Super AI is a systemic risk, but generative and agentic AI are not — Phar Kim Beng

On September 21, artificial intelligence (AI) terminology has raced ahead of reality, with generative, agentic, artificial general intelligence, and potential superintelligence being discussed as though they are part of the same technological continuum. However, these terms do not carry the same risks. Generative AI can produce various forms of content, while agentic AI allows systems to pursue objectives through actions, sometimes with limited human intervention.

Though both pose significant problems, such as hallucination, cyber vulnerability, manipulation, privacy breaches, labor displacement, and concentration of computational power, they are not necessarily systemic risks. Super AI, an artificial intelligence with capabilities vastly beyond human beings across numerous domains, combined with a capacity for recursive self-improvement, represents a deeper threshold.

The central issue is whether human institutions can retain control over systems capable of improving their capabilities quickly. Super AI could become systemic, impacting not just the laboratory, company, or country where it originated, but also geopolitics, as countries such as the US, China, and European nations view advanced AI as part of their national power.

Southeast Asian nations like Singapore, Malaysia, Indonesia, Vietnam, Thailand, and the Philippines are developing important positions within the regional AI and semiconductor ecosystems, making them increasingly close to the global AI economy infrastructure. However, Asean should resist treating every form of AI as an existential danger, as over-regulating generative and agentic AI could impose substantial economic costs.

The response should be differentiation, not panic. Generative AI requires standards related to accuracy, copyright, privacy, transparency, and responsible deployment, while agentic AI requires stronger safeguards due to the potential for autonomous actions with consequences beyond information generation. Super AI would require another level of governance, as its consequences could be extraordinary.

The analogy with nuclear weapons is instructive, as humanity differentiated peaceful nuclear applications from weapons systems while constructing elaborate regimes of safeguards, verification, deterrence, and arms control. Asean should support international discussions on advanced AI safety while preserving sufficient regulatory space for beneficial AI development, promoting a clearer distinction between various AI forms to avoid oscillating between complacency and alarmism.