What Is SOAR Cognitive Architecture?

How does a system decide what to do next?

That is the question SOAR puts at the center.

SOAR is a cognitive architecture. Allen Newell, John E. Laird, Paul S. Rosenbloom, and the Soar research community shaped its early development. The current Soar project and manual are maintained through the Soar group at the University of Michigan.

SOAR is not mainly a memory model. It is a model of goal-directed problem solving.

If you are coming from ACT-R, this is the clean difference to keep in mind. ACT-R is especially useful for understanding memory, activation, and production rules. SOAR is especially useful for understanding state, operators, impasses, and problem solving.

The simple idea is this:

state -> choose an operator -> apply the operator -> new state

A state is the current situation. An operator is a possible action that can change the state. A goal is the outcome the system is trying to reach.

This sounds plain, but it gives SOAR its power. A system does not have to solve the whole problem at once. It can keep asking a smaller question:

What operator should apply now?

The core loop

SOAR works by repeatedly selecting and applying operators.

Imagine an assistant helping Susan with a support issue. The current state might include the user, the task, the last tool result, and the open goal. Several operators may fit:

• ask for more information
• check account status
• retry the tool
• escalate to a human

SOAR would not treat those as loose suggestions. It would represent them as candidate operators. Then a decision procedure chooses one.

That choice changes the state. The loop continues.

Why SOAR matters for AI

AI builders often talk about agents as if an agent is a language model plus tools.

SOAR gives a better question:

How is the next step chosen?

The answer cannot be "the prompt decides" forever. A useful agent needs a working state, candidate actions, rules for choosing among them, and a way to learn when problem solving gets stuck.

SOAR gives names to those parts.

It does not give an AI product everything it needs. It does not decide permissions, privacy, human review, or product boundaries. But it helps separate the problem.

What to remember

SOAR is useful because it treats intelligent behavior as a cycle of state, operator selection, and action.

It is not a shortcut to human-level intelligence. It is a clean way to think about problem solving.

For AI memory, that matters. Memory should not only store the past. It should help the system choose what to do next.

Sources

• Soar homepage, University of Michigan
• Soar User's Manual
• The Soar Architecture
• Laird, Newell, and Rosenbloom, SOAR: An Architecture for General Intelligence

Next, continue with Working Memory in SOAR, Operators in SOAR, or return to the Concepts hub.