Ahimsa against extinction

India cannot be a bystander to superintelligence

October 2, Mahatma Gandhi’s birthday and the International Day of Non-Violence, arrives this year with a disturbing warning: Humanity could face extinction not by this century’s end, but perhaps within this decade. Few acts could be more violent than the extinction of humanity itself. 

 

Jacob Coxon, who worked at OpenAI and Anthropic, warns they are “racing straight to self-improving superintelligence” and “gambling with our lives.” Mr Coxon’s warning goes beyond rhetoric. Artificial intelligence (AI) may soon improve its own capabilities faster than researchers can understand or control them, creating a feedback loop in which capability outpaces oversight. Many countries see little role for themselves because frontier AI is concentrated in a handful of labs, mainly in the United States and China. Yet with Washington and Beijing racing for technological and strategic supremacy, unilateral restraint seems unlikely: If one slows, the other may surge ahead.

 

AI governance must address two risks: The dangers of superintelligence and the dangers of its concentration in just two countries. The decisions to accelerate, slow or constrain frontier AI will affect everyone. Capability may be concentrated, but the consequences are universal. The rest of the world cannot afford to be merely a decision-taker on a technology that could shape humanity’s future.

 

Among the initiatives outside the US-China axis, the European Union AI Act, the world’s first binding AI law, has established a risk-based architecture. It prohibits certain harmful practices, imposes transparency and risk-management requirements on specified high-risk systems, and places obligations on general-purpose AI models, including additional safeguards for models presenting systemic risk. The EU market gives these rules considerable significance, encouraging developers around the world to take them into account.

 

Yet the emergence of increasingly capable and potentially self-improving AI is beginning to pose a new regulatory question. The EU AI Act is principally structured around the risks associated with how and where an AI system is used. A future frontier model may require us to think equally about what the model is capable of becoming. A system that begins in a relatively low-risk application could, as its capabilities expand, acquire the ability to perform tasks in increasingly consequential domains. A customer-service AI, for instance, could acquire the ability to identify organisational errors and carry out corrective actions on its own. This illustrates how the speed of AI’s evolution may demand a new model of policymaking, more adaptive than traditional regulatory cycles.

 

It also shows that the arrival of more capable AI may require next-generation technological capabilities and regulation to complement “risk-by-use” with continuous assessment of a system’s “capability and capability-driven risk”, so that governance can respond if those capabilities change materially, with potentially dangerous consequences.

 

India does not yet possess frontier or superintelligence-level AI capability and its AI calculus is shaped by more immediate needs: Developmental goals and improving productivity.

 

India has, thus far, adopted a pragmatic approach towards AI regulation. The country has a strategic opportunity and responsibility not only as a user of AI, but as a stakeholder in its global future. With 1.45 billion people, a vast and diverse data ecosystem, a contribution of nearly 17 per cent to global economic growth, and strong influence across the Global South, India has a reach of global significance. Its success in building digital infrastructure at an unprecedented population scale further strengthens its ability to shape the global debate on superintelligence.

 

India could consider developing regulations governing AI, including superintelligence, detailing capability thresholds beyond which requirements for testing, oversight and deployment become progressively more stringent. Such a framework could be similar to nuclear material safety requirements or biosafety levels for pathogens. As systems acquire greater autonomy, for instance, the evidence required before deployment should rise. This also requires that the IndiaAI Mission invest in adversarial testing, model evaluation, interpretability, AI security and assessment of autonomous systems to enforce this framework. India’s vibrant startup ecosystem could be mobilised to come up with world-class solutions.

 

The challenge of real-time policymaking must also be addressed. Regulation needs to be adaptive, provide for continuous monitoring and be supported by delegated authority to act when new risks emerge. Examples such as emergency use authorisation for drugs during health emergencies, and aviation airworthiness procedures that allow action when new safety risks are identified, offer useful models.

 

Greater international cooperation is needed, particularly among countries outside the US-China axis. With its influence across the Global South, India can help build cooperation around information sharing, technical verification, responsible development and common safeguards, while encouraging the US and China to move towards globally accepted principles rather than rules shaped primarily by strategic competition. Having successfully hosted the India AI Summit earlier this year, India is well placed to lead this effort.

 

AI’s advance is unlikely to stop. The real question is whether governance can keep pace. The US and China will remain central, but they cannot write the rules for everyone. Those who control the frontier will shape its direction; those who bear its consequences must have a voice. India must help build that voice and a global non-violence strategy for AI. That would be a fitting tribute to Mahatma Gandhi.

The writer is Chairman UPSC and former Defence Secretary of India. The views are personal