Before evaluating life phenomena, please understand the living environment.

Regarding knowledge, there is a very interesting phenomenon; that is, after people leave school, there appear two completely opposing voices. One voice says: After graduating for so many years, I have hardly used the knowledge taught in school, why did I waste time learning it at the beginning. Another voice says: The longer I graduate, the more I feel that the knowledge taught in school is very useful, and I regret why I didn't study harder at the beginning. Under this question on Zhihu, everyone is trying to persuade the supporters of the first voice to look further into the future.

But in fact, both voices are correct.

Main Text

1. Smart Horse

To explain why, let's go back to Germany in the early 1900s.

William Austin

The person in the photo is named William Austin, who was a math teacher. Generally, bearded math teachers are not very stable, and Austin felt that the intelligence of animals was underestimated, so he started to teach a horse named Hans to learn arithmetic.

Advertisement

Training the Horse

After a period of training, Austin brought Hans to the public to show off. When asked what is the sum of two plus two, Hans stamped four times with his hooves. And when asked "If today is Monday, how many days until Saturday," Hans could still stamp five times correctly with his hooves.Demonstrating Hans' Arithmetic Skills

Germans are a very meticulous (overfed) nation, and they have even established a committee to investigate this matter, led by another bearded man named Carl Stumpf.

Carl Stumpf

After a rigorous investigation, they did not find any fraud or cheating behavior in Austin. Even when someone else was used to ask questions on behalf of Austin, Hans could still answer correctly.

However, the problem arises when the questioner is out of Hans' sight or when the questioner himself does not know the answer, Hans is unable to respond.

2. Modeling Bias

Conclusion: In fact, Hans did not really learn arithmetic, but instead, he constructed another model: to judge when to stop tapping his hoof based on the subtle expressions and body movements of the questioner that are unconsciously revealed.

PS: This is a capability that all girls assume their boyfriends have, but in reality, their boyfriends do not have at all.Features: Although the micro-expression model constructed by Hans can perform the same function as the arithmetic model under the condition that the questioner is present, the two are not the same model.

Inspiration: The story was later called "Clever Hans." It has inspired people in the fields of psychology (observer expectancy effect) and machine learning.

2.1 Machine Learning Inspiration

Clever Hans Effect: In machine learning, this is used to illustrate that some intellectual tasks that artificial intelligence seems to be able to complete are actually dependent on certain subtle features found. Once the conditions change even slightly, artificial intelligence is very prone to failure.

A.I. Example: For instance, some models that identify tanks and airplanes are actually identifying whether the background contains a blue sky color. Therefore, people say that the machine has not truly understood the knowledge.

2.2 Educational Inspiration

Human Example: However, it is ironic that humans themselves also have the same problem; the models constructed by students can achieve the same function as the model that education wants to teach on paper, but the two are often not the same model.

Modeling Pattern: Because under these conditions, whether it is Hans or students who are promoted, they do not learn what you want to teach, but construct models by maximizing a certain reward (reinforcement learning pattern), but the model that can maximize the reward is not necessarily the model of educational goals.Pet Example: For Hans or pets, their purpose is to maximize the food (reward) given by their owner. Whether they have truly learned arithmetic or shaking hands has no impact on maximizing food. The "shaking hands" that pets learn is likely just a way to "press" the food out of your hand.

Student Example: For students who need to advance in their studies, their purpose is to maximize their scores (rewards). Although true knowledge can also lead to correct answers, if there is another model that can provide the correct answers in a faster way, even a model like "choose the longest one among three short ones, choose the shortest one among three long ones" that cannot solve practical problems, students will still choose the latter.

2.3 Educational Bias Phenomenon

Exam-oriented System: In fact, it is not just students who play the role of Hans. Teachers, parents, and training institutions in junior and senior high schools are also participants in this system, with the same goal of maximizing students' scores. This has gradually led to the emergence of another seemingly practical knowledge-teaching system, which is actually a system for maximizing scores (there is no right or wrong).

Tutoring Book Phenomenon: Have you ever wondered why junior and senior high school English textbooks have dialogues and listening exercises at the end of each lesson? After all, teachers never use those parts in class. Why not just choose a good tutoring book as the textbook, so you don't have to spend extra money on tutoring books?

Reward Difference: This is because the rewards that the experts who write textbooks want to maximize are different from the rewards that the experts who write tutoring books want to maximize.

Purpose of Tutoring Books: Tutoring books are materials written by experts in academic advancement to help students learn knowledge that can maximize scores.

Purpose of Textbooks: Textbooks are materials written by educational experts to help students learn knowledge that can solve practical problems.

English Example: For example, the design of "New Concept English" is very scientific, completely corresponding to the four models of listening, speaking, reading, and writing mentioned in "Learning View 07". It has prepared input and output exercise materials for each model, which corresponds to the dialogues and listening exercises after class.

Practical vs. Exam-oriented: However, for the purpose of maximizing scores, the knowledge in textbooks is not the most effective model, and is often replaced.Translation of the provided text into English:

Example: Many people left comments under "Learning Perspective 07," saying that the word learning method introduced in the video is too slow. They are actually making the evaluation with the purpose of "maximizing scores," and in fact, it is just as they say. However, when the purpose is "effective communication," what is introduced in Learning Perspective 07 is the most effective principle.

3. The phenomenon of modeling bias

Back to the "two voices" mentioned at the beginning.

3.1. The theory of knowledge uselessness

The first voice: The proponents of the first voice, who say "knowledge is useless," actually refer to the knowledge constructed for maximizing scores.

Analogy: Then it would be like Hans, who cannot answer questions without the questioner, and when these students leave school, the models they built in school often cannot be used to solve real-world problems either.

The reason for high scores but low ability: This is also one of the reasons why "high scores but low ability" occurs, not just what people say, "high IQ, low EQ."

The second voice: The proponents of the second voice, who say "knowledge is useful," refer to the knowledge that can solve real-world problems. These pieces of knowledge are the models accumulated by the brightest brains of humanity and are the foundation of human prosperity, the effectiveness of which naturally does not need to be questioned.Conclusion: The reason why both voices are correct is that they were originally describing different things.

3.2. Educational Mismatch in Employment

Cause: It is precisely because the "model that can maximize scores" is not necessarily the "model of educational goals" that puts students in an awkward situation when it comes to employment.

Screening Method: Companies (especially small businesses that lack resources and have to rely on the education system) rely on scores and the origin of schools to screen talents (recent graduates),

Screening Objective: On the other hand, they hope to screen talents who can solve practical problems in the workplace.

Awkward Situation: The mismatch between this screening method and the screening objective makes the students who have gone through great hardships to pass the exams (recent graduates) have to re-enroll in training to learn practical knowledge after entering the workplace.

4. Elimination of Modeling Bias

4.1. Modifying the Reward FunctionQuality Education Reform: Upon hearing this, there will inevitably be voices criticizing the college entrance examination (Gaokao), hoping to implement quality education reform. In fact, the Ministry of Education proposed the slogan of educational reform as early as 2001 and tried to promote quality education, but it did not take effect. Why is that?

System Evolution Direction: As mentioned earlier, students who are promoted do so by maximizing some kind of reward to construct a model. No matter how restricted, over time, the education system for promotion will eventually evolve into students learning the "model that can maximize the reward," rather than learning the "model of educational goals."

Analogy: It's like any online game that will eventually evolve into a single optimal way of playing and become boring (an inevitable trend).

Machine Learning Perspective: As long as the main purpose of the education system for promotion is to screen talents, then the issue of quality education reform can actually be seen as: whether it is possible to change the reward mechanism so that the "model that can maximize the reward" is equivalent to the "model of educational goals," or to make the "scoring goal" completely equivalent to the "ability goal."

Feasibility: Regarding this issue, it is not absolutely impossible, but we can refer to a joke in machine learning.

AI Analogy: If you design the reward function of a vacuum cleaner as "the more garbage it picks up, the more rewards it gets," then the likely outcome is that the vacuum cleaner learns to pick up garbage, learns to put it down, and then picks up and puts down in an infinite loop.

Conclusion: Therefore, when designing the reward mechanism, it is difficult to avoid such unexpected results.

4.2 How to Reform Education

Assumption: If the quality education is implemented, and its reward mechanism still cannot make the "scoring goal" equivalent to the "ability goal,"Prediction: Ultimately, the system will also evolve into another exam-oriented education system that deviates from the educational goals based on the reward mechanism of quality education.

Examination costs: The only difference is that this new education system will increase the cost for students to obtain scores, making talent selection no longer fair.

The left picture is the issue of educational equity, and the right picture is the fairness of online games.

Example: For instance, the score limit for the rich is 900, while the limit for the poor can only be 600 (scores are just a digitalization of admission conditions, and they may not really be numbers).

Analogy: This "quality education" is like an online game that claims everyone can play with divine equipment. But in reality, you must first recharge money to have a chance to defeat the BOSS. Then, the gap between players who can't afford to recharge and those who can will become increasingly larger.

American quality education: The best example is in the United States. The scoring mechanism of American quality education involves various talents and specialties (subjective judgment), and many channels for increasing scores are only available to the children of the rich.

The biggest person involved in the recruitment scandal: William Rick Singer.

Example: The "American college recruitment scandal" exposed in 2019 revealed how the rich in the United States used the scoring mechanism of "quality education" to give their children higher scores, thus entering prestigious universities.

Response ideas: If the current situation cannot guarantee the design of a reward mechanism that makes the "scoring goal" equivalent to the "ability goal," then the most reasonable solution is not to implement "quality education" that increases scoring conditions such as morality, physique, and art. Instead, it is to reduce the cost of students' scores in "exam-oriented education," allowing every student to have the opportunity to obtain all scores just by relying on past real questions. And this "exam-oriented education" is the college entrance examination.

The college entrance examination system: It mainly solves the fairness issue of talent selection, not the quality issue of talent education. In fact, everyone knows this point, just as everyone likes online games without a recharge system.4.3 Clarify Input and Output

Is there no way to eliminate modeling bias?

Education Issue: In the final analysis, the reason why the education system is so difficult to reform is that it is not primarily an issue of pedagogy, but rather a sociological issue.

Selection Purpose: As long as a system serves as a screening (promotion) function, individuals within the system cannot escape the behavior pattern of "maximizing rewards."

Non-selection Purpose: However, as long as this system is detached from the selection purpose, students can suppress modeling bias by clarifying input and output themselves.

Example: For instance, when the input is also fixed at the same time, Hans cannot use micro-expressions and can only use arithmetic to arrive at the result. This is why clarifying input and output is the first learning principle in "Learning View."

Practical Solution: But society also needs the function of talent screening, so it can only temporarily divide the education system into two parts: one is the compulsory education stage responsible for the selection purpose, and the other is the higher education stage responsible for practical purposes.

5. Forbidden Senior Year

Now we can answer a strange provision in "Learning View."5.1. Senior High School Education Phase

Question: Why can junior high school students and those in the first and second years of senior high school watch the "Learning Perspective," but not those in their third year, yet once the college entrance examination is over, they can watch it again?

Answer: This is because students in their third year of senior high school are in the stage of compulsory education that serves the purpose of screening, and it is inevitable (and should be done) to rely on learning exam-oriented knowledge to maximize their scores.

5.2. Learning Perspective Tends to be Practical

Scope of Application: Although the learning methods introduced in "Learning Perspective" are applicable to both exam-oriented knowledge and practical knowledge learning,

Current Value Bias: However, because the current "Learning Perspective" mainly emphasizes higher education for practical purposes, such as Learning Perspective 07, which is not prepared for exam-oriented students, I am very worried that some third-year senior high school students may develop the mentality of "exam-oriented knowledge is useless" after watching the current Learning Perspective, which could delay their lives. Therefore, it is prohibited for third-year senior high school students to watch.

Reason for Skipping Sections for Update: I also mentioned in my previous article that "Learning Perspective" will have an additional section in the future, specifically for the study of exam-oriented knowledge. This is also why there was a sudden update from issue 16 to issue 18.5 to introduce the difference between exam-oriented knowledge and practical knowledge.Conclusion

In fact, throughout "The View of Learning,"

I have always avoided discussing issues such as talent and systems.

Although we all know that there are gifted individuals,

although we all wonder what the college entrance examination (Gaokao) really screens for.

But these are not factors we can change.

After all, we cannot switch games like playing online games,

when we are dissatisfied with our qualifications or when the system is unreasonable.The only thing that can be done is to gather all the factors that one can change,

and select the most reasonable step at each stage.