A Research Program
Built Around the Situation.

What Is a Situation?

A situation is the meaning of events and state, interpreted from the perspective of an observer, relative to goals, responsibilities, relationships, time, uncertainty, and possible action.

An unpaid invoice is one event and one state. For the customer it may be a temporary cash constraint. For the salesperson it may be a relationship risk. For accounts receivable it is a collection responsibility. For the CFO it may signal deteriorating receivables. The facts can be shared while the situations differ.

Context Is Necessary. It Is Not Enough.

Context engineering determines what information an AI system receives: prompts, memories, documents, tools, histories, and current state. This is indispensable. But the presence of relevant information does not itself establish what is happening or what responsibility follows.

Context is material for cognition. Situation is its observer-relative organization into actionable meaning. Modern AI recognizes situations implicitly. Situational AI proposes making that recognition explicit.

Make Situation a First-Class Object

Language models operate over tokens. Reinforcement-learning systems often operate over states. Our hypothesis is that intelligent behavior needs a native unit that binds perception, goals, judgment, expectation, action, and outcome: the situation.

Situation is to intelligent behavior what the token is to language modeling. This is an analogy about computational role—not a claim that situations are literal text tokens.

A situation object may contain

• Observer, role, actors, and relationships
• Events, current state, and relevant history
• Goals, responsibilities, expectations, and constraints
• Temporal stage, urgency, uncertainty, risk, and consequence
• Available responses, selected response, outcome, and transition conditions

From Semantic Space to Situation Space

The proposal is not to replace embeddings. It is to change what gets embedded. Semantic embeddings capture similarity in language. A situation representation should capture similarity in underlying dynamics—even when the vocabulary, objects, and domains differ.

A patient disengaging from treatment and a prospect going silent during a sales process share little vocabulary. Yet both may express the same pattern: expected progression has been interrupted, timely intervention matters, and delay increases the probability of failure.

If a latent situation space can preserve such structure, experience may transfer across domains wherever the situational dynamics are analogous.

A Situation-Centric Cognitive Loop

Interpretation is essential. Detecting an event is not the same as recognizing the situation. A situation-centric architecture maintains an explicit hypothesis about what is occurring, revises that hypothesis as evidence changes, selects a response, and observes whether the response produced the expected transition.

Situation memory

Conventional memory remembers information. Situation memory remembers experience: what occurred, what was recognized, from whose perspective, what response was selected, why it was selected, what outcome followed, and whether the original interpretation was correct.

SCBCM—the Situation-Centric Bot Cognitive Model—is one proposed implementation of this larger thesis.

What We Need to Discover

Initial experimental comparisons

• Strong context-engineered agent versus situation-explicit agent
• Semantic retrieval versus situational retrieval
• Same event viewed by different observers
• Same vocabulary describing different situations
• Different domains expressing analogous situations
• Recognition under changing goals, roles, and time pressure

Five Explorations, Summer 2026

During the summer of 2026, five UT Dallas students worked with Mindspace AI to explore different questions emerging from Situational AI: its theoretical foundations, its relationship to existing models of situation awareness, whether explicit situation recognition can reduce computational context, and working applications built on situation-centric concepts. Each has a permanent page here, and all five are presented together at the October 22, 2026 forum.

Help Build—and Challenge—the Idea

Situational AI is an open research initiative. We invite researchers, engineers, cognitive scientists, designers, operators, and critics to contribute definitions, architectures, datasets, experiments, implementations, and counterarguments.