Governing the AI Ecosystem | July 11, 2024

Introduction

On July 11, 2024, a comprehensive discussion took place to tackle various questions surrounding the governance, regulation, and innovative growth of Artificial Intelligence (AI) in India. Experts from diverse fields including law, technology, public policy, and business gathered to debate essential issues concerning AI deployment, liability, regulatory frameworks, and measures to support innovation. This article encapsulates the key points, arguments, and propositions made during the event.

AI Regulatory Frameworks: Liability and Compliance

Context and Importance

The event opened with an understanding of the various aspects of AI governance, specifically looking into liability and regulation frameworks as seen in the EU AI Act, US Executive Orders, and UK's proposed guidelines. The aim was to explore how AI systems, whether recommendation engines, generative models, or AI agents, could be regulated in India.

Discussion Highlights

1. AI as a Panel Discussion: Participants were informed that they themselves constituted the panel, fostering an environment of open dialogue.
2. AI Imperfection: All participants acknowledged the imperfect outcomes of AI tools such as ChatGPT and Claude, posing challenges for regulatory frameworks.
3. Case Studies:
   • Air Canada’s chatbot incident led a court to hold the company responsible for a discount offered, questioning liability in AI-driven customer service.
   • Alexa's unintentional mass purchasing of dollhouses after a TV broadcast highlighted the concerns over agentic AI systems acting autonomously without explicit human oversight.

Questions and Opinions

• Sectoral Regulation vs. Omnibus AI Act: Participants debated whether AI should be regulated through a specific AI Act or incorporated within broader acts like the Digital India Act (DIA). Opinions varied, with some supporting sectoral regulations focusing on specific industry needs, while others advocated for an overarching AI Act to establish baseline principles.
• Intentionality and Autonomous Decision-Making: Much of the discussion centered on where liability lies in cases where autonomous AI systems make decisions. The consensus formed around the principle that stricter liability should be assigned to AI developers and deployers, depending on the harm and risk associated.
• Public vs. Private Outputs: The issue of whether outputs from AI should be considered public or private was debated, with a general agreement that transparency and clear communication are essential.

Measures to Support Innovation

Government Initiatives

1. AI Missions and Budgets: The Indian government's commitment to the India's AI Mission with a significant budget outlay for GPUs aimed at fostering local innovation.
2. International Best Practices: Drawing from examples in the EU, US, and China on how to build a conducive environment for AI technologies through public-private partnerships, regulatory sandboxes, and national data strategies.

Recommendations for India

• Focus on Talent: Participants stressed the need for attracting and retaining AI talent in India by offering incentives, improving research funding, and creating an ecosystem conducive to high-quality research and development.
• Public-Private Partnerships: Emphasis was placed on forming alliances with private sector giants to build compute capacity and data centers.
• Regulatory Sandboxes: Issuing more regulatory sandboxes to allow startups to innovate without the heavy burden of compliance from the outset, while still maintaining a critical oversight mechanism.
• Sector-Specific Regulation: Several argued that regulations should be tailored to specific industry risks, notably in healthcare, finance, and legal sectors, thereby balancing the need for innovation with consumer protection.

Addressing Foundational Gaps

• Non-Personal Data Strategy: Strategies for sharing non-personal data between companies to drive innovation while safeguarding privacy.
• Compute Procurement and Management: The debate highlighted the need for a significant boost in GPU procurement and suggested transitioning from a capital expenditure model to an operating expenditure model to accommodate future technological advances.

Conclusion

The discussion underscored a collective call for balanced development in AI - promoting innovation while ensuring responsible and ethical use. A detailed multi-stakeholder approach was recommended to chart India’s course in AI governance effectively.

Keywords

• AI regulation
• Liability
• AI governance
• Public-private partnerships
• AI innovation
• Sectoral regulation
• Digital India Act (DIA)
• Non-personal data
• Compute procurement
• Talent retention

FAQ

What are the key principles proposed for AI regulation in India?

The key principles proposed include focusing on transparency, sector-specific regulations, and fostering a balanced regulatory environment that promotes innovation while ensuring accountability and trustworthiness in AI systems.

How should liability be assigned in cases where AI systems act autonomously?

Liability should be assigned based on a principle of proximity and control over the AI's outputs and actions. This often implies greater responsibility for developers and deployers, particularly when harm occurs beyond predictable use cases.

What measures can support AI innovation in India?

Measures to support AI innovation include government-led initiatives like the India AI Mission, public-private partnerships, regulatory sandboxes, improved research funding, and efforts to attract and retain AI talent within the country.

How should the public-private partnerships be structured in the AI sector?

Public-private partnerships could be structured to build national compute infrastructure, create shared resources like data centers, and foster collaborative research and development ecosystems to drive advancement in AI.

Why is there a debate between risk-based and harm-based regulatory approaches?

The debate exists because a harm-based approach focuses on addressing realized harms, which may be too late in some cases. In contrast, a risk-based approach aims to preemptively manage potential risks, fostering a safer development and deployment environment for AI systems.