Multi-level agency
Coarse-graining empowerment
2025-06-09 — 2025-07-09
Wherein the preferred level of abstraction for explaining behaviour is considered, and a concrete mapping between utility and biological fitness across genes, organisms and collectives is examined.
Looking at a complicated system such as a human society or a microchip or a neural network, at what level of abstraction do we want to think about it to best understand what it is doing? E.g. is technology itself an agent?
Probably some classic question about “emergence” and “complexity” and “hierarchy” and “levels of organisation”. We might also think about the modern synthesis of evolutionary biology, which is a theory of how evolution works at different levels of organisation, from genes to organisms to populations.
1 Incoming
We might care also about the suggestive-but-indirect mapping between utility and fitness…
Alignment of Complex Systems Research Group – ACS Research Program
Our research seeks to build towards the following goals:
- A science of the space of intelligent systems
- Naturalised theories of agency, including theories that account for the multi-scale nature of agentic behaviour
- Nuanced accounts of how humans reason, act and value as it pertains to solving the AI alignment problem
- An understanding of the underlying principles that govern collective behaviour between humans, between humans and AIs, and between AI systems