Nothing man-made compares to the efficiency of operating the human body so far. One gram of the human genome is capable of storing 455 billion gigabytes of data, millions of times more than any human-developed storage capacity. The creative capabilities of the human brain exceed the capacity of any artificial intelligence system. The energy used by this brain to process millions of nerve signal requests does not exceed the size of one tablespoon of sugar. The heart, liver, and kidneys perform their functions day and night without needing maintenance. The thermal energy generated by the human body during the day is equivalent to that of an electric heater within two hours of time. This body, which performs very complex functions, has become the destination of scientists during the current period to develop artificial intelligence systems after the current technology was unable to continue to develop.
The current silicon technology based on digital smart chips and binary language (Binary System) has exhausted all its tools to keep pace with the race to develop artificial intelligence, and it is facing major problems related to the energy used to operate these systems, which prompted companies such as Microsoft to move towards nuclear energy to provide the electrical energy needed by artificial intelligence systems, in addition to other problems related to the ability to process giant amounts of big data. It needs new engineering other than silicon-based to process this data.
Therefore, many scientists at research institutes and technology companies have tended to develop biological computing techniques by simulating the way the human brain works or by transplanting some brain organs and exploiting their operational capabilities to become an alternative to the silicon system to overcome energy and data processing problems during the development of artificial intelligence systems.
So what is neural network? And what are the attempts of scientists to develop biological computing? What are the challenges facing the development of biocomputing? Many questions this article seeks to answer.
Neural network
In an article published in Nature Electronics, a group of scientists at Indiana University Bloomington in the United States of America were able to grow stem cells in dedicated conditions so that nerve cells that are small brain organs that act as small brains, contain millions of brain neurons, and when connected to a computer through a set of electrodes connected to them, these neurons can perform the tasks of artificial neurons of artificial intelligence, that is, turn into a data processor capable of analyzing some requests, just like the mechanism of work of neural networks of artificial intelligence.
According to the paper, this prototype was able to distinguish some sounds with an accuracy of up to 80% after being trained for 48 hours, which opens the way for biological computing to overcome the traditional problems of silicon chips, become a potential alternative to the energy used in data processing, and bridge the gap between biochemistry and algorithms, in what is known as brainware.
Brain software is a term that can refer to several different concepts in multiple fields, but in the context of artificial intelligence and computing, it usually refers to strategies and techniques that use models inspired by the human brain to develop advanced computing systems, that is, an attempt to develop systems that simulate the form and mechanism of the human brain, in terms of the form of neurons in it and the way it performs its tasks. This includes artificial neural networks, machine learning, and computing systems that attempt to mimic the neurobiological processes of the human brain.
Brain organelles such as those implanted in the experiment (three-dimensional structures of neurons resembling the human brain that are cultured or grown in the laboratory) are used to make calculations and address problems in the same way that the brain processes information.
The goal of using “brain software” in this context, that is, the development of advanced artificial intelligence systems, is to create computing systems that offer higher levels of efficiency and the ability to handle complex tasks in a way that simulates the efficiency and flexibility that characterizes the human brain, with less energy and a higher degree of efficiency, by taking advantage of some basic aspects of the brain, such as the ability to learn through trial and error, adapt to changing environments, and process ambiguous or incomplete data with high efficiency.
Brain software helps computer systems that use biological neural networks to learn quickly and continuously. Artificial neural networks usually need several stages to process and learn from data, and in this they use a very large amount of energy and computing, known as deep machine learning, but through brain software, biological neural networks can perform multiple tasks with the least amount of data in what is known as unattended deep machine learning, as they simulate some mental processes such as attention, memory, and even awareness of its initial levels, and this is done automatically in the human brain without the need to process a large amount of data, which scientists seek to employ in artificial intelligence.
In terms of applications, this brain software may contribute to creating more advanced hardware and software that is closer to the way the human brain processes information, opening the door to new and innovative applications in multiple fields such as robotics, data analysis, and smart systems in general. We have very small wearable devices such as watches, glasses, or even less, capable of processing a very large amount of data with minimal energy. Imagine, for example, a self-driving car or even a bike that does not need to learn or connect to any kind of data until it is trained on it. This is exactly what biological computing seeks, which is to create independent artificial intelligence capable of operating itself.
Reservoir Computing
In order to understand how brain software works, it is necessary to first understand Reservoir Computing, a concept in the field of artificial intelligence and neural computing that refers to a class of neural networks that use a static dynamic system (a set of static nodes) to deal with short-term memory and that allow some tasks to be done automatically without the need for in-depth analytics such as signal processing applications, pattern recognition and other automatic tasks.
In reservoir computing, the main part of the natural neural network (brain organ) is the “reservoir”, which consists of a random and dynamic neural network that acts as a storehouse of computational dynamics (in other words, a repository for data analysis). Here, the reservoir is not trained directly. Instead, only the part responsible for the outputs is trained, that is, the part responsible for decision-making in the neural network. The rest of the neural network is static and not trained, but its natural capabilities are exploited, just like the difference between the human brain and the forelock. The brain is the center for collecting and analyzing data, while the forelock is the part responsible for making decisions, lying and error, and it is able to make some decisions without reference to the brain through temporary short memory.
In reservoir computing, only the forelock is trained, that part of the neural network responsible for decision-making. It is faster and less complex than the training of traditional neural networks, which require large data and continuous adjustments to reach the correct results. The rest of the reservoir (the brain organ) is not trained because of its high dynamic capabilities that are intertwined across a network of millions of natural neurons.
Here, the brain organ acts as a three-dimensional network with a high density of cells, simulates the function of the human brain, and is linked to a computer system to send and receive information through a set of electrodes, allowing it to deal with little training data, low energy consumption, and learn from unclear data, in what is known as unattended deep learning for artificial intelligence.
Unattended Deep Learning
Unsupervised deep machine learning (sometimes referred to as unsupervised machine learning or inferential learning) is a type of machine learning that uses deep neural networks to analyze and learn the patterns inherent in data without the need for prior training or the use of tagged data (that is, data containing predetermined answers or classifications). The model (neural network) is left to discover the structures and patterns in the data itself and try to reach solutions.
Here, unattended deep machine learning tries to simulate how the human brain learns and processes information unattended. Patterns and structures are discovered without prior guidance, which is the essence of the unattended deep learning process. Its design is inspired by the biological neural networks in the brain, and similar principles are used to design systems that simulate normal neural functions.
Because the silicon element is unable to carry out the functions of unattended deep learning with high efficiency, and in addition to attempts at brain programming, scientists have turned to another chemical element (vanadium oxide), and when looking at this element under a microscope, we find that the shape of the molecules that it consists of fully mimics the shape of the human brain, and appears as if it is an interconnected neural network, which qualifies it to be a potential alternative to silicon to carry out unattended deep learning processes, taking advantage of the properties of non-linear dynamics that characterize vanadium oxide.
Biological Computing
As unattended deep learning technologies merge with brain software, a new type of computing, biocomputing, begins to emerge, a field of research that combines biology and computing to develop computing systems that function in ways that resemble biological processes. In other words, the field is interested in designing and building computing systems that use biological molecules, such as neural networks, DNA, and proteins, instead of traditional electrical and electronic circuits. Biological computing is capable of revolutionizing how information is processed and stored, and offers promises of new computing possibilities that cannot be realized with traditional silicon-based electronic technologies.
Although biological computing may be an important step for the development of advanced artificial intelligence systems, it is still in its infancy. There are still many challenges that must be solved first, such as: integrating the software interfaces of artificial intelligence systems to fully exploit the potential of “brain software”, in addition to the need for further development, especially in the field of generating and growing “brain organisms” with high efficiency, as well as the challenges and problems of maintenance and preservation in a natural environment. They are basically living neurons and need appropriate conditions to grow, multiply and interact to increase their computing capacity.
In the end, biological computing may seem to be a solution to problems that are hindering progress in artificial intelligence technologies, such as: energy and data processing, while at the same time offering a sustainable alternative that preserves the environment from electronic waste and from thermal and carbon emissions resulting from the use of traditional energy sources in the operation of artificial intelligence systems, so that we have intelligent systems that operate with human power, so that we have machines that are half human, and half of them are human machines.
Leave a Reply