Feedback in learning

From supermemo.guru
Jump to navigation Jump to search

This text is part of: "Problem of Schooling" by Piotr Wozniak (2017-2024)

Feedback in concept networks

The power of concept networks extends well beyond simple atomic declarative learning (as in spaced repetition). Complex tasks may involve procedural learning, pattern extraction, generalization, pattern completion, forgetting, etc. The learning process should ideally be under the exclusive control of the learn drive, which optimally seeks sources of the best feedback (see: Optimality of the learn drive). In practical terms, it means that all good learning should have a form of happy free learning. The process of reinforcement learning in a concept network is analogous to training neural networks except for the dynamics of the process that are determined by the biological properties of the network.

Optimum feedback in learning is provided by the learn drive system

TV tuning metaphor

The process of reinforcement learning in a concept network can be explained by the TV tuning metaphor:

Reinforcement learning in a concept network is similar to tuning an old analog TV set. The first picture may be hazy, grainy, colorless, and hard to decode. However, corrective feedback may help incrementally tune in to the right frequency, and achieve a crisp and clear picture.

Correct feedback leads to incremental learning. Feedback on demand controlled by the learn drive is optimum and efficient (see: Optimality of the learn drive). In contrast, coercive feedback at school may be equivalent to starting the whole tuning process from the starting position all over again. The intolerance for error leads to high interference (see: On the superiority of a rat over a schooled human)

Learning to speak

Learning to speak is an example of a complex learning task based on vast conceptualization that remolds the brain for the new job. The process is based on extracting knowledge from the environment by pattern recognition, generalization, pattern completion, etc. using feedback that provides directional reinforcement.

When generating speech, the semantic brain attempts to convert a concept map activation into a string of sounds that could be used to reproduce an analogous concept map in the brain of the listener. For that purpose, the brain needs to learn to generate speech. The process has a form of reinforcement learning in which the learn drive process provides the guidance as to the level of difficulty of undertaken tasks. The feedback from listeners provides reward in the form of comprehension, or mild penalty in the shape of misunderstanding.

If a child needs to get a ball, she may point to the ball, attempt to vocalize, and receive feedback from the world. The reinforcement feedback may be the ball itself, which generates the mild pleasure of confirmation: "My knowledge is ok. My goal has been accomplished". If the vocalization is wrong, e.g. "Pa" instead of "Ball", the supervising feedback may have a form of a correction. For example, an adult may demonstrate the correct pronunciation of "Ball" using baby talk that overemphasizes the characteristics of a sound pattern. This generates significant pleasure of learning: "I achieved the goal and learned how to achieve it better next time". The corrective feedback may also have a form of mind penalty due to a lack of understanding: "What are you saying?" or just "What?". Finally, the feedback may generate a significant penalty. For example, it may be an angry retort from a parent, "What are you babbling? Again you want something?", or from a teacher "How many times do I need to teach you the same thing, you lazybones!". The best learning (and pleasure) occurs when new useful things are learned. The worst learning comes with a rebuke, which may inhibit future attempts at generating speech. One of the main flaws of schooling is the lack of feedback or a penalizing feedback (in terms of bad grades, or even an angry teacher). When the vocalization of "Pa" receives a corrective feedback of "Ball", pattern extraction may occur and the network may attempt to vocalize "Pall" or "Ba" at the next attempt. It may also replace the old pattern with the new pattern and be able to generate an approximation of "Ball". The actual outcome will depend on (1) the stabilities of knowledge in the affected concept sub-network, and (2) the resulting proactive interference.

Learning to read

The generalization process may be disrupted by incorrect feedback, excessive penalization, interference, regress, toxic memories, etc. Learning to speak occurs best when it is self-directed, and when it happens in a secure natural environment. Analogously, learning to read occurs best when it is self-directed and self-paced. Coercive extrinsic feedback may lead to educational dyslexia. Reading wars waged in educational circles are a side effect of the battle for early reading and improved literacy scores (see: Reading wars are over). In reality, understanding the optimality of feedback in learning makes it clear that early instruction in reading can be harmful. Similarly, the whole problem of phonics vs. whole language is settled naturally by the learn drive that makes it possible to maximize benefits by finding the optimum balance between phonic serialization and pattern recognition in whole language context.

Learning languages

When a student struggles to speak in a new language, the mere ability to generate an incorrect sentence should be celebrated. Instead, the teacher may explain "Wrong!" and inhibit further attempts. This explains why people learn languages easily when living abroad, and often learn to fear speaking a language when they learn it at school.

Imagine a baby who says its first word: "ma" while pointing to its mother. In (nearly) all cultures, the moment is celebrated joyously as a breakthrough milestone. The child receives a jolt of pleasure that initiates a quest for more rewards in speech production. If the child was subject to the rules of coercive feedback of schooling, it might hear instead: "No no no! That's wrong. You should be saying "MO-THER". Repeat! "MO-THER". We all intuitively know that this kind of feedback makes no sense.

Correcting the grammar of a student at school, without student's specific request (or consent), is equally harmful and discouraging. It can build a wrong habit in the student and in the teacher. It can produce a form of selective mutism for life. I know many adults who would not utter a word of English in my presence. The anxiety is just too high. I have no expectations. It is the old bad school habit based on coercive feedback.

School and SuperMemo

The same rules of feedback extend to all form of early learning that relies heavily on the semantic brain. The adult-centric approach to supervised learning known from school (or SuperMemo) is suitable for stabilization of atomic declarative knowledge, and should also be self-directed and self-paced. The old rule of SuperMemo is never to review without building a coherent knowledge structure first (Do not memorize if you do not understand). In supervised learning, the rules of optimum feedback are systematically violated at school.

Erroneous extrinsic feedback is one of the origins of school hate and school refusal



For more texts on memory, learning, sleep, creativity, and problem solving, see Super Memory Guru