Narrow AI (特化型AI) is AI that makes judgments only on a specific task, such as classifying images or translating, and is also called weak AI.
Even when it outperforms people on one task, it cannot extend its abilities on its own to other tasks it has not learned. The AI in practical use today is all of this type.
AnalogyIt is like a chef at a sushi restaurant who has mastered sushi alone. The sushi is first-rate, but the chef cannot cook anything else.
How to rememberThe name says it: narrow means the range of tasks it can handle is small.
Exam points
ExamWeak AI (弱いAI) is a mechanism that outputs judgments on a set task; it does not grasp the meaning of a question and think about it the way a person does.
Mix-upThe "weak" in weak AI refers not to low performance but to the narrow range of tasks it can handle.
General AI (汎用AI) is AI that aims to make judgments on a wide variety of tasks in any field with human-like flexibility, and is also called strong AI.
It is pictured as being able to think through and handle even tasks it meets for the first time, just as a person would. It has not been achieved yet and remains a research goal.
AnalogyIt is like an all-round butler who handles cooking, cleaning and the household budget, whatever you ask. It is an ideal that people talk about.
How to rememberThink of "general" as in general-purpose: one tool that fits every job.
Exam points
ExamStrong AI (強いAI) is another name for general AI (汎用AI), meaning AI that aims to judge broadly, as a person does, without limiting the field of tasks.
Recognition by AI (AIによる認識) is the function of identifying what is in data such as images, sound or text, or what is being said.
Type
Input
Output
Image recognition
Image
Identification of the objects or people shown
Speech recognition
Spoken voice
Text data
Character recognition (OCR)
Image containing text
Text data
Typical examples are image recognition, which tells what objects appear in a picture, speech recognition, which turns spoken voice into text data, and character recognition (OCR), which reads the characters in an image and turns them into text. Processing that analyzes the meaning of written or spoken language is called natural language processing and is used for translation and conversation.
AnalogyIt is like a person seeing "that is an apple" or hearing "someone called my name," except the machine does it with data.
How to rememberTo recognize is to "know again": the AI matches new data against patterns it has already learned.
Exam points
ExamSpeech recognition (音声認識) is a recognition process that takes in spoken voice as sound data and outputs text data transcribing what was said.
ExamNatural language processing (自然言語処理) is technology for analyzing the meaning of written and spoken language, and its processing differs from character recognition (文字認識), which reads the shapes of characters in an image and turns them into text.
Automation by AI (AIによる自動化) means AI recognizes a situation, makes a judgment based on the result, and carries out work that people used to do.
Recognize (grasp the situation)
Judge (decide what to do)
Act (carry out the work)
Remove the products judged defective from the line without human hands.
3 / 3
Its strength is that it can cope with input that differs slightly each time by applying the judgment criteria it has learned. It is used in visual inspection at factories and in self-driving cars that drive while watching their surroundings.
AnalogyIt is like a robot vacuum that sees where the furniture is, steers around it, and decides its route on the spot as it cleans.
Exam points
ExamSelf-driving (自動運転), in which a system recognizes its surroundings and takes over driving the car, is a typical application of automation by AI (AIによる自動化).
Mix-upBasic RPA is software that replays the operating steps a person has defined, exactly as they are. Automation by AI differs in that the AI itself handles the recognition and judgment.
An AI assistant (AIアシスタント) is a conversational service that understands what users say aloud or type, then gives answers or carries out operations for them.
When used by voice, it also works together with speech recognition, which turns spoken words into text. Everyday examples include smart speakers, voice assistants on smartphones, and chatbots placed at website help desks.
AnalogyIt is like a secretary who, when you call out, checks your schedule or turns off the lights.
How to rememberAn assistant helps: the user stays in charge, and the AI lends a hand from the side.
Exam points
ExamAn AI assistant (AIアシスタント) is an application of AI that builds its responses around the intent of the user's request, with natural language processing (自然言語処理), which analyzes the meaning of language, at its core.
Generative AI (生成AI) is AI that learns in advance from large amounts of data and then creates new content such as text, images or audio in response to instructions.
Prompt (instruction in words)
AI that has already been trained
New text, images and more
The instruction is called a prompt, and it can be written in everyday language (natural language). Because the content it produces is based on patterns in the data it learned from, false information that looks true can slip in, so a person should check the output before using it.
AnalogyIt is like a writer who has read a great many books composing a fresh piece on a requested theme.
How to rememberThe name comes from generate, meaning to produce something new.
Exam points
ExamThe output of generative AI (生成AI) is built on patterns in the data from its prior training, following what the prompt (プロンプト) given in natural language asks for.
ExamWhen generative AI makes up content that contradicts the facts and outputs it as if it were true, the phenomenon is called hallucination (ハルシネーション).
Mix-upRecognition by AI (AIによる認識) identifies what existing data contains, while generative AI creates new content based on patterns in data.