The Drawing Method of Artificial Intelligence (Understanding Artificial Intelligence (Illustrated))

1. Artificial Intelligence Image Robot

After the rise of artificial intelligence (AI) in recent years, many people have developed superstitions and idols due to their lack of understanding of its principles. They even believe that the scenes from Terminator and Star Wars are about to become reality. Even Hawking and Musk believe that AI may have originated from ideas and rule over humans. It's like a mountain separating them. Obviously, they don't understand the principles of AI and make jokes.

2. Artificial intelligence image generation

The principle of artificial intelligence can be summarized in one sentence as follows: Artificial intelligence refers to the level of intelligence of mathematical computing machines, which depends on the "algorithm". Initially, people found that the opening and closing of circuits can represent many circuits such as 1 and 0, which are organized together. Different permutations and changes can represent many things, such as color, shape, and letters.

3. High definition artificial intelligence image materials

In addition, the logic element (transistor) forms a mode of "input (pressing the switch button) - calculation (current passing through the circuit) - output (light on)". Imagine the dual control switch at home.

4. Artificial intelligence image creativity

In order to achieve more complex calculations, it ultimately became a "large-scale integrated circuit" - a chip circuit logic layer by layer nested, layer by layer encapsulated, and our method of changing the current state became a "programming language". The programming ape is the programmer who does this, and the entire process is fixed by the program.

5. Artificial intelligence image materials

So, in order for a computer to perform a certain task, programmers must first fully understand the process of the task. For example, the joint control elevator: Don't underestimate this elevator, it's also quite "intelligent". Consider what judgments it needs to make:

6. Artificial intelligence image high-definition

Whether the up and down directions are full or not during peak hours, whether the stopping time is sufficient for the relationship between the target floors inside and outside the elevator, whether it is a single or double floor, and so on, all the possibilities need to be considered in advance. Otherwise, there will be bugs. To some extent, it is the program ape who controls the world (program ape indicates great pressure). However, it is always a matter of personal effort. Programmers are too tired, and you can see that they are working overtime and their eyes are red.

7. Artificial Intelligence Picture Cartoon

So I thought: Can we let the computer learn on its own and solve problems on its own? And we just need to tell it a set of learning methods. Remember in 1997, IBM used a specially designed computer to win the international chess championship. In fact, its method was very clumsy - brute force calculation, the term is "exhaustive" (in fact, in order to save computing power, IBM manually trimmed many unnecessary calculations for it, such as those obvious stupid chess, and optimized it for Kasparov's style).

One of the methods for artificial intelligence images is to use object detection

The computer calculates all the moves and moves of each move clearly, and then compares them with human chess scores to find the optimal solution. In one sentence, "Strive hard to achieve miracles! But when it comes to Go, there is no way to exhaust it like this. No matter how strong the power is, there is a possibility of ultimate Go going far beyond the sum of all atoms in the universe (known). Even using the most powerful supercomputing currently, it will take tens of thousands of years.

One of the methods for artificial intelligence film reading is to use object detection

Before the maturity of quantum computer, electronic computers were almost impossible. Therefore, programmers added an extra layer of algorithm to Alfa Dogs: A. Calculate first: where to calculate, where to ignore B, and then calculate pointedly - in essence, it is still to calculate where there is "perception"! How should it determine "where to calculate" in step A?.

The process of artificial intelligence film reading is reflected in

This is the core issue of "artificial intelligence": the process of "learning". Think carefully, how do humans learn? All human cognition comes from summarizing observed phenomena and predicting, based on the rules of summarization, that in the future, when you see an animal with four legs, short hair, medium height, long mouth, and barking, called a dog, you will classify all similar objects you see in the future as dogs.

However, the learning method of machines is qualitatively different from that of humans: by observing a few features, humans can infer the majority of unknowns. For example, a three-cornered machine must observe many, many dogs in order to know if the running machine is such a stupid dog that it can expect to rule over humans. It just relies on its computational power to act recklessly! Strength work.

Specifically, its algorithm for "learning" is referred to as a "neural network" (quite deceptive). The principle is shown in the following figure:

(A feature extractor that summarizes the features of an object, integrates them into a pool, and outputs the final conclusion through a fully connected neural network) It requires two prerequisites: eating a large amount of data to try and error, and gradually adjusting its accuracy; The more layers a neural network has, the more accurate its calculations are (with limits) and the required computational power

The bigger it is, the bigger it is. Although the neural network method was developed many years ago (at that time, it was also called "perceptron"), it was limited by the amount of data and computing power, and it did not develop. Neural networks sound like perceptron does not know where the high-end is! This once again tells us how important it is to give a pleasant name to a researcher!.

Now, both of these conditions are met - big data and cloud computing. Who owns the data is the one who can do AI. Common application areas of AI at present:

Image recognition (security recognition, fingerprint, beauty, image search, medical image diagnosis) uses a "Convolutional Neural Network (CNN)" that mainly extracts spatial features to identify images.

Natural language processing (human-computer dialogue, translation) uses a "recurrent neural network (RNN)", which mainly extracts the features of the time dimension because speech is sequential. The time when words appear determines the design level of the semantic neural network algorithm, which determines its ability to portray reality. Andrew Y. Ng, a top bull, once designed a convolution layer of up to 100 layers (too many layers are prone to overfitting problems).

When we delve deeper into the meaning of computation: with clear mathematical laws, the world has quantum (random) characteristics, which determines the theoretical limitations of computers - in fact, computers cannot even generate real random numbers - machines are still clumsy. In early 2018, a company demonstrated on television their accuracy in using artificial intelligence to diagnose medical images, More accurate than a doctor's manual judgment.

What a normal thing to be surprised by! After the invention of the crane, of course, it was able to lift heavier objects than the strongest Hercules. The diagnosis of human doctors was a combination of pathogenesis and laboratory results, while the machine only estimated approximate values based on the doctor's historical diagnosis of a large number of laboratory results. Although it seemed accurate on the surface, once a new disease emerged, it did not know how to handle it.

——Can machines simulate the entire human body system? Organs, blood, mood, diet, labor intensity? Simulate the interrelationships between them to understand the pathogenesis—— It is impossible for the human body system to have quantum effects - random macroscopic manifestations, let alone the impact of human spontaneous consciousness on diseases. Humans can understand internal laws, and machines can only summarize surface laws.

In fact, we haven't even figured out what intelligence and consciousness are, and only a few philosophers argue about it. The scientific community has made no progress on this, and it's even difficult to come up with a definition! What machines have autonomous consciousness? Zhong Tian's viewpoint is very clear: AI will greatly improve productivity, there is no doubt, just like the internal combustion engine in the past.

It may also cause destruction - but it is not its own destruction, but its improper use. Killing people is not the fault of the knife. It can be certain that it is impossible for it to rule humanity. At present, it is still home to create people, and it is more thorough intelligence. The question then is - will AI lead to a wave of unemployment? In the first industrial revolution, new textile machines took away the jobs of textile workers.

At that time, the workers really ganged up and caused a commotion, smashing some factories and machines. Is it now at that stage again? Since the emergence of computers, people have been worried that they will take their own jobs. In fact, there is a simple and crude theorem in economics: human desires and needs are endless. When technological progress increases the supply of low-end products, demand will naturally move towards the high-end.

There will be employment if there is demand (note: demand=desire to have purchasing power) structural unemployment has been continuing, but on the other hand, new jobs have also been increasing because: in the AI AI revolution, there must be new things. What occupations are easy to replace? From the perspective of "artificial intelligence=mathematical calculation", it can be seen that the more specific the algorithm, the more easily mechanized and repetitive labor is replaced.

Here is a problem to pay attention to: some jobs seem to require complex intellectual labor, such as bank teller: she cordially greets you, asks you carefully about your needs, processes a lot of documents, and signs, which is not competent for a robot. But according to the "demand path theory" (to be published later, please pay attention to the official account "Kung Fu Reading"), this process is not replaced, but skipped.

This profession is also a dangerous structural adjustment, and the process is gradual. It does not mean that all AI products will be rolled out at once, and a large number of workers will lose their jobs at once. In theory, there is no absolutely irreplaceable job, and the trend is vast. It cannot be stopped but to strengthen learning and increase investment in education. However, do not be frightened and anxious by those sensationalist articles.

There is also a more thorough approach, which is to have capital. I will talk about this later. (Zhong Tian, April 27, 2018) First published on the official account Kung Fu Reading