Is goal formulation the same as planning?

No. Planning searches for actions that achieve an already specified goal, although planning results can reveal formulation problems.

Can an agent revise its goal?

In some architectures, yes, but the conditions and authority for doing so should be part of the system design.

Does a detailed goal always help?

Useful detail removes ambiguity, but unnecessary restrictions can exclude acceptable solutions or complicate the task.

AI Glossary / Graphs and symbolic AI / Intermediate

Goal formulation

Definition

Goal formulation is the process of deciding what an agent should try to achieve and expressing that outcome clearly enough to guide reasoning. In classical problem-solving agents, it identifies the desired condition before the detailed search problem is constructed. It connects a broad purpose or performance measure with a specific target for the current situation.

What it means

A broad instruction such as “organize this delivery” leaves many possible interpretations. Goal formulation chooses which outcome matters now: where the package must end up, whether confirmation is required, and which additional conditions define completion.
It is distinct from problem formulation, which specifies the states, actions, transition model, and other details needed to search for a solution. The two influence each other, but they answer different questions. Goal formulation identifies the intended destination; problem formulation decides how the relevant world and possible routes are represented.

How it works

Interpret the task in its current context, identify the conditions that would count as success, and distinguish mandatory requirements from preferences. Express the goal at a level of detail that the agent can reason about. Check whether several acceptable outcomes should be represented rather than unnecessarily requiring one exact state.
Then connect the goal to an operational test and a model of what the agent can do. If a target is unreachable or ambiguous, its formulation may need revision or explicit qualification. A good formulation also keeps constraints on the process separate from final-state conditions when they must hold throughout execution.

Three practical examples

A navigation exercise turns a general travel purpose into a target location while treating travel time as a separate optimization criterion.
A document workflow specifies the required contents and review status that make a draft ready for its next stage.
A simulated robot task defines success as an object resting in a designated area, rather than merely having been touched or moved.

Think of it this way

Think of agreeing on what a completed assignment must contain before choosing how to work on it. This captures the translation from intent to a usable target. The analogy is limited because an AI agent's internal goal representation may omit context that a human colleague would infer, making explicit examples and boundaries especially valuable.

Worked example

Consider a fictional warehouse simulator with two available routes for a parcel. Route A reaches the requested station in four steps but leaves the parcel on the robot. Route B takes five steps and ends with the parcel placed in the station's tray.
Formulation 1 says only RobotAtStation. Route A satisfies it, and so could a route where the robot travels without the parcel. Formulation 2 says ParcelAtStation and ParcelInTray. Under the simulator's stated rules, Route B satisfies both conditions, while Route A does not.
Now add the preference to minimize steps among successful plans. That preference ranks only plans satisfying Formulation 2; it should not make a shorter incomplete route count as success. If the task also forbids entering a restricted aisle, that condition belongs in the permitted-action or trajectory constraints, not merely as an untested hope about the final state. The example uses invented routes and does not prescribe a real warehouse procedure.

Illustrative case study

In a fictional editorial project, the Alder team assigns an assistant to “finish the glossary entry.” The first completion rule checks whether text exists. Reviewers then identify missing examples and an unsupported definition.
The team formulates a more specific completion target using the required sections, source-review state, and editorial acceptance. It records which checks can be automated and which depend on a reviewer. The invented case illustrates how formulation shapes completion behavior. It does not imply that counting sections alone establishes writing quality or factual correctness.

Common misconceptions and mistakes

Goal formulation is not merely choosing the first action. A target expressed as a particular action sequence may rule out better ways to achieve the same intended outcome. Conversely, a goal that is too vague can admit unwanted results.
Do not treat every preference as an absolute requirement unless that is intended. Combining incompatible conditions can make the problem unsatisfiable. Also avoid assuming that a final-state condition enforces the whole process: arriving with an intact item says nothing by itself about which route was taken. The model must express process restrictions where they matter.

Business use

The concept is useful when translating a business request into an AI workflow or planning specification. It encourages teams to agree on completion evidence and the scope of acceptable outcomes before judging the agent's efficiency.
Use representative success and failure examples to test the formulation. Ask whether an agent could satisfy the written target while omitting an essential part of the request. Where quality cannot be fully captured by a simple predicate, include an appropriate review stage instead of pretending that a convenient proxy is the entire goal.

Check your understanding

Why should “minimize steps” not make Route A acceptable under Formulation 2?

Answer: Route A does not place the parcel in the tray, so it fails a required goal condition. The step-count preference compares successful plans; it does not erase the completion requirement.

Related terms

Goal
Problem formulation
Goal test
Performance measure
Objective
task specification
planning
constraint
acceptance criterion

Learn next

Learn next about problem formulation. Take the parcel goal and list only the state features needed to determine success and action applicability. Then identify one omitted feature that would become necessary if a new requirement were added, explaining how the model would need to change.

Sources

Russell and Norvig, original Solving Problems by Searching chapter, goal and problem formulation: https://aima.cs.berkeley.edu/4th-ed/pdfs/newchap03.pdf Russell and Norvig, original Intelligent Agents chapter: https://aima.cs.berkeley.edu/4th-ed/pdfs/newchap02.pdf

Previous: Goal-directed reasoning Next: Goal monitoring