GIRAI 2024: Mapping India’s Actions on Responsible AI

Introduction

In the past two decades, Artificial Intelligence (AI) has become a ubiquitous part of our society. Its growing use has led to societal and economic shifts towards increased automation and data-driven decision-making across sectors, including healthcare, education, law enforcement, banking, and energy. Its rampant development and use have prompted questions about its ethical implications, risks, and long-term impacts, leading to emerging debates about governing these technologies through policies and regulations. Most of these discussions revolve around ‘Responsible AI’ (RAI), an umbrella term for principles and practices that promote safe and ethical development and use of AI.

Efforts to inculcate RAI principles are underway worldwide. To map different countries’ progress in developing and implementing RAI frameworks, the Global Center on AI Governance led the creation of a multidimensional index called the Global Index on Responsible AI (hereon, “Index”). It measures government commitments and country capacities, through a social, technical and political lens, through localised data collection in 138 countries. Supported by the International Development Research Centre (IDRC), the Government of Canada and USAID, its 1st edition was released in June 2024.

Digital Futures Lab served as the Regional Hub for South Asia, coordinating efforts with data collectors across eight countries. In addition, we also acted as the primary data collector for India. In this post, we highlight the key findings from India, specifically focusing on its government frameworks for Responsible AI.

Overview of the Index

As seen in the table, the Index measures RAI efforts across 19 thematic areas, clustered into 3 dimensions: (i) Human Rights and AI, (ii) Responsible AI Governance, and (iii) Responsible AI Capacities.

Each thematic area is assessed across 3 different pillars of the responsible AI ecosystem: (a) Government Frameworks, (b) Government Actions, and (c) Non-state Actor Initiatives. The different components of the Index have been depicted in the table below - highlighting key thematic areas covered by government frameworks in India.

The data used in the 1st edition of the Index covers the period between November 2021 and November 2023.

Table: Responsible AI Themes within Government Frameworks in India

Thematic areas highlighted in red represent issues not represented within existing government frameworks

Key Findings

Coverage gaps in government frameworks for Responsible AI in India

Findings under the pillar of government frameworks reveal a lack of legal and policy frameworks pertaining to RAI. Although there is some activity in specific thematic areas, the frameworks and principles are primarily outlined in white papers and strategy documents, and as such, are not legally enforceable. Further, most of the activity is very sector-specific such as the Indian Council for Medical Research's ethical guidelines for the application of artificial intelligence in biomedical research and healthcare 2023, the Ministry of Housing and Urban Affairs' AI playbook for cities, and the National Education Policy, 2020.

Human Rights & AI: This dimension measures the extent to which countries are taking steps to protect, promote, and respect key human rights implicated by AI. Findings under this dimension reveal that governments at the central and state levels are yet to take any action in the form of frameworks such as laws, policies, regulations, and guiding documents on issues such as Children’s Rights, Gender Equality, Public Participation and Awareness.

👉🏼 Note: There are some recent developments under labour protection which could not be included in this edition of the Index as they fell outside the stipulated timeline for data collection. These include legislations aimed at regulating gig work in Rajasthan, Karnataka, and Jharkhand.

Responsible AI Governance: This dimension measures the degree to which country-level governance regimes uphold effective and rights-preserving practices in RAI. Findings under this dimension reveal that governments are yet to develop frameworks that cover algorithmic decision-making and provide access to remedy and redress to those affected by algorithmic decision-making.

The lack of specific regulation for mitigating algorithmic risks has resulted in provisions from other laws, such as criminal laws in the country, being used to regulate AI harms. For example, the individual responsible for creating deepfake videos of an Indian actor was arrested under sections 465 and 469 of the Indian Penal Code (IPC) for forgery and maligning the reputation of a person. Although Section 66C (identity theft) and Section 66E (privacy violation) of the Information Technology Act, 2000 were applied, there is no AI-specific legal framework available to indict the individual responsible.

Responsible AI Capacities: This dimension evaluates the extent to which state capacities essential for advancing responsible AI have been developed. Our findings reveal that little action has been taken towards developing frameworks and blueprints to build institutional capacities for AI oversight and regulating competition.

However, other areas such as ‘Competencies’ and ‘Investments’ have been covered in the National Strategy for Artificial Intelligence, 2018. While these sub-dimensions are important, institution-building is imperative to ensure the enforcement of RAI principles and to build long-term capacity within the ecosystem to predict, mitigate, and manage AI risks.

Existing government frameworks lack enforcing power

For a majority of the thematic areas where action has been taken in India, the National Strategy for Artificial Intelligence, 2018 (NSAI) is the only listed activity. The NSAI was published by the NITI Aayog, the apex policy think tank of the Government of India and was followed by three approach papers. However, the NSAI is a strategy document which does not hold the force of law and the recommendations made in these documents are not enforceable. Additionally, the NSAI fails to cover many topics and actions pertinent to RAI in India. For example, it only discusses environmental sustainability as a use case of AI, failing to highlight the many environmental risks that energy- and resource-intensive AI infrastructures pose. Regarding AI transparency and explainability, NSAI only discusses their importance, without offering clear pathways for operationalising them within the Indian context.

NITI Aayog’s second paper, ‘Operationalising Principles for Responsible AI’, however, recommends that the government should make regulatory and policy interventions to create a trusted and responsible AI ecosystem. However, even as we reach the end of 2024, no such framework has been put in place. This highlights a noticeable disparity between policy and practice in the government's approach towards RAI.

Conclusion

The insights gathered from the Index allow us to assess the gaps in India’s efforts towards embedding RAI principles within its AI ecosystem. Efforts by sectoral actors may be seen as reactionary and failing to account for systemic and long-term needs. While attempts have been made to create a national strategy framework to govern AI, the findings indicate a lack of action to give these frameworks the force of law. Presently, a constellation of legislations respond to technology-related issues in India. This includes the Information Technology Act, 2000, the Digital Personal Data Protection Act, 2023, and the Telecommunications Act, 2023, to name a few. Since 2022, the ‘Digital India Bill’ has been extensively discussed, especially as a potential replacement for the Information Technology Act. According to statements made by the government and media reports, its scope has been expanded to include AI-related issues. However, recent reports suggest that the Bill has been delayed and the government is now looking to put in place smaller, issue-specific legislation for AI. Overall, the government has been reticent to overprescribe when it comes to AI regulation, fearing that such measures might stifle innovation in the country.

It is also crucial to consider whether India requires dedicated AI legislation, and, if so, what the framework and scope of such a law should encompass. Are there aspects of AI risks that are not covered by existing provisions? If yes, what are these risks and how do we identify ones relevant in the Indian context? Further, akin to the right to privacy being recognised as a fundamental right flowing from the right to life in the Constitution of India, could provisions of the AI legislation be constructed to enforce constitutional rights/values such as freedom of speech and prohibition of discrimination based on religion, caste, gender etc? Other than constitutional values, are there any other values, such as cultural values, that may be useful when identifying the contours of this legislation? These are some of the questions that may be worthwhile in this discussion.

It is also worth noting that government frameworks, while necessary, are only one piece of the puzzle. Efforts must be made to involve multiple stakeholders - across government, industry, and civil society, to build capacity, frameworks, and action pathways towards inculcating RAI principles into the AI development and deployment cycle in India.