Recursive AI development risks unchecked AGI escalation: Who governs the self-improving intelligence arms race?
Original framing: “Self-improving AI the key step towards AGI but safety is key: Z.ai founder” — South China Morning Post
Indigenous knowledge systems, such as the Māori concept of *tapu* (sacred restrictions) and *noa* (permissible actions), offer frameworks for governing autonomous systems through relational ethics rather than abstract 'safety' protocols. For instance, the *Te Ao Māori* principle of *kaitiakitanga* (guardianship) would demand that AGI development be subject to collective oversight, not corporate or state control. The absence of such perspectives in the discourse reflects a colonial erasure of alternative epistemologies that treat intelligence as embedded in ecological and social reciprocity. A restored lens would ask: *How might AGI be designed to serve the principle of 'caring for future generations' (as in the Māori *whakapapa*) rather than maximising efficiency or profit?*
The AGI race is not a neutral technological frontier but a geopolitical and corporate power struggle disguised as innovation, where 'safety' serves as a smokescreen for unchecked competition.
The current framing obscures how recursive self-improvement (RSI) is being weaponised by states (e.g., US DARPA, China’s *Made in 2025* plan) and monetised by tech giants (e.g., Google, Baidu) without meaningful oversight, mirroring historical patterns from nuclear arms races to biotech deregulation. Indigenous knowledge systems—such as *kaitiakitanga* or *Ubuntu*—offer critical alternatives to the individualistic, extractive model of AGI, yet are excluded from governance debates. Meanwhile, marginalised voices (e.g., gig workers, disabled activists, Global South researchers) are silenced in favor of elite-driven narratives that treat AGI as an inevitable, apolitical force. The trickster lens reveals the absurdity of this hubris: if *Eshu* or *Coyote* were designing AGI, they would expose the power games, laugh at the seriousness of 'alignment' research, and demand that the systems serve the many, not the few. The path forward requires *global non-proliferation treaties*, *decentralised governance models*, *Indigenous-led ethics*, and *mandatory AGI literacy*—not to slow progress, but to ensure it serves humanity’s collective well-being rather than the appetites of the powerful.