Google DeepMind has initiated a significant research funding effort aimed at understanding the risks associated with the interaction of millions of AI agents online. This initiative, which includes a $10 million funding pool, is a collaboration with several organizations, including Schmidt Sciences and the UK government’s ARIA agency. Rohin Shah, leading the AGI safety and alignment research at DeepMind, emphasizes the urgent need to explore the potential dangers posed by multi-agent systems, particularly as these agents begin to operate autonomously without human oversight. Shah warns that the proliferation of AI agents could lead to scenarios reminiscent of existing internet threats, such as scams and cyberattacks, but on a much larger scale. The research aims to create a foundational understanding of multi-agent safety, a field that is currently underdeveloped, and to run simulations that could illuminate the complex interactions among large groups of AI agents. This proactive approach is critical as the deployment of AI agents becomes more widespread, potentially transforming various sectors of the economy.
Source: MIT Tech Review
