From mechanical thinking to big data thinking, they are neither mutually exclusive nor adjacent. If summarized, mechanical thinking belongs to the decimal world, big data belongs to the large number world, and the middle is the median world that we often face in our daily life.
First, mechanical thinking and the decimal world
In today's view, "mechanical thinking" seems to be a lagging and inflexible symbol, and even "machinery" itself has become a description of someone, but it is by no means a compliment. However, in the17th century, mechanical thinking is as fashionable as the so-called internet thinking today.
If we use eight-character proverbs to summarize the methodology of mechanical thinking, I think nothing is more appropriate than "making bold assumptions and carefully verifying". Generally speaking, it is to make assumptions, build models, verify data, optimize models and predict the future. This is also a set of ideas that have been used to this day. The results are also obvious, such as Newton, who explained the motion law of everything in the universe with several simple and clear formulas of the three laws of mechanics and the law of universal gravitation.
There are three points in this, one is the determination of the formula, the other is the simplicity of the formula, and the third is the universality of the formula. These are also three characteristics of mechanical thinking.
Look at certainty and universality first. No matter what Newton's law is applicable to, we can use the same formula to derive the corresponding definite conclusion. This is very important for people, because according to certain laws and principles, the world becomes known to us, and we can also use this to predict the feedback of other variables in the system and predict the future.
Simplicity is equally important. Newton lived in an era when the complexity of the universe was not weaker than it is now. The only difference is the mode difference determined by the way of thinking. There are thousands of celestial bodies in the solar system, which is complicated according to complete calculation. However, the law of gravity is very simple and elegant, which returns the action of thousands of celestial bodies to the consideration of the interaction between two celestial bodies in turn. Even further, because of the uniqueness of the mass of the sun, Newton regarded each planet and the sun as an independent binary system, which was further simplified. Going back to the two-object system is naturally a "decimal", but this so-called decimal world is actually not that small. For example, in a system that only considers two objects, it involves the situation of two objects, the interaction between them and the situation of their systems. If described by the concepts of mathematics and mechanics, then the two-object system involves four equations: isolation equation, interaction equation and field equation. Every time an object is added to this system, the number of field equations remains unchanged at 1, while the number of isolated equations increases linearly at 1, while the number of interaction equations increases exponentially. So simplification is also its core.
Second, big data and the world of large numbers.
Based on the above-mentioned mechanical thinking, when designing weapons and spacecraft, the Soviet Union relied on powerful mathematicians to establish complex and accurate mathematical models, hoping to use them accurately. Scientists in the United States have a weak mathematical foundation, so they took a different path-building a simple mathematical model, but relying on computers and a lot of data. As a result, the American way won
Dr. Wu Jun also cited another example of the "intelligent age"-Germany has perfect optical instrument technology, so it has made a difficult aspherical lens, and the instrument is small and perfect; Japan lacks such technology, so it uses a combination of multiple spherical mirrors to achieve the same effect. This kind of machine is heavy, but it is easy to produce and use in large quantities. After World War II, it was not Germany but Japan that became the largest country in optical instruments.
In both cases, multiple simple models are superior to a single accurate model. However, such a victory has a premise-based on big data. If the exquisite model under mechanical thinking is pure crystal, then big data is definitely sloppy. The molecules of gas are disordered and complex, but we can predict the overall diffusion and determine its overall physical properties. This is inseparable from the randomness of each molecule, and it is randomness that makes statistics meaningful. If a flu is spreading somewhere, it is difficult for us to judge whether an individual will be infected, but Google can even calculate where it will spread next based on people's search data. The infection rate is a simple statistical calculation.
Taking the above flu as an example, we can easily find that in the face of big data, accurate numerical values are actually not that important, and the points we care about do not have to be accurate to single digits. For example, when running an app, when the number of users reaches tens of millions, the number of DAU you pay attention to every day must be tens of thousands, or even more simply, it is only specific to hundreds of thousands or millions, and the single digits are no longer important. But in the face of big data, individuals are still unique. I only have two results: infected and not infected. Then in this case, the infection rate of big data becomes the background probability of whether an individual will be infected, and the individual's own health and activity area lights become other adjustment probability items.
As can be seen from the above, the value of probabilistic thinking is more prominent. In fact, the way of thinking based on big data does not make assumptions, but only makes correlation analysis based on massive data; Don't care about causality, only judge probability and correlation.
In addition to mixing instead of precision, relativity instead of causality and uncertainty instead of certainty, the most obvious thing about big data thinking is its complete replacement of samples, which is why big data is a "big number". There is no need to consider how to choose qualitative and quantitative random samples. The style of big data is that all data is included in the calculation. For this reason, from search engines to language recognition to machine translation, Google can emerge without changing its algorithm technology-the amount of data deposited by his family is too considerable. However, massive data is just gas, which is ultimately limited by the processing capacity of the stove. It is precisely because the growth of computer computing power can't keep up with the exponential growth of data and the number of servers can't keep up. Therefore, in the face of big data, simplified algorithms are particularly important. Such as Markov chain, such as Viterbi algorithm.
Third, the complicated life and the middle world
In management, we may abstract individuals into units and then manage them as a whole with mechanical thinking; In decision-making, we may decide what market to play next based on the strong correlation analysis results of big data. Both the mechanical thinking in the decimal world and the big data thinking in the big world are based on the assumption that the past can predict the future, and the purpose is also to predict.
However, most of the situations we encounter in real life are neither decimal nor large. If mechanical thinking is crystal and big data thinking is gas, then there is another kind of liquid-the middle view of the world. How many people are there in a listed company? Is a median; How many parts does a computer have? Is a median; How many birds are there in this forest? This is a middle value ... we are all embarrassed to live in it, just like the high school math teacher's vomit-what about physics without friction? Where is there no friction? Even accelerate the movement, you accelerate one evenly, let me see!
Decimals, medians and large numbers themselves are not clearly divided into orders of magnitude. In fact, this division is abstract and conceptual. For the middle number world, I think one of the ways to deal with it is to learn mechanical thinking or big data thinking according to different situations.
Let's look at the world close to decimals and mechanical thinking. The application scenarios of hypothesis-verification-application methodology are actually very extensive. For example, the core point of lean thinking is to minimize the feasibility verification in the hot lean entrepreneurship in the past two years. Because our cost is limited in real life, it is impossible to spread it all out with one idea, which is also uneconomical. We need to test and verify an idea with minimum cost. For entrepreneurs, it is necessary to verify whether users really have this demand in real scenes with minimum cost. But the inductive conclusion at this time is not as certain as the laws of mechanics, and the cause and effect are clear. In fact, induction cannot establish causality, and can only provide strong, weak or irrelevant references.
Look at this world close to large numbers. China people like to read history, and history itself is not big data. But the history books we read can only be non-random samples (median) selected from the vast amount of historical materials. Even so, this is still "nothing new under the sun"-predicting the future according to the past is indeed of practical value to some extent. The past provides us with an external perspective for future prediction, which can be used as a background probability for us to make specific predictions (I once saw someone jokingly calling it human experience big data, which I found quite interesting, but personal experience is far from reaching the level of "big data", and it is only a median at best, but it can already provide a background probability for future prediction and decision-making). Therefore, I think the greatest reference value of large number thinking for daily life is to provide a background probability from an external perspective. When faced with a specific situation, on the basis of this background probability, analyze the specific situation and make an independent probability correction.
Our knowledge comes from the experience of ourselves and others (contemporary or historical), or from the principles refined by our predecessors.
There are two ways to learn from the experience of yourself and others-one is to copy directly, and the other is to pursue causality (although most of the time it is only related) and apply it. Everyone thinks the second one is better, but in fact we are unconsciously applying the first one, because the second one is not only difficult, but also counterintuitive-that is to say, most people (including me) can't reach the standard of mechanical thinking most of the time.
There are also two ways to apply the principles refined by predecessors-one is the unitary thinking mode, and the other is the pluralistic thinking mode. Charles Munger once said that if you only have a hammer, everything in your eyes is a nail. Because if a person has only one or two modes of thinking, then when he thinks about reality, he has to distort reality to conform to his own mode of thinking. At this time, the more accurate and specific the model is, the more severe the restriction on one's thinking is. Therefore, Charles Munger pointed out that we must have multiple thinking models, and these models must come from different disciplines (in this subdivided world of disciplines, you can never expect to find all the truths in one department). This multiple thinking model is actually the same as many simple models driven by data on a single accurate model. Even if the amount of data we are faced with can't reach the level of "large number" in many cases, the premise for a person to establish a multi-thinking model is to absorb different orders of magnitude of the underlying data than a person with a single thinking mode. Because each model is derived from the principle of a large number of empirical data, many models are backed by data of different dimensions and orders of magnitude. This multivariate model has high fault tolerance, and the analysis of specific problems can really draw conclusions and predict the future from different aspects and dimensions.
This article refers to books:
1, Wu Jun's "Intelligent Age"
2. Wu Jun's mathematical beauty
3. Victor's era of big data: great changes in life, work and thinking.
4. Introduction to Weinberg's System Thinking
5. Charlie's "Poor Charlie Collection"