One of the key announcements this year was an announcement by Google, which claimed to have designed a computer that could achieve quantum supremacism. According to Ortíz, this means the traditional computing we use on computers and mobile phones can do things that are impossible today.
Experts say: “This kind of calculation is very important because it will improve the machine learning process and make it more advanced. In addition, this new type of informatics has previously made it impossible or very difficult. You will be able to solve these problems, such as discovering which molecules can produce really effective drugs. ”
Currently, the processing speeds of the world’s most powerful supercomputers exceed petaflops, which are 1 billion floating-point operations per second, depending on the parameters used to measure the power of these huge computers. The Oak Ridge National Laboratory (ORNL) Summit holds the current record at 148.6 petaflops. But this won’t last long: China is expected to reach the expected milestone of breaking the 100 billion level this year. Machines that can exceed the superscale supercomputing threshold.
Other countries do not want to be left behind. The US Department of Energy commissioned the construction of the 2 billion mega supercomputers Aurora and Frontier to be used in 2021. The EU and Japan are developing their own projects. In addition to increasing power, overcoming the exaflop frontier is crucial. Because these machines are comparable to the human brain’s ability to process neurons, they ultimately drive projects such as the Human Brain Project.
Computers have an advantage over the human brain in terms of basic operating speed. Personal computers today can perform basic operations, such as addition operations, at a speed of 10 billion times per second. Through neuron transmission information processing and communication with each other, we can evaluate and calculate the basic information processing speed of the brain. For example: “Activate” the action potentials of neurons-the the peaks of electrical signals initiated near neuronal cells, and transmitted to axons to connect with downstream neurons.
The highest frequency of neuron activation is 1000 times per second. As another example, neurons mainly release chemical neurotransmitters on a special structure of axon terminals called synapses, transmitting information to partner neurons. At the same time, partner neurons are In a process called synaptic transmission, the combined neurotransmitters are converted into electrical signals. The fastest synaptic transmission takes about 1 millisecond. Therefore, in terms of peak and synaptic transmission, the brain can perform up to about 1,000 basic operations per second, which is 100,000 times slower than computer operations.
In terms of basic operation accuracy, the computer has more advantages than the brain. According to each number assigned by the digits (binary, or 0 and 1), the computer can express the quantity with any desired accuracy, for example, 32-bit binary equals more than 4 billion decimal. Experimental evidence shows that due to biological noise, most of the nervous system has a variability of a few percentage points, and the best accuracy is 1%, compared with the accuracy of the human brain nervous system is only 100% One in ten thousand.
However, the calculation speed of the brain is not slow. For example: a professional tennis player can observe and analyze the running trajectory of tennis. The maximum running speed of tennis reaches 160 miles per hour. According to the running position of the tennis ball, they quickly move to the best position of the court. Swing the arm and shake the racket to hit the tennis ball to the opponent’s court. The hitting action is completed within a few hundred milliseconds. In addition, the energy consumed by the brain to complete all tasks (with the help of body control) is only one-tenth that of a personal computer.
How does the brain do this?
An important difference between the computer and the human brain is the information processing model of each system. Computer tasks are mainly performed in serial steps. This can be achieved by engineers by creating a sequential flow of instructions. For this continuous cascade operation, Each step must have high accuracy, because errors will accumulate and magnify in successive steps. At the same time, the brain also uses a continuous information processing mode. In the case of hitting tennis, information is fed back from the eyes to the brain and then transmitted to the spinal cord, which controls the muscle contraction of the legs, torso, arms, and wrists.
But the human brain can perform parallel information processing, and it has an advantage in processing a large number of neurons and establishing connections for each neuron. For example: the rapid movement of tennis will activate the retinal cells-photoreceptors, whose job is to convert light into electronic signals. These signals are then transmitted in parallel to different types of neurons on the retina. When the signals from the photoreceptor cells are connected by two to three synapses, the information about the position, direction and speed of the tennis ball will be extracted by the parallel neuron circuit and then transmitted to the brain in parallel. Similarly, the motor cortex (the part of the cerebral cortex responsible for the control of motor consciousness) will issue commands to control the muscle contraction of the legs, torso, arms and wrists. The body and arms can fully coordinate and adjust the body’s optimal position to hit tennis.
This massively parallel strategy is possible because each neuron collects input information and sends information to other neurons. For mammalian neurons, there are an average of 1,000 neurons that input and output information. In contrast, the computer has only three nodes per transistor for data input and output. Information from a single neuron can be passed to many parallel downstream paths. At the same time, many neurons processing the same information can concentrate their input information to the same downstream neuron. Downstream neurons are very useful for improving the accuracy of information processing. For example, the information represented by a single neuron may be “noisy” (accuracy of one percent), and ordinary downstream partner neurons can express information more accurately Accuracy is one thousandth).
At the same time, the computer and the human brain have commonalities and differences in basic unit signal patterns. Transistors use digital signals, which use discrete values (0 and 1) to represent information. The peak value of the neuron axon is also a digital signal, because the neuron is either activated or not activated at any time. When the neuron is activated, all peaks have almost the same size and shape. Realize reliable long-distance peak propagation.
However, neurons also use analog signals, which use continuous numeric values to represent information. Some neurons (like most neurons on the retina) are in a non-peak state, and their output is transmitted through a hierarchical electrical signal. This is different from the peak signal. Their size can be continuously changed, and more information is transmitted than the peak signal. The receiving end of the neuron (usually occurs in dendrites) also uses analog signals to integrate thousands of input information, allowing the dendrites to perform complex computational processing.
Your brain is 10 million times slower than a computer. Another notable feature of the brain, which can be seen in tennis is the strength of the connection between neurons, which can be modified in response to activity and experience. This process is generally considered by neurological scientists as learning and The basis of memory. Repeated training can make the neural circuit better configured to complete the task, thereby greatly improving the speed and accuracy.