Learning
Prediction Error
A prediction error is the difference between what you expected and what occurred. Making that difference visible turns surprise into a more useful learning question.
- Expect
- Act
- Observe
- Compare
- Adjust
An expectation and an observation are compared, and their difference informs the next attempt.
You expected them to step back. They went sideways.
A light jab was meant to move your partner backward. They step sideways instead. Something differed from your expectation. Before declaring the jab ineffective, ask what you predicted and what you actually saw. Prediction error concerns that mismatch, not simply the fact that you disliked the result. Even a successful exchange can surprise you, and that surprise may be worth examining.
Expect a response, then watch it
A kickboxer practises a light jab intended to draw a guard response before moving away. Before the attempt, they predict whether the partner will raise the hands, step back or hold position. If the partner instead steps sideways, the useful question is which cue was missed and whether the original expectation was too narrow. The drill stays controlled; surprise does not justify accelerating contact.
An inaccurate prediction can come from several places. The cue may have been read poorly, the action may not have conveyed the intended threat, or the partner may simply prefer another response. Repeat at a suitable pace with more than one partner. Do not replace one confident story with another after a single attempt. The value lies in a better calibrated expectation.
The research term is narrower than the training question
Prediction error means a mismatch between an expected consequence and an observed one. In motor control research, sensory prediction errors concern the sensory effects expected from a movement. They can help recalibrate movement when the body or environment behaves differently. Reward prediction errors concern whether an outcome is better or worse than expected. These are related ideas, but they are not interchangeable.
Shadmehr, Smith and Krakauer review sensory prediction and adaptation. Their evidence does not mean every lost round is a direct measure of a neural prediction error. Skill First also uses a conscious version of the idea: state what you expect an action to produce, then compare that expectation with observation. This is a practical application, rather than a claim that conscious reflection duplicates a particular brain mechanism.
Make comparison part of the session
Choose a low risk action with an observable consequence. Write a short prediction before trying it: the partner will move backward, this grip will limit this direction, or this foot placement will make turning easier. Afterward, describe what happened before explaining why. Video or a partner's observation may reveal that the action differed from what you thought you performed.
Change one relevant element and repeat. If prediction improves only with a familiar partner, test a modest new condition. Keep a note of surprising responses as well as successful ones. Over several sessions, ask whether you recognise the important cue earlier and recover more effectively when events differ. Predicting perfectly is not the goal; responding to uncertainty remains necessary.
Feedback supplies information about an attempt. Prediction error is the mismatch that information may reveal. Bayesian updating concerns revising beliefs with evidence; a feedback loop organises repeated adjustment. These connections are useful without collapsing the concepts. Sometimes an unexpected result reflects noise, not a lesson that warrants a major change. Observe carefully, adjust proportionately and preserve room for uncertainty.
Sources and further reading
- Error Correction, Sensory Prediction, and Adaptation in Motor Control
Review of sensory prediction errors in motor adaptation; conscious training examples are an application.
- Human sensorimotor learning: adaptation, skill, and beyond
Distinguishes adaptation from other motor learning processes.
See an idea alongside another
Related models
Feedback Loop
Use the mismatch to change the next attempt, not just comment on the last one.
Bayesian Updating
Unexpected evidence should adjust confidence without creating instant certainty.
Scientific Method
Make the prediction before you know how the exchange ends.
Perception Action Coupling
Your prediction relies on information you can use while acting.