Artificial Intelligence, Machine Learning and Deep Learning

Nowadays, when Artificial Intelligence, Machine Learning and Deep Learning are compared to human intelligence, these terms are often confused with each other. Although artificial intelligence and machine learning are often perceived as synonyms used interchangeably, these words contain important differences.

Experts working in the fields of artificial intelligence, machine learning and deep learning have not yet reached a consensus on these words. In this field, new concepts are opened up for discussion every day. With the development of technology, they are now included in our lives and become a part of us. In general, artificial intelligence is an older concept than machine learning, and machine learning is seen as a sub-branch of artificial intelligence.

In 1950, Alan Turing brought up whether machines can think in an article he published. With the famous Turing Test he proposed, it is possible to distinguish whether a machine is intelligent or not. If a human cannot distinguish whether a human or a machine is behind an interaction, that machine is an intelligent, thinking machine. However, the real father of artificial intelligence is John McCarthy, who organized an academic conference on the subject in 1956. At the end of the conference, the common views of the participants were that the studies on artificial intelligence should be taken to an advanced level.

Topics included natural language development, image recognition and classification, and machine learning, which have become popular these days. However, their real development began to be seen years later, today, due to existing technologies.

Chronology of artificial intelligence, machine learning and deep learning (Source: Nvidia)

In the last 10 years, artificial intelligence has started to take place in our lives, out of a concept used in science fiction movies. Developments such as IBM's artificial intelligence Watson, which won the quiz, and Google AI, which defeated the world champion in the GO intelligence game, have put artificial intelligence on our agenda today.

All of today's biggest technology companies are investing in artificial intelligence. We are now experiencing more artificial intelligence, even machine learning, on mobile phones, social media, search engines, and more.

What is Artificial Intelligence?

As we mentioned above, if we cannot distinguish whether there is a machine behind an interaction, and if it responds to people with special abilities, we are faced with an artificial intelligence. However, artificial intelligence also has weak and strong ones.

Weak artificial intelligences are formations that do exactly what you describe/program, do not go beyond them, do not change the result you will get, and offer opportunities in the first years of technology. However, strong artificial intelligences are formations that can reach the given goals in different ways and make things easier by making comments on the data with algorithmic calculations.

To give an example from the famous Atari game breakout, you watch a game where the ball has to break bricks in a weak artificial intelligence algorithm, breaking bricks one by one since the bottom. However, strong artificial intelligence can finish the game by evaluating the data in its hand and sending the ball to the top of the bricks without moving the stick, thus minimizing the risk of losing the ball. In other words, he added comments to reach the goal in a better way. Thus, the machine developed a strategy that was not included in the software and programmed itself to get better results than a human being in just 2.5 hours.

However, we should consider the following; The machine does not see the game as bricks, balls and sticks as we see it, they have different meanings for him, for him these are just variable data in order to get a higher score.

 

What is Machine Learning?

Machine learning is considered a kind of artificial intelligence that can reveal results even for which it was not programmed.

In 1959, Arthur Samuel developed machine learning: He defined it as "the ability to learn results for which machines are not specifically programmed". Arthur Samuel has made a checkers game that can work in a computer environment, learn from his own mistakes and thus improve himself.

Machine learning, like artificial intelligence, has been idle for years. It did not become popular until the 1990s when data mining began to be used. Data mining is the application of algorithms used to uncover similar motifs with similar sequences in the data it holds. Machine learning does the same, but data mining goes one step further and changes the behavior of the program about the information it learns.

One of the applications that made machine learning popular is image recognition. This app needs to be trained first. In other words, similar images have to be shown thousands of times before the machine can learn what the image is. Thus, by recognizing similar sequences, similar motifs, and similar pixels, the machine can now identify what those pictures are. By distinguishing different pictures such as dogs, cats, trees, and houses, it can now identify the common points of the same type of pictures.

Many applications running online now use machine learning. For example; Facebook decides what to show you on the timeline with this algorithm, Amazon decides which products to recommend, Netflix decides which movies to recommend, again with machine learning. Based on the available data, it predicts what you will like and decides accordingly.

In fact, machine learning has become so intertwined with statistics, data mining, and predictive analytics that many experts believe it is necessary to separate it from the concept of artificial intelligence. To be realistic, machine learning doesn't particularly need the features used in AI. However, on the other hand, many people use the concept of machine learning more because it is not as scary as the concept of artificial intelligence. Machine learning is actively working according to artificial intelligence and we have already started to use it in our daily life.

As a result, the concept of artificial intelligence and machine learning has become so intertwined that it has begun to be presented in combination in many applications released today. For example, personal assistant applications and bots host many artificial intelligence features and machine learning features at the same time.

What is Deep Learning?

The more data there is, the better artificial intelligence features will be revealed. Things will get more complex, as they get more complex there will be shifts from artificial intelligence to machine learning. As it gets more complex, the transitions from machine learning to deep learning will begin. The more data you have, the better your system will run.

While machine learning processes in a single layer, deep learning processes in many layers simultaneously. A group is trying to reach the result in a single transaction by using machine learning algorithms at the same time.

For example; we need to separate a picture of a banana and a picture of an orange. In machine learning, we were trying to introduce the experience of human beings to the machine through parameters. Well, if it's orange, it's probably orange, and yellow is like a banana. If it's round, it's probably orange, if it's arc-shaped, it's probably a banana, so we had to define many parameters.

However, deep learning can learn this different on its own. By just showing the orange and banana pictures to the deep learning system, he creates his own rules, realizing that color and shape are the main distinguishing features in order to reveal the differences. Thus, without the need for basic human abilities, it can perform its operations by creating its own decomposing abilities.

Summary

If you are confused by these terms, know that you are not alone. Computer experts continue to debate these concepts, and they probably will. To make a short summary;

Artificial Intelligence emerged in the 1950s and is the ability of machines to perform certain operations as well as humans. While weak artificial intelligences only perform what you have programmed, strong artificial intelligences are systems that can improve what you have programmed by making algorithmic calculations and learn from mistakes.

Machine learning emerged in the 1980s and started to become more popular with the use of data mining. They are self-training systems that can make better determinations than you, reveal what you have not programmed, by making simulations with the data and parameters you have presented.

Deep learning is a system that started to be used in the 2010s, makes calculations used in machine learning in many layers, not in a single layer with a sea of big data, at once, discovers even the parameters that you need to define in machine learning, and maybe can make evaluations with better parameters.

If we call all these processes artificial intelligence in colloquial language, this system that learns faster than humans and produces more qualified results becomes important. Putin's rhetoric, "He who is a leader in artificial intelligence, becomes the ruler of the world" is not something to be taken lightly in the light of this information. It is also necessary to consider the determination of Elon Musk, "Artificial intelligence can cause the 3rd World War".