Agent Foundations 01: Embedded Agency
What does it mean for an intelligent system to exist inside the world it is trying to understand?
Classical AI often imagines an agent as something separate from its environment, observing and acting from the outside. Real-world intelligent systems, however, are embedded within the very world they model.
They have limited computational resources, can influence both themselves and their surroundings, and must reason while being part of the system they are trying to understand.
In this first DASC Agent Foundations session, we’ll introduce Embedded Agency, one of the central problems in theoretical AI alignment. We’ll explore why classical assumptions about agents and environments break down, what new challenges emerge when an agent is embedded in its environment, and why this question matters for building safe and reliable advanced AI.
This session is designed as an accessible introduction. No prior background in AI alignment is required.
Topics
What is an embedded agent?
Why classical agent models are insufficient
Self-reference and world models
Bounded reasoning and computation
Why embedded agency matters for AI alignment
Format
Guided discussion · Presentation · Open Q&A
About DASC
DASC (Delhi AI Alignment & Safety) is a Delhi-based community for exploring AI alignment, safety, and the theoretical foundations of intelligent systems.
DASC operates under Amyndra Labs and is part of the broader AI Safety India community network.
Hosted by
DASC · Delhi AI Alignment & Safety Collective
AI Safety India · DASC