Monitoring Desk
Samsung described the ability to copy the brain
At the end of September, the specialists of the Korean company, together with scientists from Harvard University, presented the concept of imitating the human brain using semiconductor chips. The engineers were able to describe a way to copy and transfer the structure of the brain into a neuromorphic chip. Such a chip of the future will allow to reproduce unique computing characteristics – adaptation to the environment, ease of learning, low power consumption, autonomy and cognition. “Lenta.ru” tells why you need to copy the brain and transfer it to physical memory, why the human mind is superior to the most powerful computer and what prevents the development of artificial intelligence.
Neuromorphic chips differ from conventional chips in the way they work with information. These types of processors process data in much the same way the human brain reacts to incoming data – thanks to billions of neurons working together. The connections between neurons are constantly changing in response to external stimuli – this is how learning occurs. According to experts, neuromorphic processors are necessary to create a full-fledged artificial intelligence.
It is theoretically possible to create such a superprocessor, which would be built from transistors that mimic the operation of neurons. For this, there is an area such as neuromorphic engineering, which is dedicated to the transfer and reproduction of analog computations of the human brain in the form of a digital solution.
One of the most striking and successful examples in this area is the TrueNorth modular system, created by IBM for DARPA in 2014. It is an artificial network built on the basis of several processors. The system contains 4.5 billion transistors with 256 million emulated synapses. To replicate the way the brain works as closely as possible, the engineers managed to create an energy efficient network. So, if for the functioning of an Intel processor, about 140 watts of energy are needed, then TrueNorth consumed only 70 milliwatts.
A key feature of the human brain is the processing and transmission of information through chemical compounds. If there are billions of neurons, then there are an order of magnitude more synaptic connections between them. Therefore, it is still impossible to copy the so-called human thinking in its pure form and transfer it to digital. Samsung engineers estimate that the brain consists of one hundred billion neurons and even more synaptic connections, so it would take several trillion units of independent memory to digitize it.
The process of adaptation of a living brain in an artificial form is so complicated that, in addition to computer science, scientists turn to biology, mathematics, electronic engineering and physics.
The architecture of most modern computers is based on a concept that was developed by the Hungarian-American mathematician John von Neumann in the middle of the 20th century. It describes the principle of joint storage of commands and data in the computer’s memory. The von Neumann architecture is built on the principles of memory homogeneity, targeting, and programmatic control, but it has many bottlenecks. According to the expert on neural networks Catherine Schumann, in order to invent a true neuromorphic system, it is necessary to move away from the principles of von Neumann and rely more on neuroscience.
Neuroengineering has two fundamental tasks – to create a mechanism that could learn in the image of a living mind and uncover the secret of how the human brain works.
Thanks to evolution, the human brain is an almost perfectly optimized system. Firstly, the brain is completely autonomous from the external environment, and secondly, it consumes very little energy. If, to solve a problem, a computer based on the von Neumann concept repeatedly transfers information between various components, then the brain selects a separate group of neurons for a specific process. According to neuroscientists, the brain requires about 20 watts of energy to operate, which is about half what a modern laptop needs. Third, it is extremely fault-tolerant, since information is stored in multiple locations and in excess.
Fourth, a highly organized mind is flexible and capable of handling several processes at the same time. Unlike a machine, a person is able to quickly adapt to new conditions and solve new types of problems.
How will an artificial brain change our lives?
If in the future scientists can repeat the structure of the human brain at the microprocessor level, then a revolution will take place in the segment of artificial intelligence. The chips found in laptop computers, smartphones, cars, and IoT devices will be an order of magnitude smarter and smarter. Perhaps the machine will not be able to copy the mind and creativity of a person, but there will be minimal differences between the human brain and the neural network.
Using the best qualities of the brain (namely, the ability to process information autonomously and in parallel), the machine will gain greater freedom in solving problems and a wider range of actions. For example, analyzing an obstacle on the road, an unmanned vehicle will be able to process information and make a decision without contacting the manufacturer’s servers. This eliminates the lag that occurs when the autonomous vehicle communicates with the remote hub.
Having more freedom, smart home gadgets will become more self-reliant and autonomous. The quality of local analysis of information will not be inferior to the quality of data processing using external systems. Machines will use energy more economically, the likelihood of a cyberattack will decrease – all due to the fact that the information will not go beyond the processor of a conventional smart socket or a robot vacuum cleaner.
Courtesy: (lenta.ru)
