What is learning?
Learning is a lasting change produced by experience.
The change may appear in knowledge, skill, expectation or behaviour. A child learns a word. A driver learns to judge a narrow turn. A body learns to react more quickly to a familiar threat.
A temporary change caused by tiredness or mood is not usually learning.
Learning occurs when experience changes what a system can do next.
Is learning the same as memory?
No, though they depend upon one another.
Memory preserves information so that the past can affect the present. Learning is the process by which experience changes what is preserved and how it is used.
One can remember a telephone number briefly without learning anything general. One can also learn a motor skill without being able to describe every experience that shaped it.
Memory keeps traces. Learning reorganizes them into future ability.
How does learning begin?
Learning begins when a system detects a relationship.
A baby hears a repeated sound while seeing the same person. A dog notices that a particular sound is followed by food. A student discovers that dividing both sides of an equation preserves equality.
Repeated events alone are not enough. The learner must notice which events belong together.
Attention, timing and earlier knowledge all influence what is learnt.
What role does error play?
Error reveals a difference between expectation and reality.
Suppose you expect a door to open when pushed, but it does not. The failure directs attention towards the handle, the lock or the possibility that the door opens the other way.
A learner uses such differences to correct a model.
Without an expectation, an event may merely happen. With an expectation, surprise becomes information about what needs to change.
Does learning require repetition?
Often, but not always.
Repetition strengthens a relationship and shows whether it remains stable across occasions. Practice also makes some operations faster and less demanding of attention.
But a single event can be enough when its consequence is strong. Touching a hot surface once may teach a lasting lesson.
What matters is not repetition alone. It is how strongly the experience changes the model, memory or value attached to an action.
What is practice?
Practice is repeated action aimed at improving performance.
Useful practice contains feedback. A musician hears whether the note is correct. A bowler sees where the ball lands. A programmer observes whether the program behaves as intended.
Repeating the same mistake without noticing it can strengthen the mistake.
Good practice keeps the task difficult enough to reveal error, but clear enough for the learner to correct it.
What is understanding?
Understanding is learning the relationships that make knowledge transferable.
A student may memorize that a particular answer is 24. Understanding appears when the student can explain why and use the same principle in a new problem.
Memorization preserves an answer. Understanding builds a model that can produce answers.
Both have value. Facts must sometimes be remembered, but disconnected facts are difficult to use when the question changes.
What is generalization?
Generalization is the ability to apply what was learnt to a new case.
A child who learns the idea of a chair can recognize a chair never seen before. The child does not compare the object with a perfect stored picture. A broader pattern has been learnt.
Generalization requires ignoring some differences while preserving important ones.
Too little generalization makes every case seem unrelated. Too much treats important differences as if they did not matter.
Can we learn the wrong thing?
Yes.
Experience does not explain itself.
A person who becomes ill after eating one food may blame that food even when the cause was elsewhere. A child praised only for quick answers may learn that appearing certain matters more than thinking carefully.
Bias, coincidence and poor feedback can all produce mistaken learning.
What is learnt should therefore be tested against new cases and, when possible, compared with stronger evidence.
How do emotions affect learning?
Emotion assigns importance.
Fear can make a danger memorable. Curiosity can hold attention on a difficult question. Pleasure can strengthen an action that led to reward.
This is useful because a living system cannot remember every detail equally.
But emotional strength is not proof of truth. A vivid story may be remembered more easily than a quiet fact. Good learning respects emotion while checking the model it creates.
How do machines learn?
A machine-learning system changes an internal model using examples or feedback.
It may be shown correct answers, discover patterns without labels, or learn which actions lead to reward.
An algorithm determines how the model changes. Training data supplies experience. A measure of error or reward guides adjustment.
The system has learnt when this adjustment improves its behaviour on relevant new cases, not merely when it remembers its training examples.
Is machine learning like human learning?
In some respects.
Both can detect patterns, change through feedback and generalize beyond earlier examples.
But human learning is shaped by a living body, needs, emotion, social relationships, language and conscious experience. Present machine-learning systems are built and trained for more limited purposes, even when their outputs cover many subjects.
Similar performance does not prove an identical process.
Does learning guarantee improvement?
No.
Learning means change through experience, not necessarily change towards truth or goodness.
A system can learn a prejudice, a harmful habit or a successful deception. It can optimize the wrong measure.
Improvement requires a standard: more accurate prediction, greater skill, reduced harm or some other valued result.
The quality of learning depends upon the experience, the feedback and the goal.
How does learning support intelligence?
Intelligence cannot remain effective in a changing world without learning.
Learning corrects models, adds skills and changes expectations. Inference uses what has been learnt to reach new conclusions. Decision uses those conclusions to select action.
The action creates another experience, and the cycle continues.
An intelligent system is not one that never fails. It is one that can use failure to become less wrong.
So what is learning, finally?
Learning is how experience changes future possibility.
It preserves useful information, corrects models, strengthens skills and allows old lessons to guide new situations.
But experience can also teach badly. Learning becomes reliable when feedback is honest, errors can be examined and conclusions are tested beyond the examples that formed them.
To learn well is not merely to change. It is to change in closer contact with reality.
