Shulou(Shulou.com)06/02 Report--

Article Source: Citic Institute of Technology | Technology sharing of CITIC payment and settlement team II-Liu Chuang, head of the payment and settlement machine learning technical team, shares "AI and first principles"

Shared by: Liu Chuang, head of machine learning technical team of Yixin payment and settlement

The original article was posted on a personal blog: pigs in the zoo.

1. AI and first principles

AI, that is, artificial intelligence, has been very popular in recent years, but what is the nature of AI? Using first principles to explore the nature of AI may be easier to understand. Next, let's explore AI, first principles, and how to use first principles to understand the nature of AI.

II. First principle 2.1 what is the first principle

The first principle, this concept comes from the ancient Greek philosopher Aristotle.

"in any system, the existence of the first principle is the most basic proposition or hypothesis that cannot be omitted or violated." -- Aristotle

It sounds a little obscure, but the explanation is simple: everything has its inherent "first principle". My understanding is that the most essential thing is that this thing is actually there, based on the most basic assumptions, and you have to recognize it, it doesn't need any premises or proof, and with it, you can deduce other conclusions that eventually form the whole system.

Give me a few examples.

We began to learn Newton's mechanical theory in junior high school. His first principle is gravitation. His so-called "basic proposition" is that all things attract each other and maintain the inertia of motion when there is no external force. These two hypotheses construct the whole edifice of Newtonian mechanics and guide all aspects of architecture, industry and even astrophysics.

Elon Musk, who reintroduced this concept to the world, is also a practitioner of first principles. Musk had a dream from an early age: to move to Mars, so he began to devote himself to the development of civilian rockets. At first he went to cooperate with the Russian Rocket Company, and the price offered to him was more than $65 million, which was too expensive for him. So he reconsidered the question, and Musk devoted himself to studying the principles of rockets for two years, understanding the principles and processes of rocket manufacturing, and thinking about the core costs in the process of rocket manufacturing. After continuous thinking and practice, Musk's SpaceX company cut the cost of launching the rocket to 1/10.

Then he began to think about how to reduce the high cost of electric car batteries in Tesla, his electric car maker. At that time, the price of energy storage batteries was $600 per kilowatt-hour, which was too expensive for the whole car. Musk led the team to carefully analyze the composition of the battery, thinking from the first principle: what exactly is the battery pack made of? What is the price of these battery materials in the raw material market? After inspection, they found that the battery is made of carbon, nickel, aluminum and some polymers, and if you buy these raw materials from the London Metal Exchange, it will cost as little as 80 US dollars per kilowatt-hour, which is as much as eight times the previous price. In other words, if he can master the technology and methods of making batteries, then only the cost of raw materials will be left. In the end, they did, greatly reducing the cost of batteries, thus enabling the entire electric car industry to flourish.

2.2 first principles and deductive methods

The first principle and the deductive method go hand in hand. In fact, the deductive method is the derivation of syllogism. Syllogism refers to "major premise, minor premise and conclusion". The major premise refers to the general axiom, while the minor premise is often some special facts. with the major premise, the small premise can be deduced from this deductive method.

For example: "people will die. Aristotle is a man, so Aristotle will die." This is an easy-to-understand example of syllogism.

Do you think this method is very simple and useless when you see it here? In fact, there is something here, I don't know if you have noticed. That is, the major premise is very important, and if the major premise is wrong, or if the major premise is not a very solid assumption, then the conclusion may not be correct.

How to ensure that the major premise is correct? This requires that the major premise is also deduced according to the deductive method. In other words, the major premise is the conclusion of another deductive reasoning, and the major premise has its own major premise. In another deductive reasoning, it exists as the conclusion of that major premise. This forms a recursion that can be traced back to the first principle.

Just like the proof questions in school, I give you some assumptions and let you get the final conclusion according to some reasoning formulas, so the conclusion is very credible.

One of the most classic examples of the first principle and deduction is Euclid's "geometric primitive". He deduces 476 geometric conclusions from the first five axioms and five postulations, which are solid and solid.

The first principle, supplemented by deduction, is a very important way for us to think about the world. We should use this method to establish our view of the world. For anything, we should constantly think about its most fundamental cause and essence. Found this origin, essence, its first principle, other phenomena, conclusions and problems, can be easily solved.

Of course, this is not an easy thing, in the process, we need to keep learning, thinking, distinguishing false things, keeping the real things, constantly thinking about whether we have explored the most authentic things, constantly questioning, verifying, and repeating until we are sure that we have found it.

III. Artificial intelligence

Back to the topic of AI. In 2016, AlphaGo defeated Lee se-dol, and AI began to become popular again, so that the entrepreneurial circle, industry and even the national level are all buzzing about artificial intelligence, everywhere advocating that artificial intelligence will surpass human beings, and human beings will face extinction. But is that really the case? Let's think about what's really behind this hottest concept.

Let's start with a very popular AI chestnut, ZAO~.

First of all, why is AI so popular? Maybe what you see is the victory of AlphaGo, while what I see is the lack of technology as productivity.

We have experienced the IT revolution, the Internet tide, has lasted for more than 30 years, but now these two technological support points have begun to face the disappearance of dividends, people are in urgent need of a new form of technology to take over the baton of productivity. Look from left to right, the blockchain is too much hype, the Internet of things has been unable to land for a long time, and 5G is currently limited to the communications industry. Only AI- artificial intelligence seems to be the next stick of technology as productivity. In this way, AI was pushed to the altar, although it is not yet mature, it has experienced three troughs, but the desire for scientific and technological productivity, people can not wait to push it onto the stage.

Smart people may ask: why do you need new productivity? This is the way of thinking of the first principle, you are trying to explore the deeper origin of things, very much like you.

This question is actually very difficult to answer. To talk about my understanding, it is because without new productive forces, goods and values cannot be created more and more efficiently, and the ever-expanding consumer desire and credit expansion will come to an abrupt end. The world economy will fall into stagnation and recession, thus giving rise to a variety of economic, social and political problems. Jobs Ray Dalio of the financial world once made a 30-minute video about the workings of the economy, which is easy to understand. If you are interested, you can Google it.

Putting aside these flattering words on the Internet, let's take a look at the development of the AI industry. Knowing this, you may not follow others to carnival with "them", or fear the future "extinction" of mankind.

Then the question becomes: what can I do to get the most quick, efficient and comprehensive understanding of the current AI industry, instead of just listening to one-sided comments in the media?

I have come up with some ideas: check the financing of AI-related companies on 36 krypton, because I firmly believe that investors are far superior to ordinary people in terms of ideas and information, and that if they invest real money, they will consider it more carefully and comprehensively. So, by observing their investment in these AI companies, trends and quotas, you can have a basic understanding of the industry.

I also found Baidu that I think is the best in the direction of AI (of course, you can think of it as other enterprises, it doesn't matter, this is very subjective). Go to the websites of these AI giants and observe their industry cases, solutions, and technical white papers, you can quickly know where the first-tier enterprises have landed AI.

I will also look for some AI enterprise research reports, through the eyes of professional consultants, quickly understand the situation in their eyes of the industry, you know, the value of an industry research is the crystallization of quantitative indicators and subjective feelings after professionals have spent time and money.

Is there a better, more objective, more cost-effective way to help you understand the real situation in the industry? this is something you should think about, so that you will be more convinced of the conclusion you have come to.

Let's go back to the professional field of AI, when is the most authentic thing in AI? When I first started learning, I didn't answer this question, so the learning efficiency was not very high, and it took me a long time to touch the door.

To understand a discipline, you must first understand the framework of the whole discipline, and then understand the most essential content behind each large classification. The subject of artificial intelligence is indeed too big. Take the most popular deep neural network at present, for example, it is actually only a branch of machine learning, but the effect of neural network is surprisingly good. That's why it stands out from many machine learning methods.

If we continue to dig deeper, the whole machine learning is essentially looking for statistical laws in the data, which need to be expressed, either through linear methods, or through nonlinear methods, in a more general sense, through probability distribution. The density function of probability distribution is itself a function. Since it is a function, it can be expressed and fitted mathematically, and the fitting method can be expressed by the combination of countless neurons of the depth neural network.

If we explore further, the essence is the optimization theory in mathematics and the omnipotent approximation theorem in functional, including the reverse gradient descent in the process of finding parameters, and so on. Many theories are supported by rigorous mathematical theorems in many branches of mathematics such as information theory, probability statistics, functional, optimization and stochastic processes.

Is it familiar? As if back to the support of Euclid's five axioms and five postulates, yes, the whole artificial intelligence is based on modern mathematics, which is the perfect interpretation of the first principle!

Conclusion

Therefore, do not be confused by the media and experts, go deep into them, to explore the so-called artificial intelligence in the end, explore its nature, all the fog will disappear. This is the way of thinking of the first principle.

Whether it is the process of understanding AI, or the problems encountered in work and life, do not be confused by appearances and noise, settle down, seriously explore the real things, put aside the layers of others to smear the coat, see its most internal origin, in this way, you will be a lot less confused, to work, to life, and even to life, more self-confidence and calm.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.