Did you know that within the developments achieved by science in the last century, it has been possible to create an alternative for machines to get knowledge on their own?
Human beings have always thought about the possibility of building machines that think for themselves, are able to speak, that can solve different problems and, above all, that learn autonomously. For example, Homer, around the 8th century BC, describes the maidservants of the god Hephaestus as beings made of gold, endowed with intelligence and capable of performing both physical and intellectual work. In another legend, we find the story of Talos, a gigantic bronze robot whose task was to protect the city of Crete and prevent it from being attacked by pirates. References of this nature imply that for the Greeks, the gods were capable of creating machines that possessed a certain degree of intelligence; that is, artificial intelligence.
However, the above is not the only reference in the literature on the creation of artificial intelligence systems where the thing that is built has the ability to learn by itself. In the literature, the term “automaton” is addressed as a machine capable of performing different tasks autonomously and having the ability to solve some particular problems presented to it. In the last century, this possibility has been explored from all horizons within Science Fiction literature and cinema. Therefore, it has been contemplated the creation of machines that acquire self-awareness and decide to dispense with humanity , making this type of narrative an apocalyptic affair.
But beyond Science Fiction, do you know that within the developments achieved by science in the last century, it has been possible to create an alternative for machines to get knowledge on their own? This type of development in artificial intelligence is known today as “machine learning” or “Machine Learning”.
Machine Learning is a form of artificial intelligence that allows an automated system to be able to “learn” from experience, very similar to the way we humans do; that is, it can learn to perform certain tasks from a compilation of data that it constantly receives, instead of doing so through explicit programming. This makes it improve its performance progressively, according to the amount of data it is fed with, and the more data it receives, the better it can perform.
Let’s see an example: until very recently, climate models were made manually from data received by meteorologists over the years and made predictions about the state of the weather in some regions with a certain degree of accuracy. However, much of this data was susceptible to errors and usually took several days to produce results. Today, thanks to the large amount of information available from historical records and that which is generated in real time, there are automated programs that are capable of making an assessment of the climate situation in seconds based on various factors, giving increasingly accurate real-time forecasts of both short- and long-term climate variations.
How does it work?
Machine Learning can make use of two learning techniques by which it can find patterns in the midst of the data load that help it make better decisions and predictions: Supervised and Unsupervised learning.
Through supervised learning, classification and regression techniques are applied to develop predictive models. While the first ones are focused on classifying data according to certain categories, the latter can make use of available data to make predictions about certain activities or continuous responses, such as changes in temperature or air currents, or even variations in the consumption of resources such as energy.
Unsupervised learning is the use of hidden patterns or intrinsic structures found in the data and is used when there is a massive amount of unclassified data. Under this method, anomalies that are not visible can be found and data can be grouped according to their similarity (clusters), which is very useful for solving real-world problems, such as finding customers with common interests based on their purchase or search history, detecting anomalies or grouping documents in a system.
