What can AI (artificial intelligence) do?

AI (artificial intelligence) is now gradually entering our society and life as new products and services. We will explain the current state of AI, such as what is AI and what types of AI are available.

What kind of Technology is AI?

AI (Artificial Intelligence) is a system that imitates and reproduces the abilities and activities of the human brain, such as recognition, thinking, and learning, using computers and other devices.

However, there is still no clear definition of AI, and artificial systems that perform intellectual activities that are completely equivalent to those of humans have not yet been created. Despite this, AI is currently attracting a great deal of attention because the emergence of new AI-related technologies such as “deep learning” and “machine learning” has paved the way for commercialization. .

The recent movement that began around 2011 is called the “third AI boom.” Prior to that, the first boom occurred in the 1950s and 1960s, and the second boom in the 1980s, both of which failed to produce significant results.

It was in February 2011 that IBM’s “question-answering system” called Watson appeared on a popular quiz show in the United States and won against humans. Watson is considered to be similar to AI, but it has functions such as voice recognition and image recognition, and can search for answers by referring to text on the Internet by itself.

Artificial Intelligence

Artificial Intelligence

This is one way to train computers to perform human-like tasks such as speech recognition, image identification, and prediction. In 2012, Professor Hinton of the University of Toronto won an overwhelming victory in a global artificial intelligence image recognition contest using a system using deep learning. After that, in 2015 and 2016, Google’s “AlphaGo”, which utilizes deep learning, defeated a top Go player and created a sensation.

Deep learning is a method of “machine learning” that reproduces human learning ability on a computer. Machine learning requires a variety of data as input values. The data is processed by the machine learning algorithm in the AI ​​program, and output values ​​such as classification, recognition (discrimination), and prediction (regression) are derived while learning by itself. Deep learning, on the other hand, is the development of a mechanism called a “neural network”, which is a mathematical model of neurons in the human nervous system, and related technologies. In addition, deep learning is said to be particularly effective as an analysis technology for big data.

What AI can do?

Let’s list the types of “what AI can do” that are currently under development or have already been realized and put into practical use.

1. Sentence comprehension AI

It understands written texts, translates and summarizes them, and creates articles using given data. AI has already been used in several media for articles on themes such as financial information and sports match results. Such text understanding by AI is being advanced by research in the field called “natural language processing”.

2. AI for speech understanding

With Google Assistant, Siri, and chatbots, you can have voice conversations with virtual assistants. Such speech understanding (speech recognition) is also one of the technologies related to natural language processing. Already, smart speakers have been put into practical use for voice conversations and home appliance control. It is also said that in the near future, it will be possible to recognize emotions contained in speech.

3. AI for image recognition

Among AI-based image recognition systems, facial recognition systems are becoming more and more practical. Face recognition using deep learning can be used to recognize faces even when the face is tilted or when the face is wearing sunglasses, and is expected to be used in surveillance cameras.

4. Inference AI

Inference is finding answers to new data based on past knowledge. AI learns through training to make this inference. Currently, AI has come to show reasoning abilities that surpass human ability in games such as Othello, chess, shogi, and Go, where the rules are fixed.

5. Machine control AI

Many companies are working to fully utilize AI in the control of automobiles, industrial robots, sensors, servomotors, construction machinery, and more. Machine control by AI differs from conventional automation by computers in that AI “understands” the meaning of data as knowledge and performs optimal control while “learning”.

There are various types of AI, and the technology has already begun to be used in various fields. And in the very near future, AI will evolve to be able to do even more things. In the next few years, we cannot take our eyes off the trend of AI technology.

AI can be divided into several types according to the purpose of use and functions.

Specialized type and general type

AI that specializes in a specific area and demonstrates its capabilities is called “specialized AI”, and AI that solves various problems in different areas is called “general-purpose AI”.

All of the general AIs that currently exist, such as autonomous driving systems, image recognition, shogi/chess/go, and AI that recognizes voice and text words, are specialized types. On the other hand, general-purpose AI is said to be able to autonomously perform various intellectual tasks in a single system like humans, but there are still many problems to be solved before such an AI appears. I’m here.

Strong and weak AI

According to University of California professor and philosopher John Searle, a “strong AI” is an AI that “has human-like mental abilities and behaves like a human.” It has a brain, spirit, and consciousness similar to those of humans, so to speak, it is like Doraemon.

AI other than this strong AI is “weak AI”. All existing AIs are philosophically weak AIs. Weak AI has no self-awareness and is nothing more than an imitation of human power. However, they are better at things like the ability to work more accurately, quickly, and with more patience than humans.

Leave a Reply

Your email address will not be published.