Artificial neural networks are information processing systems based on natural nervous systems such as the human brain. They consist of many individual artificial neurons which are interconnected, can solve problems together and have the ability to learn. A recurrent neural network (RNN) particularly resembles the human brain because it contains feedback loops. These allow signals to travel both forward and backward, creating a more complex and less stable system. The recurrent neural network is dynamic and, after each input, the state of the system continually changes until it reaches an equilibrium.
Human brains can be described as biological recurrent neural networks. An artificial recurrent neural network shares the brain's ability to learn processes and behaviors. This is not possible with methods of traditional machine learning. In common with other types of neural networks, a recurrent neural network is especially good at recognizing patterns and spotting trends. A number of potential uses have been found for this kind of computational model, including recognizing disease from medical scans, modeling body systems, speech and handwriting recognition and stock market forecasting.
Typically, a recurrent neural network will be used to solve a problem in which it is known, or strongly suspected, that there is some kind of relationship between the data input and the unknown output. The network will be trained, or will train itself, to work out that relationship and provide a possible output value. A recurrent neural network is able to handle large complex problems in which some values are missing or corrupted. Its ability to learn from example makes it powerful and flexible, and removes the need to create an algorithm for each specific task.
Recurrent neural networks can be described as non-linear statistical data modeling tools. The presence of feedback loops means that they are adaptive systems, able to respond to change. A recurrent neural network used in the field of robotics can enable a robot to learn from experience, allowing it to make decisions about which direction to take in order to reach a target. It might even be possible to develop curiosity in robots by making it rewarding to focus on things that are unpredictable, though not completely random. Some scientists believe that consciousness itself is a mechanical process and that it might be possible to develop a conscious form of recurrent neural network one day, although this would lead to ethical questions about the rights of robots and machines.