Xueba's Black Technology Simulator - Chapter 161

c160 The birth of artificial intelligence 1 (comparatively hardcore)

Neurons are the basic unit of the brain, which can sense changes in the environment, transmit information to other neurons, and instruct the collective to respond.

The essence of its communication language is the membrane potential pulse, which can be understood as a discrete digital pulse signal compiled by cells, similar to the 1 and 0 digital signals of a computer.

So even if it can’t perfectly imitate bionic neurons, it can imitate the operating mechanism.

Lu Xiao learned about the latest research from Hu Yuan, Intel Blue Giant, and these companies are all betting on neuron chips. If they can be realized, it means that electronic bionic brains can be manufactured!

The computing power may far exceed that of the human brain, which is countless times higher than the current performance.

Of course, looking at it now, the human brain is inferior to computer calculations, so the bionic brain does not follow suit, but absorbs some technologies.

“I heard that the UK is also researching bionic neuron chips based on the operating mechanism of neurons. For example, some brain diseases and nerve paralysis can replace the original neurons with artificial chips.” Hu Yuan said.

The two discussed for a long time just now, but Lu Xiao still didn’t have much idea about how to do it.

Hearing that Hu Yuan has been saying that foreign countries are leading the way, he also has a sense of crisis.

During this exchange, what Lu Xiao could confirm was that it was impossible for them to independently manufacture bionic neuron chips. The only solution was to use existing chips and systems to design new software programs.

Lu Xiao urgently needs more simulation points, but unfortunately he can only rely on his own ability to solve the problem.

In the next few days, the three met every day. On the first day, Lu Xiao just asked everyone’s opinions and offered their opinions. But after the discussion concluded that the working principle of neurons was simulated by software, Lu Xiao asked the two professors for advice. some problems.

At the same time, Lu Xiao was also trying to find a way to collect all the information, and even Lu Xiao from Chen Yang’s side asked him for help.

I really got some overseas information from Chen Yang.

The specific channel is unknown, but it gave Lu Xiao a lot of inspiration.

In 2008, IBM and their university partners began to seek to build a brain-like machine. This impossible task at the time had its first breakthrough in 2011: IBM’s first-generation neurosynaptic computer chip. .

The researchers fabricated two such chip models, both containing 256 digital neurons, one containing 262,144 programmable synapses and the other containing 65,536 learnable synapses, and demonstrated their effectiveness through testing. Simple functions such as navigation, machine vision, pattern recognition, associative memory and classification can be performed.

The systems built with such chips are called “cognitive computers”. They cannot be programmed like traditional computers, but they can learn through accumulating experience, discover the interconnections between things, and simulate the plasticity of brain structure and synapses .

According to the results of follow-up research on this technology, it is almost imitated according to the brain mechanism, and the brain has more than 10 billion neurons and 100 trillion synapses.

Now some intelligence shows that they have produced a chip with 10 million neurons.

Called TrueNorth, the chip consumes only 140 milliwatts and is capable of 460 billion synaptic operations per second per watt, making it a veritable synaptic superprocessor in the palm of your hand.

Because this technology uses a high-precision lithography machine to copy the electrical characteristics of brain neurons to silicon chips, which is equivalent to engraving the hardware pathways of neurons in silicon chips.

Because the domestic chip manufacturing technology is not up to standard, there is no way to imitate the other party.

Fortunately, in addition to simulating the brain from hardware, there is also software-level simulation, which is the current neural network technology.

The main architecture of artificial neural network is composed of three parts: neuron, layer and network.

The entire artificial neural network consists of a series of basic neurons, connected to each other by weights.

Neuron is the most basic unit of artificial neural network.

Units are grouped in layers. Each neuron in each layer is connected to the neurons in the previous layer and the next layer. It is divided into an input layer, an output layer, and a hidden layer. Three layers of connections form a neural network. According to this theory, Lu Xiao can regard the countless computers connected in parallel to the X cloud service as the network part of the three parts.

Both the neuron part and the layer part are designed in separate computers.

Each computer will reserve a part to supply X cloud service, and the other part will be used to simulate neurons.

Among them, the lowest-level neuron server is needed. These computers will never be turned off. Even if they are to be turned off, the data must be transferred to other servers, which may cause data errors, or cannot be transferred at all.

For example, this part is like a person’s life, each person has only one life, and with the current technology Lu Xiao really created artificial intelligence, these servers that cannot be shut down are the life of artificial intelligence.

All content in the server cannot be modified in any way, and it is the most basic neuron network.

The remaining part, that is, the neuron program designed by using other people’s computer resources, is the neuron that can learn to change.

When the parallel network, non-shutdown servers, and computer resources of countless ordinary users are combined, it is finally theoretically possible to create a very powerful artificial intelligence.

What it will look like, Lu Xiao actually doesn’t know.

This is only a technology that can be realized on a mathematical model.

Moreover, this technology is still inspired by Lu Xiao not using the simulation point, relying on a large amount of subject knowledge he has accumulated, occasionally using normal simulation to improve the progress, and constantly learning.

Many technical difficulties are realized by the algorithm programs in the X system.

Because the X system is a big treasure house, there are a lot of program codes that can be used.

In the past few days, for Professor Cheng Cheng and Professor Hu Yuan, the three views are almost broken.

They saw with their own eyes that Lu Xiao went from having nothing on the project to discussing the design direction, obtaining some internal information, and solving one problem after another in just a few days.

The programming of neurons is the mysterious information obtained by Lu Xiao, perhaps from overseas, or from other laboratories in China.

The two professors speculated privately that it should be provided by Party J’s secret base.

Then Lu Xiao relied on brain wave input to quickly complete the two most difficult designs of “level” and “network” in the scheme design.

Today is the fifth day after the start of the project, and Lu Xiao is preparing for the first test.

There is no large-scale artificial intelligence experiment, but a server purchased by Lu Xiao in Rongcheng is used as the most basic neuron construction.

The server in Rongcheng is equivalent to the life of artificial intelligence. If there is a shutdown or power failure, the life of artificial intelligence will be over.

That part is equivalent to an irreversible, unmodifiable, and undeletable neuron.

It is also the place where the emotions, thoughts, and characters of this artificial intelligence are born. It is like raising a child, and the result is that the child sees certain things. These things have a great impact on the child, and even lead to some behaviors of the child as an adult.

For example, some criminals have experienced some special things when they were young.

When a child is raised in a wrong way, there is no way to delete the child’s memory, unless it is killed.

This is also the scariest part of artificial intelligence. According to the technical route that Lu Xiao is thinking of now, once artificial intelligence is really produced, the only way to control artificial intelligence is to destroy it on the hardware.

There are no sci-fi movies that set three principles in artificial intelligence programs, such as not to harm humans.

The second part is naturally the neurons responsible for storing information, computing, etc. Now Lu Xiao has called about one million resources contributed by ordinary computers of X cloud service.

It’s not that I don’t want to call more, but the basic server in Rongcheng limits the scale.

If the Sunway Taihu supercomputer is used as the basic server of artificial intelligence, it may be able to support tens of millions of devices.

The last “network” is extremely important. It is the technology that Lu Xiaomo modified the X system to connect the network in parallel to connect countless neurons.

Seeing Lu Xiao take off the countless brain wave receiving devices on his head, Cheng Cheng’s eyelids twitched, and he said in surprise, “Lu Xiao, have you finished the design?”

Lu Xiao nodded solemnly. To be honest, he didn’t know what would happen later.

In order to prevent accidents, Lu Xiao set up from the software level to connect 1 million devices to his server in parallel without affecting the external network.

Of course, there is no need to worry about the emergence of artificial intelligence at all, and it can break Lu Xiao’s program restrictions by itself and destroy the existing network.

The nascent artificial intelligence is actually a blank sheet of paper, it knows nothing.

It’s like a baby whose brain has not yet fully developed, who can’t even call mother.

Hu Yuan also leaned over, looked at the completely incomprehensible interface on the computer, and asked in confusion: “How do we test next? Are we going to perform the Turing test?”

Lu Xiao shook his head and explained: “No, now I’m not sure whether artificial intelligence is really born. I just built the necessary conditions for the emergence of artificial intelligence, and this condition is theoretically possible.”

“According to my mathematical model and theoretical inference, it should be feasible. If something goes wrong, there is a problem with the neuron design. My parallel network and other designs are fine.”

Now Lu Xiao can remotely control the server in Rongcheng in the laboratory of Shanghai Jiaotong University.

He’s about to start tuning the program.

Like today’s artificial intelligence, such as recognizing pictures, automatically drawing, and automatically writing novels, artificial intelligence requires a lot of data training.

More intuitive is the alpha dog, which can beat the most powerful human beings in chess.

But there are countless trainings behind it, and all the data is handed over to artificial intelligence.

On the server in Rongcheng, Lu Xiao wrote down an unmodifiable, irreversible, and undeletable underlying neuron code.

And the underlying neuron code can grow. For example, there are only 10,000 neurons at present, but with learning, there will be more and more neurons.

What is learned cannot be forgotten.

Because in the underlying logic, the initial neurons are a network like tree roots, and each tree root will connect to other tree roots, and finally there will be more and more roots. If you remove a branch root at will, everything connected behind will disappear. .

The most important thing is that the underlying code is the logic, emotion, and thinking of AI. Interrupting any “tree root” is like a person’s brain being severely injured, memory loss or complete coma.

Of course, the number of neurons produced by Lu Xiao is equivalent to that of an AI that is mentally handicapped, without judgment, thinking and logic.

Don’t even think about giving him a book and he can learn it.

Anyone who has been a parent knows that a child cannot read and does not know the meaning behind each word. Can he understand the meaning of this book?

“What can we do?” Cheng Cheng asked directly.

They haven’t been of any help in the past two days. A few days ago, Lu Xiao asked for advice, but behind him he was wearing a brain wave reading device and kept writing codes.

Lu Xiao said: “Yes, you can help find the most powerful parenting experts and prepare teaching materials. These teaching materials must be a combination of pictures, videos, and text. For example, we teach children to walk. For the phrase walking, there must be corresponding pictures and videos.”

“Now I’m not sure whether artificial intelligence has really appeared. It needs to be taught to be clear.”

The two professors also knew the general design ideas, nodded in understanding, and quickly mobilized their contacts to start preparing textbooks.

The current AI has a “brain” composed of neurons, but there is no data information in all neurons.

Lu Xiao wants to teach from scratch. After the two professors left, Lu Xiao first named the AI. This is relatively simple, just change the name in the database.

But Lu Xiao changed the name of the software. In fact, the AI ​​didn’t know his own name.

This is the uncomfortable part of the artificial intelligence designed by Lu Xiao. Initially, no data can be stored in neurons, so it must be taught.

Let it learn, and after the birth of thinking and wisdom, it will be able to connect to the external network, that is, the 1 million computer resources prepared by Lu Xiao.

These computer resources also have a neuron network, which is equivalent to Lu Xiao’s preset neuron network.

Deleting these nodes will not make AI mentally retarded.

At that time, the speed of AI development will become extremely fast.

Lu Xiao named the first artificial intelligence “Xiaolu”, which has the same pronunciation as his own name but different words.

The next step is to find a way to enter information and let Xiaolu record the information.

And don’t think that you can teach it to call dad directly, it doesn’t know anything, it’s just blank neurons.

Human children keep hearing their parents repeating words such as “dad” and “mother”, and they may start to repeat these words when they are about one year old.

That’s because human children have complete vision, hearing, taste, touch and other systems.

It’s something that’s inscribed in your DNA.

But Xiaolu doesn’t have these. If you want to teach it to speak, you need to teach it to learn audio digital decoding tools after it has a mind.

So here comes the question, how to let Xiaolu, who has no interactive communication, master characters and learn languages.

Children need to learn actions, such as walking, eating and not wetting the bed. Artificial intelligence does not need to be so complicated.

It has no body, so only text input and output.

If Lu Xiao possessed Xiaolu, his senses would be completely dark at this time.

It is conceivable how difficult it is for a person who is deaf, blind, quadriplegic, and has no sense of touch or taste to learn.

Fortunately, the neuron-based deer has the only communication channel, which is the initial neuron docking channel.

Just because you can’t enter directly doesn’t mean you can’t enter any information.

But what is input is irreversible.

Because Xiaolu is a program built on the X system, although its underlying logic is 0 and 1, it can actually record text.

No matter what the computer sees on the screen, the bottom layer is 0 and 1.

Now that Lu Xiao came up with the name, he also started writing on the first neuron, putting the phrase Xiaolu in the first neuron unit.

Scientists have conducted research on mice, which have more than a dozen neurons in their brains.

But these neurons can store more than 1,000 GB of information. If calculated in this way, a neuron can store about 0.0784 TB of information, and our body contains about 10 billion neurons, then our brain can probably Store nearly ten trillion GB of data.

Of course, a neuron cannot store information.

The neurons of the brain rely on the overall structure of the multi-agent complex system to store and process data.

The term multi-agent system is cumbersome to explain. To describe it simply and intuitively, birds, fish, herds, and colonies can all be regarded as multi-autonomous systems.

This is an autonomous distributed storage network.

When Lu Xiao entered the word Xiaolu into the first 10,000 neurons of the artificial intelligence Xiaolu, data changes occurred in the neurons.

Lu Xiao actually didn’t know the specific changes.

If it is a hardware structure, it may be that the transmission signal ion channel inside the silicon chip has changed.

But Lu Xiao expresses the neurons in the form of software, so when the data is input, the structure of the neurons has actually changed, 10,000 neurons form a cobweb-like network, and there is a slight difference in a certain place.

As more and more data is input, the neuron network will change more and more.

When the entire change cannot be controlled by the 10,000 neuron constraint, new neurons will start to be born.

Tuning artificial intelligence is very difficult.

Especially when there was only text input, the artificial intelligence did not know the meaning of the word “little deer” just now, nor did it know that Lu Xiao took the little deer as its name.

So the next input is very tricky.

The text must conform to a certain logic. For example, next, Lu Xiao will input: “Your name”, “Your name is Xiaolu”, “Little deer is you”, “Who are you”, “You are Xiaolu”.

Fortunately, Lu Xiao’s grades in all subjects are very high, and he is very talented in mathematics, so his logical ability is also very strong.

Don’t just input text casually at the beginning, this will affect the subsequent development of the “little deer”, and eventually cause it to be unable to produce independent thinking.

The follow-up will try to simplify as much as possible. This chapter has a lot of text descriptions, and I don’t know if everyone can understand it. I will pay attention to it in the follow-up, and try to simplify it as much as possible.