Reflections on the Intelligent Age —— On Mechanical Thinking and Big Data Thinking

It took me about two days to finish reading Teacher Wu Jun's "Intelligent Times". There were a lot of dry goods and a lot of gains. I think the most important thing is the third chapter about the revolution of thinking. I will share it with you as an article after thinking about it.

I remember that I was a philosophy class in high school and learned something about mechanical thinking. I don't remember it clearly, only that it is mechanical, and I don't understand flexible thinking methods like machinery. After reading Wu Jun's book, I realized that mechanical thinking used to be a very advanced way of thinking, including Newton's three laws of kinematics and Einstein's theory of relativity, all of which can be said to be the result of mechanical thinking. Is it another breakthrough?

To a great extent, mechanical thinking originated in ancient Greece. The only reason why Europe can lead in science lies in the ability of speculative thinking and logical reasoning established in ancient Greece. In China, except in the Spring and Autumn Period, hundred schools of thought contended, and the main idea behind it was Confucianism and Taoism. Confucianism stresses the doctrine of the mean, neutrality, moderation, balance and people-oriented, while speculation is an extreme way of thinking and pursuit. Taoist thought pursues the nature of Taoism, the unity of man and nature, the balance of yin and yang, and the mystery of yin and yang; The pursuit of Tao is that Tao can be Tao, Tao is extraordinary, and name can be very famous, that is to say, the unchangeable Tao cannot be expressed in words. To sum up, in my opinion, our ancestors in China have long discovered that the world is very complicated. Confucianism cares more about people from the perspective of benevolence, while Taoism cares more about the harmony between man and nature, taking heaven and earth as teachers, learning from heaven, learning from land and learning from nature.

In his book, Mr. Wu summarizes mechanical thinking as follows:

In my opinion, most people may have this mode of thinking, and there is another mode of thinking that does not exist at all, that is, doing things blindly, which is not as good as mechanical thinking. The core idea of mechanical thinking can be summarized as follows: the world can be fully understood, there are certain laws, there is a certain causal relationship between everything in the world, and the search for laws requires "bold assumptions and careful verification", which is now considered as a scientific attitude and scientific way of thinking. Many achievements in modern times, including Einstein's mass-energy equation, were completed in the mode of mechanical thinking.

The positive side of mechanical thinking that the world is certain and can be recognized also has its limitations, that is, denying uncertainty and unknowability. Einstein famously said, "God doesn't roll the dice". This is what Einstein said when arguing with Bohr, the inventor of quantum mechanics. Today we know that Bohr was right in this argument, Einstein was wrong, and God rolled the dice.

From the expert prediction of the stock market to the weather forecast, to the simple roll of dice, many predictions are inaccurate. Where does the uncertainty come from:

First, the more detailed we know, the more variables that affect the world, and the results cannot be calculated by simple methods or formulas. If all the variables that affect the transaction are exhausted, it is of course predictable, but it is impossible in reality, and the formula will be very complicated and cannot be described clearly with simple formulas;

Secondly, due to the influence of our subjective thinking, our understanding of the world is inaccurate because of the limitation of our way of thinking, which also hinders our understanding of the world.

Third, objectively, the world is complicated. For example, we know that electrons rotate rapidly around the nucleus, but the position and speed of electrons at a specific moment are uncertain, which is the characteristic of atoms themselves, similar to the uncertainty principle in quantum, and our measurement activities themselves affect the measurement results. This is similar to buying stocks according to a certain index in the stock market. If everyone buys stocks according to this index, then these behaviors of buying stocks themselves will affect the price and trend of the stocks themselves.

The world is uncertain, but not unknown. It is also the movement of electrons around the nucleus. Although we don't know the specific position and speed of electrons, we can estimate the position and probability of their appearance. So many things in the world that are difficult to be described by specific formulas can usually be described by probability models. Shannon, a genius, linked the uncertainty of the world with information and formed the information theory, which is not only a communication theory, but also a new way for people to know the world.

Information theory first solves the problem of how much information there is. The two simple sentences "Tomorrow the sun will rise in the east" and "xxx star and xxx star have been secretly registered to get married" are very informative. From our intuitive thinking, the sun will rise in the east tomorrow, which is almost certain, so it is equivalent to nonsense with no information content; The probability of the latter sentence is relatively low, so it contains a lot of information. Dr Shannon linked information with the uncertainty of event determination and introduced the concept of entropy. Entropy was originally a concept in thermodynamics. The two containers are separated by baffles, and the gases on both sides are in an orderly state. If the baffle is removed, the gas state will become more and more disorderly and tend to be stable in macro. This gas has gradually changed from the original ordered state to the disordered state, and its entropy has been increasing, that is to say, the entropy determined by order is low, and the more chaotic it is, the higher the entropy is. If entropy is to be low, or things are to be orderly, there must be external forces.

Still a little abstract. For a simple example, if you don't tidy your room, it will become more and more messy in the future. At this time, the entropy is getting bigger and bigger. Looking for it for a long time may not find it. Why? Because chaos leads to the increase of uncertainty, how to deal with it is to tidy up the room, improve the order and reduce the entropy.

Dr Shannon uses entropy to measure the amount of information. The greater the amount of information, the greater the uncertainty, so the greater the entropy. If you want information to be certain, you must introduce more information. The amount of information to be introduced depends on the uncertainty of the required event. Contrary to mechanical thinking, information theory is based on uncertainty.

If you have studied machine learning, you must be familiar with the decision tree algorithm. It doesn't matter if you haven't studied it. Let's use an example to briefly explain how to judge whether a watermelon is a mature watermelon. We need to judge the color depth of melon pattern, the thickness of melon pedicle and the sound of knocking on melon. We can judge step by step according to these conditions, choose one condition for each step, and finally judge whether the melon is ripe or not according to multiple conditions, as shown in the following figure:

First of all, we are not sure whether a melon is ripe or not. How to determine, we introduced a lot of information, such as the color of melon pattern, the thickness of pedicel, and the order of judging the introduction conditions is also very important. For example, I think the depth of melon pattern color is very important, and this information is relatively large. After the introduction, the information entropy of ripe melons will decrease rapidly, which makes us more and more sure about the information. This principle is used to judge decision trees. The color of the pattern and the thickness of the pedicel are not randomly selected, and the information to be selected is related to whether the melon is mature or not (according to information theory). Mutual information in information theory explains the size of correlation.

The important principle of information theory is that when we want to find a probability model for unknown events, this model should satisfy our existing data, but make no assumptions about the unknown. This is the principle of maximum entropy, which is different from the previous "bold assumption and careful argument". The premise of not making subjective assumptions is that there is enough data.

First of all, we should understand three characteristics of big data: 1) The data volume should be large enough; 2) The dimension of data should be sufficient; 3) If the data is complete and fully covered, sampling is not allowed.

The essence of this world is an uncertain world. The more information we know, the easier it is to eliminate uncertainty. With the development of big data, many artificial intelligence problems can be solved because our data is large enough.

If the amount of data is large enough, we will have enough information. The more the uncertainty in related fields is reduced, the faster the related research progress will be. The more dimensions of data, the better the correlation matching with the problem we want to solve. With multi-dimensional information, cross-validation can be done, thus further reducing the uncertainty of information; The integrity of data prevents the occurrence of small probability events, which covers the whole range of the environment where the events occur. Because of the progress of technology, the completeness of data collection can be called possible.

Big data emphasizes relevance, not causality. Since the world is uncertain, we can't find its causal relationship in some laws, but it doesn't prevent us from finding its relevance, such as placing snacks advertisements on movie rental websites, such as placing credit card advertisements and mortgage advertisements on coffee review and sales websites. This is the result of clicking on the big data sharing advertisement, although we don't know the causal relationship. But this correlation also helps us to improve the click-through rate of advertisements. We should learn to accept this answer without knowing the reason. If we are willing to accept it, we will jump out of the way of mechanical thinking and only pursue causality.

The era of big data has arrived. Our way of thinking should stop thinking only in a familiar mechanical way and dare to accept answers without cause and effect.

I wish everyone can improve their thinking.

-Yi Ming was in Chengdu on August 3 1 2009.