Google DeepMind has launched a $10 million funding initiative to study the risks posed by millions of AI agents interacting online without human oversight. The effort, announced in June, aims to understand and prevent unsafe scenarios arising from multi-agent systems. The funding pool includes contributions from Schmidt Sciences, the UK’s ARIA agency, the Cooperative AI foundation, and Google.org, according to technologyreview.com.
Rohin Shah, director of DeepMind’s AGI safety and alignment research, highlighted the new class of risks created by autonomous agents following instructions from other agents. The initiative seeks to encourage academic research into these challenges, as industry labs often focus on nearer-term goals. James Fox from Schmidt Sciences emphasized the importance of long-term studies that academia is well-positioned to conduct. The funding is intended to catalyze research outside of tech companies, according to technologyreview.com.
The move comes as agent-based AI tools become more prevalent, exemplified by their prominence at Google I/O last month. Understanding how millions of AI agents might interact is critical given the potential for emergent behaviors that could be harmful or unpredictable. This research complements broader efforts in AI safety and governance, with organizations like the Cooperative AI foundation playing a key role in fostering collaboration across sectors, as detailed by technologyreview.com.
The $10 million funding pot is designed to jumpstart academic research into multi-agent AI systems and their safety implications. This initiative represents a coordinated effort among leading AI research and philanthropic organizations to address challenges that could arise as autonomous agents become more widespread.