← LibraryState UncertaintyProject Management · Project Risk ManagementLesson 30/31← PrevNext →
ArticlePublished 6 Jul 20263 min readBy Kevin JoginDec-POMDPBayesian gamesincomplete informationinformation asymmetry

Project ManagementProject Risk ManagementAlgorithms for Decision MakingChapter 25

25Part V · Multiagent Systems

State Uncertainty

Now no one sees the whole truth, and each party knows things the others don't. You must reason about the hidden state — and about what everyone else knows.

Chapter 25 of 26 12 min read Original KEVOS® synthesis

Combine the strategic uncertainty of Part V with the state uncertainty of Part IV and you reach the most tangled situation of all: no one sees the whole picture, and each party knows things the others don't.

This chapter is where partial observation meets multiple agents. Now you must reason on two fronts at once — about the hidden state of the world and about what the other parties know, believe, and will do given their own private information. It is the formal home of information asymmetry, and it underlies procurement, bidding, disclosure, and disputes.

1Games of incomplete information

When agents hold private information — a contractor knows its true costs; a bidder knows its own valuation — the interaction becomes a Bayesian game. Each agent has a "type" capturing what it privately knows, and each reasons about the probability distribution over the others' types. A competitive tender is the archetype: you choose a bid without knowing rivals' costs, so you reason about the distribution of what they might do and best-respond to that. Strategy becomes inseparable from belief about who you're dealing with.

true state hidden from both partial view partial view Party A Party B "what does B know… and what does B think I know?" "what does A know… and how much have they seen?"
Figure 1. Each party sees only a slice of the hidden truth and must reason not just about the state, but about what the other knows — and what the other thinks they know. These nested, recursive beliefs are the essence of strategic interaction under information asymmetry.

2Decentralised decisions and recursive beliefs

When several agents each act on their own partial observations, possibly without sharing what they see, you have a decentralised POMDP — among the hardest decision problems known, precisely because each agent must reason about others who are themselves reasoning under uncertainty. That opens an unnerving regress of recursive beliefs: what I believe about the state, what I believe you believe, what I believe you believe I believe, and onward. Exact treatment is intractable at any real scale, so practical work leans on bounded reasoning — modelling a few levels of this regress, not infinitely many.

Key idea

When information is held unevenly and no one sees the whole state, strategy and belief fuse: your best move depends on what others privately know and on what they think you know. Reasoning a level or two into that regress is usually as far as anyone can — or needs to — go.

What it means in practice

Information asymmetry is the water project professionals swim in — in procurement, claims, and negotiations, the other side always knows things you don't, and vice versa. Reason explicitly about what each party privately knows and, one level deeper, about what they assume you know; a surprising amount of negotiation turns on that second level. And recognise asymmetry as something you can act on: sometimes the winning move is to reduce it (due diligence, disclosure, independent verification), and sometimes it's to manage your own signalling with care.

Continue learning

Actor–Critic MethodsArticle · Project Risk ManagementAlgorithms for Decision MakingArticle · Project Risk ManagementApproximate Value FunctionsArticle · Project Risk ManagementBeliefsArticle · Project Risk Management