The achievement of AI (Artificial Intelligence) development has been remarkable since the beginning of the 2000s, and recently I hear it in the news and so on. When it comes to AI, it tends to draw attention to the great achievements of recent years, but in fact, the research and theory on which it was based have been much elder. The oldest dates back to ancient Greek myths. This time, I would like to introduce the basic knowledge of AI for those who do not understand what AI (artificial intelligence) is in the first place. In addition to that, the important achievements from ancient times that are indispensable to talk about modern AI are summarized as “History of AI”.

Voice assistant function “Google Duplex” which attracted attention

There are much AI-related news and TV shows, but recently we have talked about Google Duplex, a voice assistant feature developed by Google.

What is AI (Artificial Intelligence)? We thoroughly explain the history from ancient myths to the latest AI!

The achievement of AI (Artificial Intelligence) development has been remarkable since the beginning of the 2000s, and recently I hear it in the news and so on. When it comes to AI, it tends to draw attention to the great achievements of recent years, but in fact, the research and theory on which it was based have been much elder. The oldest dates back to ancient Greek myths. This time, I would like to introduce the basic knowledge of AI for those who do not understand what AI (artificial intelligence) is in the first place. In addition to that, the important achievements from ancient times that are indispensable to talk about modern AI are summarized as “History of AI”.

A comprehensive guide to the IT industry is an industry information media operated by the Internet Academy of Japan’s first Web-specialized school. We introduce the latest industry information to beginners in an easy-to-understand manner.

Voice assistant function “Google Duplex” which attracted attention

There are much AI-related news and TV shows, but recently we have talked about Google Duplex, a voice assistant feature developed by Google.

Google Duplex is a function that allows AI to make a telephone reservation for a restaurant or hairdresser on behalf of the user. This is very useful if you can not make an online reservation or if the user can not make a direct call. Google Duplex surprised people with the natural nature of the conversation. The above video shows how Google Duplex calls the actual hairdresser to make a reservation.

Google Duplex not only speaks very fluently but also uses words such as sumo wrestling and gap filling very naturally, so it is almost impossible to discern which is human and which is AI by just listening to the voice. There is no doubt that the spread of Google Duplex will bring about significant changes in our lives. In recent years, AI technology has progressed in various fields, including such voice assistants.

With AI (Artificial Intelligence)

AI (Artificial Intelligence) is an acronym for “Artificial Intelligence”. The term “Artificial Intelligence” is the first term used by American computer scientist and cognitive scientist John McCarthy at the 1956 “Dartmouth Conference” and is defined on his homepage as:

It is the science and engineering of making intelligent machines, especially intelligent computer programs.

Professor From John McCarthy’s homepage ” What is AI? / Basic Questions “

“It’s science and engineering that makes intelligent machines, especially intelligent computer programs.”

Intelligence here refers to human intellectual ability. In other words, it refers to reproducing on a computer the intelligence necessary to carry out the activities and tasks that humans originally perform, resembling those of humans. In the past, computers have been required to think and learn in the same way as humans. It has been applied gradually to our familiar places, and research is being conducted in various fields such as natural language processing such as speech recognition and translation, and image recognition.

With machine learning (machine learning)

Learning is necessary for computers to make guesses as AI. Just as AI itself does, we must obtain the necessary judgmental material for guessing through knowledge and experience. We will construct an algorithm to solve the next problem by deriving regularity, expression method and index from the collected data. The method of learning this data is called machine learning.

Machine learning (machine learning) can be broadly divided into three types: “supervised learning,” “unsupervised learning,” and “reinforcement learning.” Next, I will introduce these differences one by one.

What is supervised learning?

“Supervised learning” is a way of presenting problems and answers simultaneously to the machine as if the teacher taught the students how to solve the problems. It is efficient to have machine learning in this way if the answer to the problem is clearly defined.

For example, if you want the machine to learn “the face of a person,” analyze the face picture of the “person” and the answer “this is the face of a person” simultaneously. Even if it is simply a “face of a person”, it varies from individual to individual, but the machine grasps features while analyzing tens of thousands and millions of pieces of such information, “Why this photo is on the face of a person We will learn that it will be classified. A method that allows you to analyze problems and answers simultaneously and improve the accuracy of problem-solving in this way is called supervised learning.

What is unsupervised learning?

In contrast to “supervised learning”, the method of learning when the problem is given but the answer is not given is called “unsupervised learning”. In this case, there is no teacher who will tell you the answer, so it is suitable for making the machine read a large amount of data for analysis, classification, guessing and new discoveries.

This “unsupervised learning” is often used in “data mining” to reveal the features of countless data whose law is not known. “Data mining” is a technology to extract information that has not been clarified until now, especially useful information, from a huge collection of data. Knowledge such as statistics and pattern recognition is also used during data analysis. In addition, there is “semi-supervised learning” as a method of machine learning that shifts to “unsupervised learning” and analyzes a huge amount of data if learning is started by “supervised learning” and the tendency is grasped to some extent.

What is Reinforcement Learning?

“Reinforcement learning” is a learning method in which the machine itself performs trial and error to maximize the reward. This method of learning is also a kind of “unsupervised learning”, considering that only the problem is given and the answer is not given, that is, there is no teacher. However, the difference from “unsupervised learning” is that the machine itself repeats trial and error so as to obtain better results, rather than merely classifying and inferring data. In order to get the machine to do high-scores in games such as Go and Shogi and to perform complex and advanced things such as holding objects deftly with an arm, it is effective to have the machine try and error.

What is needed to guide the machine to maximize the final effect during repeated trials is to give “reward”. If you happen to be rewarded by chance, even if you’re repeating a nonsensical task at the beginning, the machine will get a hint that “you could get paid this way”. By repeating these processes, their accuracy will gradually increase, and in the end, they may be more powerful than human beings. The computer Go program Alpha Go, which quickly became famous after defeating the world champion of Go, was able to become so strong by “Reinforcement Learning.”

LEAVE A REPLY

Please enter your comment!
Please enter your name here