Definition
Plain language
A way of describing a decision problem where the agent can't see the whole situation and must act on partial information.
As stated in the literature
Partially Observable Markov Decision Process, an MDP extension where observations only partially reveal the underlying state; used to formalize active-perception video agents that carry a persistent text notebook as their belief state while discarding raw frames.
Also called: partially-observable decision process, partially observable Markov decision process
Why it matters: It's the standard way to reason about decisions under uncertainty, which matches most real-world agents that never see the full picture.
For example, a robot in a dark room that can only feel what's directly in front of it must decide where to move without seeing the whole layout.
Heard on the show
“They formalize it as a partially-observable decision process, but honestly the jargon is heavier than the idea.”Episode 154 — How a 7B Model Out-Investigates a 72B One by Choosing What to Look At