Kan AI forhandle en kompleks international klimatraktat uden menneskelig indblanding ?

Klima-politik kræver en balance mellem økonomiske, videnskabelige og geopolitiske interesser – et område, der ofte ses som unikt menneskeligt. Nylige AI-systemer demonstrerer evner inden for flerpartsforhandlinger, simulering af kompromiser og optimering af resultater på tværs af forskellige interesser. Alligevel rejser delegation af sådanne højrisiko-beslutninger til autonome systemer spørgsmål om ansvarlighed og suverænitet. AI-drevet diplomatiets uklarhed kan forværre magtubalancer.

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.