Artificial Intelligence. Past, Present and Future (V)
Many of today's artificial intelligence successes are due to a set of tools, heuristics and models defined or inspired primarily by nature, and some of the conceptual discoveries John McCarthy referred to began to occur in the 1990s. Advances in neuroscience, new computer models, heuristics,...
Artificial intelligence begins to be considered as an emulation of human behavior. But instead of programming machines to perform a task explicitly, a different approach is used. Machine learning began to be exploited in more and more fields.
From the year 91 with the creation of the Internet and from now on with the explosion of available data, interest in AI techniques has been reactivated. It is based on the experience understood as data, to imitate the functioning of the nervous system. Machine learning models are defined that, using the large amounts of information available, manage to learn and offer solutions that are not approachable with a classic approach.
An algorithm is, in general terms, a series of organized steps that describe the process that must be followed to solve a specific problem. Instead, genetic algorithms propose to use the laws governing natural evolution to give that solution to the problem posed. In the 1970s, with John Henry Holland, one of the most promising lines of artificial intelligence emerged, that of genetic algorithms. This line of solutions started in the 19th century with Charles Darwin!
Another example of nature-based solutions is found in our brain. Today, concepts such as “Machine Learning” are widespread and can be implemented using various techniques, but one of the most common are neural networks, recently evolved and renamed with the concept of Deep Learning. It's just that machine learning mode that's revolutionizing the expectations of AI.
Examples of genetic algorithms and neural networks
Let's see a series of videos where the process is shown
of learning using AGs and Neural Networks techniques.
In the following video you will see a graphical representation of a neural network that is capable of recognizing numbers. An input layer is shown as a set of pixels representing the image (equivalent to our retina) and a layer of output neurons. There is a set of hidden layers in which neurons are reconfigured during the training phase to give a correct solution.
3D simulation of neural networks
The following video shows a recreation of an ecosystem with various life forms created randomly. The ones that adapt the worst disappear. They begin to develop smarter behaviors, and create a true ecosystem.
Ecosystem with various forms of “life”
A genetic algorithm learns how to shoot and self-improvement to be more effective. In the first generations they shoot randomly, but as time passes they develop more advanced shooting strategies.
A genetic algorithm learning to shoot in a game
Genetic algorithm learning to jump an object. A few rules of movement are defined and from there the system learns how the object should jump.
Genetic algorithm learning to dodge an object