Pode a IA negociar um complexo tratado internacional sobre o clima sem intervenção humana ?

A política climática requer equilibrar interesses económicos, científicos e geopolíticos — um domínio muitas vezes visto como exclusivamente humano. Sistemas de IA recentes demonstram capacidades em negociação multipartidária, simulando compromissos e otimizando resultados em interesses diversos. No entanto, delegar decisões tão críticas a sistemas autónomos levanta questões de responsabilidade e soberania. A opacidade da diplomacia impulsionada por IA poderia agravar desequilíbrios de poder.

Background

Climate policy routinely pits economic growth against emissions cuts and requires navigating entrenched national interests—traditionally a human domain. Recent AI systems have demonstrated an ability to engage in multi-party negotiation, exploring trade-offs and synthesizing compromise texts when trained on structured data. Still, the domain remains beset by concerns: even benign-looking AI could tilt the balance of power if its internal reasoning remains hidden from diplomats. Current AI tools are confined to support roles—processing climate projections, drafting non-binding annexes, or forecasting the ripple effects of proposed carbon schedules—yet they stop short of autonomous consensus-building. The canonical limitation is sovereignty: any binding treaty must ultimately be ratified by sovereign states, whose political legitimacy rests on visible human judgment. Fully automated negotiation also sits in legal limbo; international law assigns final authority to human representatives, not algorithms. Ethical frameworks such as the 2023 Asilomar AI Principles explicitly caution against ceding high-stakes diplomatic decisions to autonomous systems whose decision paths cannot be publicly scrutinized. As of 2026, no operational AI has negotiated, signed, or ratified an international instrument; existing deployments merely act as advisory engines inside human-led processes.