Types of AI. Brief overview of ML.
Types of artificial intelligence
There are two main types of artificial intelligence. The first one is applied. Applied artificial intelligence includes systems that can perform a small set of tasks and that can learn performing these tasks. For example, robots that trade stocks or robots that control autonomous vehicles belong to this category. Such robots can make decisions when it comes to their narrow fields, but are useless in other fields.
The second category of artificial intelligence is general AI, meaning robots that in theory can handle tasks in a variety of fields. General artificial intelligence is less common than applied AI.
Machine learning from 1959 to now
The term “machine learning” came into existence in 1959. Arthur Samuel was a scientist who has invented the term. It was Samuel who came up with the idea that instead of teaching the machines everything there is to teach about the world, it could be possible to teach them to learn on their own. Samuel has created software that allowed a machine to play checkers with a person, which was one of the first successful uses of artificial intelligence in the human history.
The importance of machine learning has been increasing since the invention of the Internet, because currently humans have access to more data than any person or group of people could ever process.
One of the biggest breakthroughs in the development of machine learning has been the creation of neural networks. A neural network is a set of algorithms that try to mimic the algorithms created by human brains for human behavior, including learning. One of the main tasks of such algorithms is pattern recognition. For example, if a person touches a hot item on a stove, the person is likely to become very careful with what he or she touches on the stove. This means that an event led to the creation of a pattern and now the person acts according to this pattern. This is what neural networks are about. They get data and then interpret the data using specifically designed perception algorithms. The patterns the machines recognize can be images, sounds or text.
Neural networks function similarly to neurons in the human brain. These networks consist of multiple parts (or “neurons”), all of which have the functionality that allows them to perform tasks related to the action they are trying to accomplish just like a person would do. Examples of such tasks are recognizing information or a pattern, failing to see commonalities and treating something as new information, matching different pieces of information together in an old or a new way, and answering questions about information or a relationship between different pieces of information. Each part of the network is capable of making a decision and then informing the neurons that surround it about the decision. The neighboring neurons can accept the decisions as input and process them further. Because neural networks can change their approach to information and are able of processing large volumes of data, in a way they function similarly to human brains.
Practically speaking, neural networks are a way to sort and classify different types of data. It is convenient to think about them as a layer on top of the layer with raw data. Examples of neural networks at work include facial recognition, recognition of emotions on people’s faces, identification of specific objects in images and videos, such as stop signs, weapons, and so on, detection of voices and speakers, classification of messages as spam.
Essentially, machine learning neural networks use probability to make decisions and then use the information about these decisions to improve on future decision-making. It is the addition of this second part, use of decisions to make decisions, is what makes machine learning different from the algorithms in the past. A machine first makes decisions. Then, it learns whether these decisions were right or wrong. Finally, it uses this information to modify the approach it will use in the future to make the same types of decisions.