Uncertainty

You hear a crash from another room.

The sound tells you something. Something happened. A glass may have fallen, a door may have slammed, or an object may have been knocked from a shelf. Some explanations fit what you heard better than others, and many possibilities can be ruled out immediately.

But the sound alone does not tell you exactly what happened.

You have relevant information without having certainty.

Many ordinary problems of perception, interpretation and action have this structure. Sensory signals can be noisy. Important causes can be hidden. Situations can support more than one interpretation. Future events have not happened yet. Yet organisms still have to perceive what is around them, form working interpretations and act.

Uncertainty is therefore not merely a temporary failure of cognition. It is one of the normal conditions under which cognition operates.

What Uncertainty Means

Uncertainty is an informational condition in which the evidence currently available does not uniquely determine the relevant state, cause, interpretation or future outcome.

Put more simply, more than one possibility remains compatible with what is currently known.

Those possibilities do not have to be equally plausible.

You may be uncertain about a present state: What am I seeing through the fog?

You may be uncertain about a hidden cause: What produced the sound I just heard?

Or you may be uncertain about a future outcome: What will that approaching vehicle do next?

In each case, substantial information may already be available. The problem is not necessarily that nothing is known. It is that what is known does not settle everything relevant to the question.

This is why uncertainty should not be treated as another word for ignorance. There are situations in which little relevant information is available, but there are also situations in which evidence rules out many possibilities, strongly supports others and still leaves something unresolved.

Science also uses the language of uncertainty at several related levels. Researchers may describe uncertainty in the information available to an observer, represent uncertainty within a formal model, or investigate whether an organism behaves differently when information becomes more or less reliable. These are related questions, but they should not automatically be treated as descriptions of the same thing.

The central concept here is simpler: evidence can be informative without uniquely determining what is true.

Uncertain Does Not Mean Unconstrained

This distinction is especially important once perception is understood as constructive.

As established in Perception Is Constructive, perception is not a passive recording of the world. Sensory information has to be organised and interpreted before it can support a usable representation of what is happening.

But construction does not mean invention without constraint.

Suppose you see a dark shape through thick fog.

The available visual information may be poor enough that several interpretations remain possible. Perhaps it is a person, a signpost or part of a tree moving in the wind.

Yet the visual signal still matters.

The shape's apparent size, location and movement constrain what it could reasonably be. As you move closer and more information becomes available, some interpretations may become less plausible while another becomes better supported.

The fact that the evidence does not immediately determine one answer does not make all answers equally defensible.

Uncertainty and constraint can exist at the same time.

Evidence can rule possibilities out without identifying one possibility with certainty. It can favour some interpretations over others. It can narrow the range of reasonable conclusions while still leaving more than one open.

Controlled perceptual experiments demonstrate versions of this principle. When researchers vary the reliability of visual, auditory or tactile information, people can adjust how strongly different sources influence their judgements. A less reliable signal can exert less influence than a more reliable one.

Experiments involving auditory and visual information likewise show that people can remain uncertain about whether two signals came from a common event while still using their spatial and sensory properties to constrain their judgement.

These experiments do not establish that human inference is always optimal, nor that every perceptual problem is solved in the same way. They establish a narrower and more important point for this concept: behaviour can remain sensitive to the structure and reliability of evidence even when that evidence does not determine one certain interpretation.

Constructive perception under uncertainty is therefore not unconstrained guessing.

Uncertainty Can Change

Uncertainty is not simply present or absent. The informational situation can change.

Return to the crash in the next room.

At first, you know little beyond the fact that something happened. Then someone says, "I dropped a plate." The range of plausible explanations narrows considerably. You enter the room and see broken ceramic on the floor, strengthening that interpretation further.

Other questions might remain unresolved. Did the plate simply slip? Was someone carrying too many things? Was it already damaged?

Additional information can therefore reduce one uncertainty without eliminating every uncertainty surrounding an event.

New evidence can also change the structure of what remains possible. It may rule out alternatives, make one interpretation better supported, weaken another, or reveal a possibility that had not previously been considered.

Researchers often use mathematical quantities to represent particular aspects of uncertainty and changes in informational reliability. Depending on the problem and the model, these representations may involve probabilities, variance or other formal quantities.

Such tools can be extremely useful.

But, as established in Scientific Models Are Tools, a representation should not automatically be confused with the phenomenon being represented.

A model may use a probability distribution to characterise uncertainty. That does not establish that uncertainty universally is a probability distribution, nor that a nervous system must literally represent uncertainty in exactly the same form.

Formal descriptions tell us how a particular model represents a problem. Establishing how an organism actually processes uncertain information is a further empirical question.

Ambiguity Is One Source of Uncertainty

Ambiguity and uncertainty are closely related, but they are not identical.

Information is ambiguous when it supports more than one relevant interpretation. An indistinct figure in the distance may be interpreted in several ways. A sentence may have more than one plausible meaning.

Ambiguity can therefore produce uncertainty.

But uncertainty can arise for other reasons. Information may be noisy or incomplete. A relevant cause may be hidden. Or the question may concern a future event for which several outcomes remain possible.

Different informational problems can therefore leave different things unresolved.

More observation may supply missing information. Better measurement may reduce noise. Additional context may distinguish competing interpretations. Some uncertainty about future events may remain even after every presently available source of information has been used.

Ambiguity is one route to uncertainty, not a synonym for it.

Feeling Certain Is Not the Same as Being Certain

There is another distinction that matters.

A person can feel certain even when the available evidence leaves relevant possibilities unresolved.

Subjective confidence describes how sure someone is about a judgement. Informational uncertainty concerns what the available evidence does and does not determine.

The two can be related. Clearer or more reliable evidence can increase both accuracy and confidence. But the relationship is imperfect.

People can be highly confident and wrong. They can also recognise substantial remaining uncertainty while correctly identifying one interpretation as better supported than its alternatives.

Research on metacognition—the ability to evaluate one's own judgements—illustrates this distinction. In controlled perceptual tasks, people can perform similarly while differing in how accurately their confidence distinguishes their correct judgements from their errors.

Confidence therefore contains information that can be useful without becoming a guarantee of correctness.

Feeling certain does not transform incomplete evidence into certainty. Conversely, feeling uncertain does not necessarily mean that the available evidence supports nothing useful.

A person may recognise that one conclusion is strongly supported while also acknowledging that another possibility cannot yet be ruled out.

Recognising residual uncertainty can itself be an accurate representation of the evidential situation.

When the Evidence Changes

Working interpretations formed under uncertainty sometimes need to change.

Suppose the crash from the next room initially seems most likely to have been a dropped object.

That may be well supported by what you heard.

Then someone calls out that a shelf has collapsed.

The new information changes what the evidence supports.

Revising your interpretation does not necessarily show that the earlier interpretation was poorly formed. The earlier judgement may have been reasonable given the information then available.

Research on perceptual decision-making shows that revision can occur even after an initial choice has been made. When further evidence becomes available, people can sometimes change an earlier perceptual decision rather than treating it as final.

This does not mean that cognition always updates appropriately. People can ignore relevant evidence, place too much weight on weak evidence or interpret new information badly.

The narrower principle is that an interpretation formed under uncertainty can remain provisional. Additional evidence may strengthen it, weaken it or justify replacing it.

Whether an interpretation was well supported therefore depends partly on the evidence available at the time.

Uncertainty makes revision possible without making every conclusion disposable.

Action Under Uncertainty

Uncertainty does not disappear simply because action becomes necessary.

Consider driving toward an intersection.

You can observe another vehicle's position, speed and direction. Traffic rules and road conditions provide additional information. Yet you cannot know with certainty what the other driver will do next.

Driving nevertheless requires continuous action.

The same basic problem appears throughout sensorimotor behaviour. Catching a moving object requires acting on information about its trajectory before its future position is known with certainty. Moving through a changing environment requires responding while relevant information is still unfolding.

Controlled studies of perceptual decision-making show the same general principle in simplified settings: people can reach decisions while sensory evidence remains noisy or incomplete.

This does not mean that uncertainty should be ignored.

Sometimes uncertainty provides a reason to wait, inspect, measure or seek additional information. In other situations, delaying carries costs or removes opportunities to act.

The relevant point is that effective action does not require complete certainty.

Organisms routinely have to act on the information currently available while remaining responsive to what happens next.

Models of Uncertain Inference

Because uncertainty is fundamental to many problems of perception and action, researchers have developed formal frameworks for describing inference when information is incomplete or unreliable.

Bayesian models are one important example. They provide mathematical ways of representing how existing information and new evidence can be combined under uncertainty. Behaviour in some controlled perceptual and sensorimotor tasks has been well described by models of this kind.

That does not establish that all human cognition is Bayesian, nor that a good fit between behaviour and a model uniquely identifies the cognitive or neural mechanism that produced the behaviour.

Predictive-processing theories, introduced in Predictive Processing, are another influential family of approaches to perception and inference. They attempt to explain aspects of cognition through relationships among predictions, incoming information and the revision of internal models.

But uncertainty does not depend conceptually on predictive processing.

We can establish that organisms encounter incomplete and ambiguous evidence, respond to differences in informational reliability, revise interpretations and act without certainty without assuming that predictive processing—or any specific probabilistic framework—is the uniquely correct account of cognition.

Nor is uncertainty the same as prediction error.

Within predictive frameworks, prediction error concerns a discrepancy between a prediction and incoming information. Uncertainty concerns what the available information leaves unresolved. The two concepts can interact without being interchangeable.

Likewise, a formal model of uncertain inference is not automatically a literal description of neural implementation.

Questions about what predictive-processing frameworks genuinely explain, what kinds of uncertainty they represent and where their explanatory claims may exceed the available evidence belong to Predictive Processing: Scope and Limits.

Certainty Is Not the Requirement

Return once more to the crash in the next room.

You heard something. The sound constrained what might have happened. It did not tell you everything.

That is not an exceptional failure of perception. It reflects an ordinary feature of encountering a world in which information arrives partially, relevant causes can remain hidden and future events are not fully determined from the evidence currently available.

Human cognition therefore does not operate only after uncertainty has disappeared.

It forms working interpretations from incomplete evidence. It distinguishes among possibilities even when several remain open. Confidence can rise or fall without becoming equivalent to certainty. Interpretations can change when the evidence changes. Action can proceed while relevant uncertainty remains.

None of this means that evidence is powerless.

The existence of uncertainty does not make every interpretation equally defensible, and it does not mean that nothing is known.

Evidence can constrain without completely determining.

That is the central condition uncertainty names: not an absence of information, and not unconstrained guessing, but the ordinary situation in which what is available tells us something without settling everything.