Why in the News
The summit between the Chinese President and the US President in Washington produced only a modest package on artificial intelligence (AI). The two powers could not even agree on what to call the technology, and their rivalry now shapes the AI rules India must live under.
What did the summit deliver on AI?
- Groundwork in New York: At the first AI talks under the two countries’ trade mechanism, on 20 September, Washington proposed a channel to notify AI incidents, meaning AI events with national security consequences.
- Modest package: The summit confirmed a dialogue on risks and benefits, a channel for incidents and a next meeting by November.
- Dispute over the name: The White House says the leaders agreed to use the term “super intelligence”. China’s list speaks of a “China-US dialogue on artificial intelligence”.
- Nuclear control dropped: At Lima (November 2024), the two sides affirmed human control over nuclear-use decisions. Neither fact sheet repeats it, and only China’s list has a military crisis communication memorandum.
- The takeaway: The two powers agreed to talk about AI risk without agreeing on what it is, so the channel can handle incidents but not set rules.
Why will rivalry, not dialogue, shape the AI order?
- Race to be won: The US President calls the case for slowing AI a “hoax”. He prefers prosecutors to police AI harms afterwards over regulators acting in advance.
- Language of control: The Chinese President wants AI “always under human control”. His call for both sides to “play to their strengths” criticised US export controls.
- Safety as containment: Beijing reads US safety talk as a cover for holding back China. State media accused a US firm of calling distillation (training a cheap model on a stronger model’s outputs) a threat.
- Regime security first: China’s security minister calls AI “the main battlefield” of rivalry. Party control of labs limits what models may say, not how fast they grow more capable.
- Tools of competition: The US relies on chip controls and Pax Silica, a US-led network of trusted chip suppliers. China pushes chip self-reliance and open-weight models, which anyone can download and adapt.
What concerns does this raise for India?
- G2 overlay (a US-China duopoly): The two powers hold most frontier compute, the computing power behind top models, so they can set the rules. In the nuclear order, such deals justified discriminatory controls on others.
- Two AI systems, two sets of standards: India faces pressure from both camps:
- Washington expects trusted partners to keep China out of their AI ecosystems;
- Beijing proposed a BRICS open-source AI community at the New Delhi summit;
- India is unlikely to join China’s World Artificial Intelligence Cooperation Organisation (WAICO), a China-centric initiative like the Belt and Road Initiative.
- Pull of cheap Chinese models: Chinese open-weight models are competitive, cheap and adaptable. Eg. Alibaba’s Qwen has spawned more than 150,000 derivative models.
Challenges
- Private adoption: Stopping private firms building low-risk apps on Chinese models is hard, as low cost attracts Western and Indian companies. Eg. Singapore, Malaysia and Brazil use them.
- American dependence: Access to US models can vanish quickly. Eg. June’s brief cut-off of foreign access to top American models.
- No rival offer in BRICS: India has no alternative yet to the Chinese AI offering in the grouping.
Way Forward
- Multilateral norms: Welcome US-China risk reduction, but insist on multilateral frontier AI norms and a place in incident-notification arrangements.
- Nuclear human control: Make human control over nuclear-use decisions part of India’s nuclear discourse, and press all nuclear powers to affirm it.
- Dual and differentiated de-risking: Bar Chinese models from government systems, critical infrastructure and sensitive data; require security testing and local hosting for private use.
- Capability at home: Build compute, chips, models, datasets, talent and the ability to test frontier systems independently.
- Shape BRICS: Insist that BRICS AI initiatives be consensus-based and technology-neutral, rather than cede the space.
Conclusion
The new channel manages incidents between two rivals but leaves the rules for everyone else unsettled. What to watch is whether India is admitted to incident notification, and on what terms the BRICS AI community takes shape.
Government Initiatives for AI capability in India
- IndiaAI Mission (2024): Treats AI as a public good, built on shared compute, open datasets and decentralised talent development.
- Shared compute: More than 38,000 graphics processing units (GPUs), the chips that train AI models, form a national compute grid for startups and researchers.
- AI Kosh: Offers over 360 curated non-personal datasets across sectors such as agriculture, health and climate.
Matching Previous Year Question
“[2025, GS2, 10 marks] With the waning of globalization, post-Cold War world is becoming a site of sovereign nationalism. Elucidate.”
