Does qualitative physics prohibit numbers?

No. Qualitative and quantitative information can be combined; the distinctive feature is reasoning with meaningful qualitative abstractions.

Can it predict an exact time?

Usually not from qualitative information alone. Timing requires additional constraints or numerical data.

Is an ambiguous answer a failure?

Not necessarily. It may accurately show that the available information permits several behaviors and identify which comparison or observation would resolve them.

AI Glossary / Graphs and symbolic AI / Intermediate

Qualitative physics

Definition

Qualitative physics is an area of AI concerned with representing and reasoning about physical systems using meaningful qualitative distinctions, such as increasing versus decreasing, positive versus negative, and above versus below a relevant threshold. It aims to explain possible physical behavior without requiring an exact numerical value for every parameter and initial condition.

What it means

People can often reason that a container's level will rise when more liquid enters than leaves, even without calculating the precise rate of rise. Qualitative physics seeks explicit representations and reasoning methods for such knowledge.
Its categories are chosen because they matter to behavior. A level reaching a container's rim can start overflow; a difference changing sign can reverse a flow direction. These landmarks are more informative than arbitrarily rounding every number to fewer digits. The approach can complement equations and numerical simulation. It is not a claim that quantitative physics is unnecessary or that a coarse description always determines one unique future.

How it works

Identify the objects, quantities, processes, and constraints relevant to the question. Represent magnitudes relative to landmarks and represent changes with signs or other qualitative relationships. Propagate the consequences of physical constraints to determine possible states and transitions.
If available information does not settle a comparison, retain multiple possible behaviors or seek more detail. For example, knowing that inflow and outflow are both positive does not determine whether a tank fills or empties. Their relative magnitudes matter. A qualitative simulation can therefore produce a branching set of possible evolutions rather than a single numerical trajectory with exact timestamps.

Three practical examples

A teaching model can explain why a tank level rises, falls, or stays steady under different flow relationships.
A troubleshooting prototype can compare observed trends with qualitatively expected behavior to identify model-consistency questions.
A robot-planning simulation can reason about whether a modeled pouring action may reach a spill threshold before requesting a more detailed calculation.

Think of it this way

Think of a map that marks uphill, downhill, and important crossings without listing every elevation to the millimeter. It supports some useful judgments while leaving others unresolved. The analogy is limited because qualitative physics must preserve the physical relationships relevant to the task; simply discarding numbers without a principled representation can produce misleading predictions.

Worked example

Consider an invented rigid tank with liquid volume V, inflow rate q_in, and outflow rate q_out. Assume no leakage, evaporation, or other volume-changing process, and consider the tank before it is empty or overflowing. Conservation gives dV/dt = q_in − q_out.
If q_in is greater than q_out, V is increasing. If the rates are equal, V is steady. If q_in is smaller, V is decreasing. These conclusions need the ordering of the rates, not their exact values. Merely knowing that both rates are positive leaves all three behaviors possible.
Add a qualitative landmark V_max for the rim. If the volume is below V_max and increasing, reaching the rim is a possible future event; to assert that it will reach it, the model needs the increasing trend to persist long enough. Exact arrival time is not determined by the qualitative description alone.
For a numerical cross-check, suppose V starts at 7 liters, V_max is 10 liters, and constant rates are 4 and 3 liters per minute. Net inflow is 1 liter per minute, so the rim is reached after 3 minutes. That timing uses extra quantitative and constancy assumptions, beyond the earlier qualitative conclusion.

Illustrative case study

Fictional case: Brook Learning builds an interactive tank lesson. An early explanation says that an open inlet always makes the water rise. A reviewer adds an open outlet and shows that a larger outflow makes the level fall instead.
The lesson is revised to compare the two rates and to report uncertainty when their ordering is unknown. It also distinguishes “could reach the rim” from a timed forecast. This fictional case illustrates the importance of physical constraints and explicit missing information. It does not report a real engineering deployment or a validated control system.

Common misconceptions and mistakes

Do not equate qualitative reasoning with vague intuition. The representations and inference rules can be precise even when their values are non-numerical. Do not assume two positive influences can be combined without considering opposing effects and constraints.
Coarse abstraction can admit behaviors that are not realizable in a more detailed model, so an apparent possibility may need refinement. Conversely, omitting a relevant process can hide a real behavior. “Increasing” does not by itself specify how fast, for how long, or whether a threshold will be crossed in finite time. A qualitative explanation also remains conditional on the chosen physical model and its scope.

Business use

Qualitative physics can help produce understandable explanations where exact parameter values are unavailable or unnecessary for the question. It can organize early analysis, support training, and identify when additional measurement or numerical modeling would change the answer.
For a business prototype, define the distinctions the user needs: direction of change, possible threshold crossing, or compatibility with an observation. Check simple cases against known equations and retain explicit “undetermined” outcomes. Avoid presenting every branch of a qualitative simulation as a forecast with equal probability. The representation describes possibilities under constraints, not necessarily a probability distribution over outcomes.

Check your understanding

If both inflow and outflow are known to be positive, can the model conclude that the tank level rises?

Answer: No. It must compare their magnitudes. Greater inflow produces a rise, equal rates produce a steady volume, and greater outflow produces a fall under the stated conservation model.

Related terms

Landmarks
Abstraction
Qualitative reasoning
qualitative simulation
physical model
landmark
sign algebra
causal explanation
conservation law
numerical simulation

Learn next

Learn qualitative simulation and model refinement next. Change the tank so outflow increases with liquid level. Explain why the original constant-rate timing calculation no longer applies, and identify the additional relationship needed to reason about an eventual steady level.

Sources

Kenneth Forbus, Qualitative Physics: Past, Present, and Future, original author chapter: https://www.qrg.northwestern.edu/papers/Files/Q3PF.PDF

Previous: Qualitative data Next: Qualitative probabilistic networks