When talking about artificial intelligence(AI), many people think that neural nets should be as intelligent as the human brain. Some people don’t even imagine how widely they are used in their life. Let’s narrow us to something more simple and understandable.
The most exciting usage of AI is robotics. Today isn’t tough to build a simple robot with a few sensors and a couple of motors. But the harder part is to make it react to the real-world as you expect it to be. When programming the robot brain, there is checking sensor states and reacting to particular circumstances like hitting the wall, line on the ground, etc. But more sensors robot has more complex reactions may be. Programming of more complex systems may be painful without using simple neural networks.
In neural networks, we have two subjects: “knowledge” and “learning.” This means that intelligent systems have some knowledge, or so-called experience and ability to learn and improve. Let’s have an example: Suppose we have a T form maze. In this maze, there is a mouse. On one side of T, there is electric shock and on another side is cheese. During some tries, the mouse will learn which side to choose to avoid electric shock and get cheese. You see – we have touched both subjects: learning and knowledge. On the first try, the probability of choices ara 50×50 percent, speaking of science, but probability grows due to experience gained after repeated tries. Everything seems fine until we face hardware – a robot. How to make robots feel hunger, anger, thirst, pain, satisfaction?
We, humans, have these standard reflexes on birth brought by genes. Using them, we grow and learn. Once you burned a hand in the oven, you will always remember how painful it is and try to avoid the heat. With artificial intelligence, there is almost the same, just at a different level. In the modern world, AI is taking the first steps in understanding how our consciousness works, how neurons interact, and how the brain works.
In digital electronics, neurons could be interpreted as multiple-input AND, OR, or XOR elements. But in reality, the neuron is an analog element with multiple inputs with different sensitivities. Some of these input signals define the activity of neurons. The output signal may be processed as a result or forwarded to another neuron’s input.
In general neural network is a set of interconnected elements where each of them has its own input signals and outputs some resulting signal. For instance, a simple robot platform:
Mathematically everything is possible to describe with formula:
The function may be any logical algorithm, finite state algorithm, or simply as a set of operations of any programming language. Understanding that algorithm isn’t a reaction to one or another input, but describes a method used by a neural network. Teaching neural networks is done by examples: network inputs are affected by some actions, and output signals are compared to our expected reaction. If it differs – then we get so-called an “error.” To reduce this error, sensitivities of inputs are reduced so that error would be minimal. This process is repeated many times until the network reacts as expected.
Some science has to be done while choosing the right structure of neural networks. If the neural network is too small, it will be ineffective, and it won’t be able to learn what you want. If there will be too many neurons, then learning can take much more time than expected.
After initial training – a neural network may be used to control a robot platform. It can learn by itself, reacting to real-world objects. For instance: it can learn to find the most effective way to turn to light with minimal collision number and so on…
This is a very narrow area we have touched on in this article. Building effective neural networks require some knowledge and experience. But such simple solutions can be a good starting point.
Looking further, many more interesting things touch neural networking like genetic algorithms. They are used to copy the real-world example. It acts as the population of several neural networks and can duplicate and form an effective neural net that is most effective for particular purposes. But this is early to talk about artificial intelligence as it is more likely artificial reflex.