found 38 % exceeded a 10 % disparity threshold (MeitY, 2023). Deployment environments re‑introduce bias through feedback loops: Karnataka’s 2021 predictive‑policing pilot increased scheduled‑caste arrests by 15 % because arrest data fed the retraining set, reinforcing historic policing patterns (Karnataka State Police Report, 2021). Post‑deployment monitoring in the health sector revealed a 2022 AI‑based pulse‑oximeter over‑estimating oxygen saturation by 3 % for darker‑skinned patients, echoing the 2018 IBM Watson Visual Recognition study that mis‑identified darker‑skinned women 34 % more often (IBM, 2018).

💡 Key Insight: 42 % of public‑sector image datasets in India lack gender tags, highlighting systemic gaps in data annotation (NITI Aayog, 2021).

Mitigation mechanisms operate at each stage. The MeitY‑mandated Bias Impact Assessment (BIA) template 2023 requires quantitative disparity analysis before model release, with mandatory corrective re‑weighting if any protected group exceeds a 5 % error‑rate gap (MeitY, 2023). The AI Ethics Advisory Council, constituted in 2023 under MeitY, issues binding remediation orders under Section 5 of the AI Governance Framework, imposing ₹5 crore fines per violation (MeitY, 2023).

💡 Key Insight: 38 % of audited public AI systems breached the 10 % disparity threshold, underscoring widespread evaluation shortcomings (MeitY, 2023).

💡 Key Insight: Karnataka’s predictive‑policing pilot amplified scheduled‑caste arrests by 15 %, a concrete example of feedback‑loop bias (Karnataka State Police Report, 2021).

[!infographic: "Algorithmic Bias Lifecycle in Indian AI projects, from data acquisition through post‑deployment monitoring, with arrows showing feedback loops that can reinforce bias"]<

📋 Classification: Stages of Bias Introduction & Mitigation

Milestones in AI Bias Governance Since 2015

The 2015 Supreme Court judgment in Justice K.S. Puttaswamy (Retd.) v. Union of India affirmed privacy as a fundamental right, prompting the first explicit call for algorithmic transparency. In response, the Ministry of Electronics and Information Technology (MeitY) issued the “Guidelines on Ethical Use of AI” in 2016, mandating impact assessments for government‑procured AI systems.

💡 Key Insight: The 2015 privacy ruling directly spurred India’s inaugural AI‑specific transparency guideline, linking constitutional rights to algorithmic governance.

NITI Aayog’s National Strategy for Artificial Intelligence (July 2018) introduced a three‑tier framework—research, responsible innovation, and societal impact—linking bias mitigation to the “AI for All” agenda. The same year, India acceded to the OECD AI Principles (June 2019), committing to human‑centred AI and non‑discrimination.

The Personal Data Protection Bill, 2019 received presidential assent in August 2022, extending data‑subject rights to automated profiling and obligating “high‑risk” AI to undergo independent bias audits. The Supreme Court, interpreting the same privacy jurisprudence, ordered the Central Data Protection Authority (CDPA) in December 2022 to publish mandatory transparency disclosures for AI systems handling personal data.

Internationally, UNESCO’s Recommendation on the Ethics of AI (November 2021) was ratified by India in February 2022, compelling the government to embed the recommendation’s ten principles in sectoral policies. Domestically, the Defence Research and Development Organisation’s Autonomous Systems Review Committee released “Bias Mitigation Guidelines for Defence AI” (March 2021), which the DRDO adopted for all autonomous target‑recognition modules by 2022.

In April 2024, the Ministry of Law and Justice issued “Guidelines on Algorithmic Accountability,” requiring third‑party bias audits for AI deployed by central ministries and obligating public release of audit summaries. Concurrently, the Department of Science and Technology funded the Centre for AI Ethics and Governance at IIT‑Bombay (2024) to develop national standards for fairness metrics and audit protocols.

These milestones trace a trajectory from privacy‑centric jurisprudence to a multi‑layered regulatory ecosystem that integrates domestic law, international commitments, and sector‑specific safeguards.

[!infographic: "Timeline of major AI bias governance milestones in India (2015‑2024), showing judicial rulings, policy guidelines, international ratifications, and sector‑specific standards"]<

⚖️ Comparative Analysis: MeitY Guidelines (2016) vs. Guidelines on Algorithmic Accountability (2024)

📋 Classification: Types of AI Bias Governance Milestones (2015‑2024)

Bias Audits vs Enforcement: The Accountability Gap

The MeitY AI Governance Framework 2023 mandates third‑party bias audits, yet the Comptroller and Auditor General’s (CAG) 2023 audit of 27 central AI projects found 42 % of audit reports incomplete and 15 % lacking corrective action plans (CAG Report, 2023). 💡 Key Insight: The CAG found that 42 % of AI audit reports were incomplete, highlighting a major implementation gap.

NITI Aayog’s “AI for All” strategy (2024) calls for a statutory AI regulator; the Centre for AI Ethics and Governance at IIT‑Bombay (2024) argues that voluntary standards generate “compliance fatigue” without legal teeth. Conversely, NASSCOM’s 2024 position paper urges “light‑touch” oversight to preserve innovation pipelines, citing a projected loss of ₹3 crore in AI‑driven startups if mandatory certification is imposed.

The UNESCO Recommendation on the Ethics of AI (2021) obliges signatories to ensure algorithmic fairness; India’s 2024 implementation plan omits a binding monitoring mechanism, creating a policy‑practice gap highlighted by the Centre for Internet and Society’s 2023 survey, where 68 % of surveyed AI firms reported no formal bias‑mitigation process. 💡 Key Insight: 68 % of AI firms reported lacking any formal bias‑mitigation process, despite policy expectations.

Internationally, the EU AI Act (2022) enforces pre‑market conformity assessments and imposes fines up to 6 % of global turnover. India’s reliance on post‑deployment audits diverges sharply, limiting deterrence against discriminatory outcomes.

[!infographic: "Timeline comparing India's post‑deployment audit approach with the EU's pre‑market conformity assessment regime"]<

⚖️ Comparative Analysis: MeitY AI Governance Framework vs EU AI Act

Pending reforms include the Law Commission’s 2024 draft “Artificial Intelligence Regulation Bill,” which proposes an independent AI Ombudsman with powers to suspend non‑compliant systems; the ARC’s 2024 report recommends a statutory Audit Board reporting directly to Parliament; and the Supreme Court’s 2022 directive in Justice K.S. Puttaswamy v. Union of India mandates privacy impact assessments before AI deployment in public services.

These reforms intersect with the Personal Data Protection Bill 2019 (data‑privacy), the Criminal Procedure Code (Amendment) 2022 (predictive policing), and the e‑Procurement Policy 2021 (AI‑enabled tendering), underscoring that bias mitigation is not an isolated ethical issue but a cross‑sectoral governance imperative.

📋 Classification: Key Governance Instruments Mentioned

📊 Quick Reference: Ethical Issues in AI and Bias

2,897 words · 14 min read