Does a conditional effect make an action nondeterministic?

Not by itself. Known starting conditions can determine exactly which effects occur in a deterministic model.

What happens when an effect condition is false?

That effect is inactive. The affected fact may persist or be changed by another applicable effect; its opposite is not automatically asserted.

Can conditional effects be represented without special syntax?

Sometimes the cases can be split into separate actions with distinct preconditions, although the resulting representation may become larger.

AI Glossary / Graphs and symbolic AI / Intermediate

Conditional effect

Definition

A conditional effect is an outcome of an action that occurs only when an additional condition holds. In a planning model, the action may be available in several states while producing different changes depending on the starting state. The condition governing an effect is distinct from the precondition governing whether the action can be taken at all.

What it means

An action description can contain effects that always occur when the action executes and other effects that depend on particular facts. This lets one action capture several related situations without duplicating its entire description.

For example, a modeled processing action could always mark a record as processed but mark it as priority-handled only if it already carries a priority flag. The extra flag determines one result; it need not determine whether processing is allowed.

Conditional does not mean random. If the starting state is fully known and the action model is deterministic, the active effects can be determined exactly. Uncertainty arises when the relevant facts or the model itself are uncertain, not simply because an effect has a condition.

How it works

In a standard instantaneous deterministic planning interpretation, first check the action's precondition. If it holds, evaluate the conditions of its effects against the state immediately before execution. Collect the unconditional effects and the conditional effects whose conditions are true, then apply the resulting changes together.

This before-state evaluation matters. An unconditional effect that makes a condition true does not automatically activate another conditional effect within the same action. A different interpretation would need to be specified explicitly.

Facts not changed by the applicable effects ordinarily persist under the model's frame convention. Conflicting active effects need a defined treatment or should be excluded by a well-formed domain description.

Basic STRIPS descriptions lack direct conditional effects, but a finite set of relevant cases can sometimes be represented by separate actions with different preconditions. Richer representations express these cases within a single action description.

Three practical examples

1. A planning model represents a packaging action that adds a protective insert only for an item already marked fragile, while sealing every eligible package.

2. A game action opens a passage only when a switch was active before the action, but consumes a turn regardless of that switch's value.

3. A workflow simulator records completion for every processed item and adds a review marker only when the item's starting state meets a specified exception condition.

Think of it this way

Think of a conditional effect as an optional attachment in an instruction: carry out the main change, and include this extra change if the stated situation applies.

The analogy does not mean the executing system freely chooses the attachment. Under a deterministic planning model, the specified condition determines whether that effect applies. Ordinary human instructions may leave timing and interpretation less precise than the formal model requires.

Worked example

Consider a toy action called Prepare. Its precondition is Ready. Its unconditional effect sets Prepared to true. It also has a conditional effect: if Tagged was true before the action, set Routed to true. Another unconditional effect sets Tagged to true.

Start with Ready = true, Prepared = false, Tagged = false, and Routed = false. Prepare is applicable because Ready holds. The Tagged condition is false in the starting state, so the routing effect is inactive.

After execution, Prepared and Tagged are true, while Routed remains false. The action's new Tagged value does not retroactively activate its routing effect.

Now start instead with Tagged = true and the other values unchanged. The same action makes Prepared true and Routed true. This comparison isolates the effect condition while keeping the action's applicability unchanged.

If Prepare is allowed again in the first resulting state, a second execution can activate routing. Whether repetition is permitted depends on the complete action model.

Illustrative case study

Fictional case: Hazel Workflow builds a planning demonstration with the Prepare action. A developer updates flags sequentially and then tests the routing condition, causing the first execution to route an initially untagged record.

The modeling team identifies the mismatch: its specified semantics evaluate conditions in the state before the action. It changes the simulator to collect applicable effects before updating the state and adds the two starting-state examples to its review material.

The case illustrates a semantic implementation error. It makes no claim about a real workflow platform's behavior or measured business results.

Common misconceptions and mistakes

Do not move every effect condition into the action precondition. That would prevent the whole action in states where it should still produce its other effects.

Do not treat a false condition as evidence that the effect's opposite becomes true. An inactive effect usually makes no change; another effect or persistence convention determines the resulting value.

Do not confuse conditional effects with conditional plans. An effect belongs to an action model. A conditional plan chooses later actions according to available observations or conditions.

Business use

For business teams reviewing planning-based automation, conditional effects make outcome assumptions visible. They help explain why the same operation can produce different modeled results for records or objects in different starting states.

Reviewers should distinguish eligibility requirements from outcome modifiers and ask when each condition is evaluated. Small examples covering true and false conditions can reveal misunderstandings before a larger plan is trusted. If the real process has uncertain outcomes, the model needs an appropriate uncertainty representation rather than a deterministic conditional effect alone.

Check your understanding

An action is applicable, but one conditional effect's condition is false. Must the entire action be rejected?

Answer: No. Applicability comes from the action precondition. The false effect condition prevents that particular effect, while other applicable effects can still occur.

Related terms

Precondition
Conditional plan
Concurrent action list
Action model
STRIPS
state transition
frame assumption
deterministic planning

Learn next

Compare an action with conditional effects against equivalent separate actions for a small finite domain. Then examine nondeterministic and probabilistic effects to see how unknown outcomes differ from known state-dependent changes.

Sources

Poole and Mackworth, Artificial Intelligence: Foundations of Computational Agents, Representing States, Actions, and Goals: https://artint.info/3e/html/ArtInt3e.Ch6.S1.html . The primary textbook treatment of state-dependent effects and STRIPS representations was consulted. The Prepare calculation and fictional implementation case are original illustrations with explicit before-state semantics.

Previous: Conditional distribution Next: Conditional entropy