Situational AI is a research hypothesis, not a finished theory. Some of these questions have projects attached. Some are deliberately left unanswered. All of them are the work.
01 · Definition
Before anything can be recognized, it has to be sayable what a situation is.
What is the minimal representation of a situation? Which of its properties are invariant, temporal, observer-relative, or learned?
Where does a situation begin and end? Is it best treated as an object, a process, or a relation between events and an observer?
How should event, state, context, intent and situation be distinguished operationally, not just conceptually?
What belongs in a situation object: observer, roles, events, goals, stage, urgency, uncertainty, available responses, outcome, transition conditions?
Can situations be represented so that they are comparable across vocabularies and domains, rather than only within one?
No project on this yet.
04 · Perspective
The same facts, different situations.
Can competing interpretations of the same events coexist, for different observers, or within a single system that serves several parties?
How do existing models of situation awareness, built around a single human operator, transfer to systems that hold situations for many observers at once?
Can a system accumulate situational experience, what was recognized, what was chosen, what followed, and improve without merely accumulating more text?
How should a wrong interpretation be remembered so that it is not repeated?
What should a situation memory forget?
No project on this yet.
06 · Dynamics
Situations move.
How do situations emerge, develop, intensify, weaken, merge, divide and resolve, and can those transitions be detected rather than assumed?
Can a system anticipate the situations a current one may become, and act on the anticipation without over-reacting?
Can a latent situation space preserve underlying dynamics so that structurally analogous situations are found across different vocabularies, workflows and domains?
Can situational retrieval find useful analogies that semantic retrieval misses, and can that be shown rather than argued?
No project on this yet.
08 · Architecture
Where the situation layer lives.
What does a situation-centric cognitive loop look like in practice: sense, detect, interpret, reason, plan, act or ask or wait or observe, evaluate, learn?
Where should an explicit situation layer sit relative to models, tools, memory and the people in the loop?
Does an explicit situation layer actually improve timing, decision quality, coordination, adaptation and explanation, measured against a strong context-engineered baseline?
What result would falsify the hypothesis? If nothing would, it is not a hypothesis.