Platform approach to AI policy

India's AI Safety Norms:? a platform approach is needed

By K Yatish Rajawat and Dev Chandrasekhar

Artificial Intelligence (AI) is transforming worldwide societies, economies, and governance structures. However, its rapid evolution has raised concerns about critical safety, ethics, and accountability. Governments worldwide have begun implementing regulatory frameworks to ensure AI development aligns with societal interests. India, too, has initiated efforts to establish AI safety norms, but a fundamental question remains: Whose safety are these regulations prioritizing? Are they aimed at protecting citizens, the government, data privacy, or corporate interests? A comparative analysis with global approaches provides insight into India’s stance and implications.

On 30 January, India’s Union Minister for Electronics & Information Technology, Railways, Information & Broadcasting, Ashwini Vaishnaw, announced the establishment of an AI Safety Institute; according to the release of the government’s PIB, the aim is to address artificial intelligence's complex challenges through a strategic, multi-institutional, techno-legal approach. Eight projects will work parallel to ensure data privacy, reduce biases, and make AI systems transparent and accountable.

The 8 Projects Addressing AI Governance

The first of these projects, "Machine Unlearning," is being spearheaded by IIT Jodhpur. Since data privacy is paramount, this initiative aims to allow AI systems to "forget" specific data, providing a selective eraser for sensitive or outdated information.

IIT Roorkee's "Synthetic Data Generation" project creates artificial data that mimics real-world information. This allows AI systems to be trained without compromising individual privacy. It's a clever workaround that could set a new standard in data protection.

The "AI Bias Mitigation Strategy" being developed at NIT Raipur tackles one of the most pressing concerns in AI ethics. Identifying and reducing biases does not just make AI fairer; it actively works towards a more equitable society.

The "Explainable AI Framework" developed by the Defence Institute of Advanced Technology, Pune, and Minecraft Technologies focuses on transparency. This effort aims to demystify AI decision-making, transforming these systems from inscrutable black boxes into understandable, accountable tools.

The "Privacy Enhancing Strategies" project, a collaborative effort between IIT Delhi, IIIT Delhi, IIT Dharwad, and the Telecom Engineering Centre (TEC), focuses on developing tools and techniques to protect user data using AI systems.

IIIT Delhi and TEC are working on the "AI Ethical Certification Framework," dubbed Tool Nishpaksh. This initiative will create a certification system to ensure AI systems meet ethical standards, acting as a quality check for AI fairness and safety.

The "AI Algorithm Auditing Framework," known as Tool Parakh, is being developed by Civic Data Labs. This tool will audit AI algorithms to ensure they work as intended and do not cause harm, checking for issues like bias, inefficiency, or unethical behavior.

Finally, the "AI Governance Testing Framework," a joint effort by Amrita Vidyapeetham and the Telecom Engineering Centre, will test and evaluate AI systems to ensure they comply with governance rules and regulations.

A Holistic Approach to AI Governance

Unlike isolated initiatives that tackle individual AI challenges, the eight projects are designed to work synergistically to create a comprehensive ecosystem. Machine Unlearning complements Privacy Enhancing Strategies by providing mechanisms to remove sensitive data. The Synthetic Data Generation project supports the AI Bias Mitigation Strategy by creating training datasets that minimize potential biases. The Explainable AI Framework directly supports the AI Algorithm Auditing Framework by making AI decision-making processes more transparent and analyzable.

The AI Ethical Certification Framework is an overarching quality assurance mechanism integrating insights from all other projects. The AI Governance Testing Framework ensures that the cumulative efforts meet regulatory standards, creating a robust, end-to-end approach to responsible AI development. This interconnected strategy transforms AI governance from a piecemeal approach to a comprehensive, mutually reinforcing system. Each project actively enhances the effectiveness of the others, creating a robust, multilayered framework for ethical AI development.

How India Stacks Up Globally

While India’s efforts align with global initiatives, they are more comprehensive in many ways. The U.S. lacks a centralized regulatory framework, relying instead on a patchwork of state-level laws and voluntary industry standards. and moreover, the new President has done away with the executive order passed by the previous administration on AI regulation. In contrast, India’s AI Safety Institute provides a centralized and multi-institutional approach to oversee AI governance, ensuring consistency and accountability across the board.

The EU’s AI Act is one of the most comprehensive regulatory frameworks in the world, mandating strict requirements for high-risk AI systems. However, the EU’s approach is overly restrictive, potentially stifling innovation. India’s techno-legal approach strikes a better balance, combining regulatory oversight with technological innovation to ensure ethical AI deployment without hindering progress.

The UK has taken a more decentralized approach, with organizations like the Alan Turing Institute focusing on ethical AI and bias mitigation. While this fosters innovation, it lacks the centralized oversight that India’s AI Safety Institute provides. India’s focus on algorithmic auditing and ethical certification ensures higher accountability and transparency.

Singapore’s AI Governance Framework is widely regarded as a model for ethical AI development, promoting voluntary certification and testing. However, India’s approach goes further by developing mandatory frameworks like Tool Nishpaksh for ethical certification and Tool Parakh for algorithmic auditing. This ensures that all AI systems meet stringent ethical standards, providing more user protection.

China’s AI strategy prioritizes state control and surveillance, with less emphasis on individual privacy and ethical considerations. In contrast, India’s focus on privacy-enhancing strategies and bias mitigation ensures that AI systems are fair, transparent, and accountable, making it a more user-centric approach.

Whose Safety Do India’s AI Norms Ensure?

While India’s AI safety norms appear to address various stakeholders, the prioritization suggests a hierarchy:

1. Government Interests: The primary focus is AI’s role in national security, digital governance, and law enforcement.
2. Data Sovereignty: Ensuring AI systems operate within India’s jurisdiction aligns with the government’s broader push for data sovereignty.
3. Economic Growth and Corporate Innovation: India’s AI initiatives aim to foster digital innovation, benefiting businesses and startups.
4. Citizen Protection: While fairness and bias mitigation are emphasized, enforcement mechanisms remain limited.

Adopting a Platform Approach to AI Safety

India can adopt a platformic approach to AI governance to foster innovation while ensuring AI safety. This involves creating an open and standardized AI ecosystem with modular and scalable regulations, tools, and frameworks. By developing interoperable AI safety protocols, India can encourage startups, researchers, and corporations to contribute to a shared governance infrastructure while maintaining compliance with ethical standards. The eight initiatives must merge into a single platform that controls and manages access to public and private data.

A platform approach would include:

• Open AI Governance Platforms: Establishing a centralized AI safety hub where developers can access regulatory frameworks, compliance tools, and certification mechanisms.
• Public-Private Collaboration: Encouraging government, academia, and industry partnerships to co-develop AI safety solutions. This is happening bilaterally, but it will have to occur in multi-lateral form as that is where innovation will leapfrog across singular company initiatives that are part of the current development model.
• AI Sandboxing Environments: Creating controlled testing environments to evaluate new AI models against ethical and safety benchmarks before deployment.
• Incentivizing Responsible AI: Providing tax benefits, funding, or certifications for AI systems that adhere to ethical AI frameworks.

By leveraging a platform approach, India can drive AI safety while ensuring an innovative, competitive AI landscape that benefits all stakeholders.

-ends