What is AI? Definitions, History, Mechanism
January 17, 2025
In recent years, artificial intelligence has become an integral part of our lives. Many people are aware that AI stands for artificial intelligence, but they are unsure what it is.
In this article, we will explain how AI, which is currently receiving a lot of attention, works in an easy-to-understand manner even for beginners.
If you identify with any of these categories, this is the article for you. You will gain a better understanding of AI and its history after this article.
Artificial Intelligence (AI) is, as the name implies, an artificial brain with human intelligence.
The name was coined in 1956 by computer scientist and cognitive scientist Professor John McCarthy at the Dartmouth Conference, which was held at Dartmouth College in the United States.
Machine learning and other AI technologies are increasingly being used in applications such as translation, self-driving cars, and medical image diagnosis.
However, current artificial intelligence is incapable of understanding all aspects of human cognitive abilities, common sense, or emotions and performing all tasks.
It is a developing field in which various ideas are proposed to create a general-purpose AI capable of handling any challenge.
Many people confuse robots and artificial intelligence (AI).
A robot does what it is programmed to do, whereas an AI does not move unless it is given an external command. AI can also think and learn independently like humans, as well as quickly absorb human words and actions. It outperforms humans in terms of learning and processing ability.
AI is now being used in a variety of fields, such as the following:
| Categories | Examples of application areas |
| Computer Vision | Image Classification, Image Generation, Object Detection |
| Natural Language Processing | Machine Translation, Language Modeling, and Question Answering |
| Medical Care | Medical Image Segmentation |
| Methodology | Distributed Representations (Word Embeddings) and Representation Learning |
| Gaming | Video games and board games |
| Graph | Link prediction and node classification |
| Speech | Speech recognition and synthesis |
| Time Series | Time series classification and assignment |
| Audio | Music Generation and Audio Classification |
| Robot | Calibration and self-location recognition |
| Music | Music information retrieval and music modeling |
| Computer Code | Dimension reduction and program synthesis |
| Inference | Decision-making/Common sense reasoning |
| Knowledge Base | Knowledge Graph Causal Discovery |
| Hostility | Attack, Defense, and Hostility Text |
| Others | Recommendation Topic Model |
AI is also spreading in familiar fields. For example, many Apple products are equipped with “Siri.” Voice assistants such as “Google Assistant” have also become familiar.
Recently, air conditioners equipped with a human detection center have become popular. The reason why humans do not need to operate the remote control is because AI automates it.
Cleaning robots, such as the well-known “Roomba,” also use AI to collect information about the size of a room and obstacles.
According to the Japanese Ministry of Internal Affairs and Communications report, “IoT, Big Data, AI: New Value Created by Networks and Data,” there are three booms in AI research.
Since the Dartmouth Conference in 1956, AI has advanced to the point where it can conduct computer-based “reasoning” and “exploration,” allowing it to tackle specific challenges.
During the Cold War in the United States, machine translation using natural language processing gained popularity.
While AI at the time could prove simple hypotheses, it struggled to solve complex problems involving multiple factors.
The second boom began with the development of “expert systems,” which enabled the “representation of knowledge.”
An expert system is a computer system developed through artificial intelligence research that responds to specific problems as if it were an expert. It lacks a self-learning mechanism, so it is necessary to anticipate situations and record countermeasures and judgments.
Because it was necessary to write down all of the required information, the amount of knowledge that could be applied was limited to a specific field, so its popularity declined around 1995.
The third AI boom, which has continued from 2000 to the present, has enabled AI to acquire knowledge using “big data.” This is called machine learning.
In 2006, “Deep Learning” was introduced. “Deep Learning” is a mechanism that allows a machine to automatically extract features from data without human knowledge by preparing a sufficient amount of data. The emergence of Deep Learning sparked a boom in the AI field, and it has become even more popular.
However, to spread Deep Learning to society as actual products and services, environmental facilities are often required, and there are still many challenges to overcome.
A neural network is a combination of mechanisms inspired by the structure of the human brain (neurons).
A neural network consists of three layers: an input layer, an output layer, and a hidden layer. The final result can be calculated by inputting data into the input layer and inputting neurons into the output layer.
Originally, there were no intermediate layers, also called hidden layers, which meant that it was not possible to solve linearly inseparable problems. However, by having multiple intermediate layers, it became possible to solve more complex problems.
A neural network is a combination of mechanisms inspired by the structure of the human brain (neurons).
A neural network consists of three layers: an input layer, an output layer, and a hidden layer. The final result can be calculated by inputting data into the input layer and inputting neurons into the output layer.
Originally, there were no intermediate layers, also called hidden layers, which meant that it was not possible to solve linearly inseparable problems. However, by having multiple intermediate layers, it became possible to solve more complex problems.
A genetic algorithm uses the mechanisms of biological evolution.
The idea was inspired by a phenomenon that occurs during the evolution of living organisms: “The stronger individuals that adapt to the environment survive, while the weaker individuals that cannot adapt are selected out.”
Choosing the optimal answer from many options is called “multi-objective optimization.” The more objectives you want to meet, the more difficult and time-consuming it becomes to make a selection, but by using a “genetic algorithm” in such a case, you can quickly derive the optimal solution.
The expert system, which was introduced earlier, is also an important AI algorithm.
We listen to experts about possible situations and how to deal with them, as well as their judgments and predictions, and we use that to set rules.
It then uses logic to determine which data it believes to be the correct answer and clearly explains to the user why and what the thinking behind choosing that answer was.
What makes it different from other AI algorithms is that it does not have a mechanism for self-learning. Also, while the more rules there are, the more accurate it becomes, it has a weakness in that if there are too many rules, it becomes difficult to maintain consistency between the rules.
Machine learning refers to making a machine capable of dealing with various problems by having it learn data. It is broadly classified into three types: “supervised learning,” “unsupervised learning,” and “reinforcement learning.”
Supervised learning refers to teaching artificial intelligence in one direction by giving it a set of example problems and model answers. It generally requires a large amount of data, and the neural network itself uses the given data to determine whether the output result is correct or not.
Unsupervised learning is a method in which AI learns by accumulating data based on its activities without the need for model answers. While it does not require a large amount of data, it does require an “environment in which it can learn correctly.”
Reinforcement learning is a method of finding optimal actions and values through repeated trial and error in the environment in which the AI is placed. It is necessary to make the AI fully aware of its actions and current situation. Reinforcement learning has a wide range of applications and is highly effective when the object to be learned cannot be modeled.
Deep learning is a type of machine learning that uses multi-layered neural networks. Representative examples include CNN (Convolutional Neural Network) and RNN (Recurrent Neural Network).
Let me explain each one.
CNN (Convolutional Neural Network) is used for image recognition and motion detection and consists of a “convolutional layer” that extracts image features and a “pooling layer” that analyzes the features. It enables quick identification.
It is a neural network with an autoregressive structure. It can handle variable-length time series data and is used in speech recognition, video recognition, natural language processing (NLPs), and other applications.
Subscribe to our newsletter for updates on DEHA news and highlights in the IT industry today.
