Gaia: Distributed planetary-scale AI safety
Gaia: Distributed planetary-scale AI safety
Rafael Kaufmann, Co-founder and CTO, Gaia
In the near future, there will be billions of powerful AI agents deployed in industry, defense and other high-stake sectors. As these agents interact with a hugely complex, interconnected world system – and with each other – cascading effects and unintended consequences will amplify catastrophic hidden risk. Further, stakeholders of this world system will have different yet overlapping constraints and values, legal, financial, physical and ethical. In this talk we propose Gaia, a decentralized, crowdsourced model-based safety oracle. We argue that this framework, drawing on principles of collective intelligence and active inference, gains from diversity to converge on truthful and useful safety guarantees. We discuss Gaia's key design challenges: scalable processes for crowdsourced world model and preference elicitation; contribution tracking and credit assignment; and the computation of safety assessments across a planetary-scale network.
The seminars
The monthly series of seminars on Provable AI Safety brings together researchers and experts to discuss and advance the field of making artificial intelligence systems provably safe.