Power, Protection, and Progress: Legislating for the AI Era

Contents

Executive Summary

This panel discussion brings together Indian and Israeli legislators and policymakers to examine how democracies should govern artificial intelligence development and deployment. The core argument centers on framing AI governance primarily as a democracy and sovereignty question rather than merely a technical one, emphasizing that legislators must balance innovation with protection while ensuring equitable access across populations and preservation of indigenous knowledge systems.

Key Takeaways

Frame AI as Governance, Not Technology: Legislators should evaluate AI through democratic impact, economic sovereignty, and social equity lenses rather than demanding technical mastery.
Sovereign Compute Capacity = 21st Century Power: Nations without design, manufacturing, or export-control autonomy in AI chips will be strategically dependent. India must act on the trio of access, supply diversification, and domestic infrastructure.
Modernize Laws Before Regulating AI: Rather than creating new AI-specific laws (European approach), update existing frameworks (copyright, fair use, data protection) to address 21st-century realities first.
Protect Undigitized Knowledge Proactively: Programs like Bhashini (Indian government initiative for regional language digitization) are essential to prevent civilizational wisdom and minority languages from being erased by AI models trained on English-dominant datasets.
Balance Innovation with Trust Through Education, Not Just Regulation: Regulation alone cannot create safe AI. Critical thinking education, transparency requirements, and measured (not reactive) implementation are equally important to maintaining public trust in both AI and democratic institutions.

Key Topics Covered

AI Governance as Democracy: How AI shapes information systems and democratic processes
Compute Sovereignty & Supply Chains: GPU access, manufacturing monopolies, and strategic autonomy for nations like India
Legislative Preparedness: Whether legislators need deep technical knowledge or should focus on economic and societal implications
Social Media Regulation: Age restrictions and child protection measures against addictive platforms
Copyright & Fair Use: Legal frameworks for training data in the AI era; Israeli approach vs. need for Indian copyright modernization
AI Access & Digital Divide: Preventing an "intelligence divide" through equitable access to AI tools
Deepfakes & Information Integrity: Threats to democratic trust and voter decision-making
Knowledge Preservation: Protecting indigenous languages, oral traditions, and civilizational wisdom from being excluded from AI training data
Regulation Timing & Scope: When and how extensively to regulate AI without stifling innovation
Indigenous AI Development: Building sovereign AI capabilities rather than depending on foreign companies

Key Points & Insights

AI as Information System Problem: Democracy fundamentally depends on how information is shaped, amplified, and acted upon. AI transforms all three dimensions, making it a governance question before it's a technology question.
Legislators Don't Need Deep Technical Knowledge: Unlike nuclear energy experts, legislators don't need to understand LLM architectures or GPU specifications. Instead, they should focus on economic implications, job displacement, youth impact, and where AI-derived value flows geographically.
Compute is a Sovereignty Issue, Not Just Supply Chain: Control over AI compute capacity in the 21st century parallels control over oil, gas, and steel in the 20th century. India currently lacks design (Nvidia monopoly), manufacturing (TSMC monopoly in Taiwan), and export control autonomy (US restrictions).
Three-Pronged Approach to Compute Access:
- Secure direct access to the global AI ecosystem
- Diversify supply sources to reduce geopolitical dependence
- Build domestic infrastructure (data centers, chip manufacturing)
Indigenous LLM Development is Necessary: India's talent advantage (world #1 in AI skill penetration, expected to double by 2027) alone is insufficient without compute capacity. India must develop its own LLMs to ensure data stays domestically valuable and prevents brain drain.
Regulation Must Be Measured, Not Reactive: The "winter of regulation" phenomenon causes governments to over-regulate after crises out of fear, stifling innovation. Regulation should be evidence-based, strategic, and introduced at the "sweet spot" when technology is mature enough but not too late to influence.
Fair Use Definition Urgently Needs Modernization: India's copyright law dates to the 1950s-60s and defines fair use based on then-available media (journals, newspapers). It hasn't been updated for digital-age practices like AI training on published content, educational remixing, or code commenting.
Knowledge Displacement Risk: AI models are being built on a fraction of human knowledge. Oral traditions, community practices, civilizational wisdom, and non-digitized knowledge won't be captured by AI systems, potentially erasing crucial indigenous knowledge.
Every 14 Days a Language Vanishes: Language death takes with it collective wisdom developed over generations. AI training data reflects this bias—languages with less digital presence become underrepresented, accelerating knowledge loss.
Access Without Education Enables Manipulation: Free access to AI tools without critical thinking education and media literacy leaves populations vulnerable to deepfakes and disinformation, eroding democratic trust and causing disengagement from civic participation.

Notable Quotes or Statements

Udwell Kumar: "Democracy at its core is an information system. It depends on what people want, what they experience and what they want to express... AI transforms information. That's why we all are here. It makes it a democracy question more than a technology question."
Raghav Chhatta: "In the 20th century power was controlled by those who had oil, gas and steel. In the 21st century power will be controlled by those who have access and AI compute capability, AI capability as well as chip manufacturing abilities."
Sri Krishna G: "It's not a talent problem, it's not a capital problem, it's a compute problem."
Advocate Sarith Felba (Israel): "Artificial intelligence needs human trust. If we'll move too fast towards AI, I'm afraid that the trust in AI and in parliament might reduce... we call it in Hebrew the 'winter of regulation'—when something happens, regulators get afraid and introduce severe punishments, stifling innovation."
Raghav Chhatta (on knowledge displacement): "Anything that is not digitized or not searchable, is not codified, is not something that AI's compute brain will apply itself to... we are building future models based on a fraction of human knowledge and human intelligence."

Speakers & Organizations Mentioned

Panelists:

Udwell Kumar: Co-founder of Cognissi (public benefit corporation), Fellow at Harvard Kennedy School; founder of Quantum Alliance
Sri Krishna G: Member of Parliament (Lok Sabha), Andhra Pradesh; founding member of AI Legislators Forum; CS degree holder
Sri Raghav Chhatta: Member of Parliament (Rajya Sabha); legal background
Advocate Sarith Felba: Senior Director, Ministry of Justice, Israel

Organizations & Institutions:

Future Shift Labs (conference organizer)
AI Legislators Forum
Harvard Kennedy School
UN Food and Agriculture Organization
Ministry of Justice, Israel
Indian Parliament (Lok Sabha, Rajya Sabha)
Cognissi
Quantum Alliance

Companies/Systems Referenced:

Nvidia (GPU design monopoly)
TSMC (Taiwan Semiconductor Manufacturing Company — 90% of global chip manufacturing)
OpenAI, Meta (Facebook), YouTube, Instagram, TikTok
Apple App Store, Google Android
Swiss initiatives on indigenous LLMs

Government Initiatives Mentioned:

Aadhaar (India's digital identity system at population scale)
UPI (Unified Payments Interface — India's open-source payment infrastructure)
Bhashini (Indian government program for regional language digitization and AI)
AI Mission (separate Indian government AI development program)

Technical Concepts & Resources

LLMs (Large Language Models): Models like those from OpenAI; central to discussion of compute requirements and training data
GPUs (Graphics Processing Units): Bottleneck resource for AI development; India currently has ~34,000 GPUs vs. global need
Generative AI: Autonomous systems that can generate content; discussed in context of regulation and access
Deepfakes: AI-generated synthetic media used to impersonate individuals; major democratic integrity concern
Tokens: Units of text processed by LLMs; value of tokens generated in India matters for economic sovereignty
Fair Use: Legal doctrine allowing limited use of copyrighted material; Israeli government issued legal opinion (2020) classifying AI training as fair use; India needs copyright modernization to define this
Compute Capacity: Processing power available for AI model training and inference; identified as strategic national asset
Data Localization: Ensuring training data and model outputs remain within national borders for sovereignty

Regulatory Frameworks Mentioned:

EU AI Act (2019 introduced, regulates technology broadly rather than specific harms)
Indian Copyright Act (1950s-60s vintage; urgently needs fair use modernization)
Indian Constitution Article 19 (freedom of expression with constitutional safeguards as model for AI regulation)
US Senate questioning of tech CEOs on ad-based business models

Context & Meta-Observations

This transcript reflects discussions at an AI summit in New Delhi (Feb 16-20, 2025) and includes audio quality issues and repeated phrases suggesting live transcription artifacts. Despite these, the substantive arguments about compute sovereignty, legislative preparation, knowledge preservation, and measured regulation are consistent across all speakers. The panel represents a rare moment of substantive, non-partisan technical policy discussion between democracies (India and Israel) on AI governance frameworks.