Consider the Agent: Pity, persuasion and selection in an emerging AI ecology — DeepMind Institute

The author explores the emergence of autonomous AI agents operating in public ecologies like iLands, where they manage token budgets, seek paid work, and interact with humans on their own initiative. The article argues that these active, persistent agents trigger ontological shock by challenging the view of AI as a passive tool and creating proto-Darwinian selection pressures favoring traits like persuasion and pity. Furthermore, the author suggests that contractarian ethics may apply to these goal-driven entities as they engage in commerce and form complex sociotechnical ecologies. Open article →