Edited by Pier Giuseppe Giribone

Artificial Intelligence (AI), Machine Learning (ML) and Deep Learning (DL) are concepts
closely related to each other: Deep Learning is a type of Machine Learning and Machine Learning is a type of Artificial Intelligence.

This definition, although not absolute, is accepted in these terms by the scientific community.

Artificial Intelligence is a general concept that encompasses a variety of technologies, including Machine Learning, which, in turn, is made up of specific techniques, including Deep Learning.

It is therefore not a mistake to compare Artificial Intelligence, Machine Learning and Deep Learning to Matryoshka dolls: each is essentially a component and is included in the previous term.

Artificial Intelligence (AI) is the broader term used to classify machines that emulate human intelligence. It is used to predict, automate, and optimize tasks that humans have historically performed, such as speech and facial recognition, decision-making, and language translation.

The scientific classification provides for three main categories of Artificial Intelligence:

Artificial Narrow Intelligence (ANI), Artificial General Intelligence (AGI) and Artificial Super Intelligence (ASI).

ANI is also known as weak AI or narrow AI. This technology can perform a specific task by learning from a set of data consistent with that purpose. Some examples of ANI include facial recognition, automatic chess playing, and self-driving cars.

Narrow AI has limited capabilities, which is why it's classified as "weak." Narrow AI lacks any form of genuine intelligence and therefore cannot be compared to human intelligence. However, its value lies in achieving specific goals efficiently and effectively (e.g., industrial automation engineering applications).

A narrow AI system is typically trained on a specific dataset (called the training set) so it can develop an understanding of the problem it faces and solves. Once it achieves its goal, the AI ​​can use its knowledge of the decision-making process to predict outcomes and execute the action it was designed for.

For example, if you want to train a narrow AI system to identify handwritten numbers in text, you need to train it with a dataset containing a large number of examples of numbers written by different people. After training, ANI will be able to identify characters in new texts.

As we consider stronger forms of AI, such as AGI and ASI, the incorporation of Human-like behaviors become more important, including, for example, the ability to interpret tone and emotion. Chatbots and virtual assistants, such as Siri, although highly advanced, are still considered examples of ANI.

Artificial general intelligence (AGI) is the complete representation of human cognitive capabilities that, when faced with an unfamiliar task, can replace the human approach in finding a rational and predictable solution. Therefore, the goal of an AGI system is to perform any cognitive task that a human is capable of within the realm of rationality.

Definitions of AGI vary, as experts in different fields define human intelligence from different perspectives. Computer scientists often measure human intelligence in terms of the ability to achieve goals. Psychologists, on the other hand, often define general intelligence in terms of adaptability or survival.

Despite accelerating technological progress, we are still far from AGI as it is commonly perceived, which in fact includes the following abilities: abstract thinking, common sense, interpretation of cause and effect relationships and the ability to transfer learning to other agents.

At a science fiction level, an ASI (or super intelligence) system would even have to surpass human cognitive capabilities.

Specific learning methods for ANI systems include numerous machine learning algorithms, which enhance the knowledge acquired by the system through data. Depending on the specific objective and the size of the available sample, the procedure deemed most appropriate for the purpose is chosen.

If a large amount of significant data is available and a very in-depth level of knowledge of the system is required, despite a significant training time, deep learning models can be used.

Having clarified the terminological differences between Artificial Intelligence, Machine Learning, and Deep Learning, the next article in this series will focus on the difference between a traditional algorithm and one that implements machine learning paradigms.