AI is technology that lets computers do part of the work humans do with their minds, such as writing text, predicting from data, and telling photos or sounds apart. What's making headlines now is generative AI, which can "create" text and images from a plain request. AI as a technology has been studied since the 1950s. So why is it causing such a stir only now? In this chapter, you'll get the big picture in one go, without jargon: what AI is, why it matters now, what it can and cannot do, and how to start using it.
The full picture in 3 minutes
What's Behind the AI Boom – Why Is Everyone Talking About It Now?
The answer is simple: "AI has finally reached a level that ordinary people can use."
The turning point came in November 2022, when OpenAI released ChatGPT. Until then, AI was a difficult tool used by programmers and researchers. ChatGPT, however, only requires you to open a browser and type a question. Type "Make a packing list for my business trip next week," and an answer comes back within seconds. This ease of use changed the world.
In the years since, services such as Claude and Gemini have followed ChatGPT, and using AI to write work emails or draft reports has become far from unusual. It is said to be as big a change as the spread of smartphones.
How Is This Different from Past AI Booms?
In fact, this is not the first AI boom. There have been times in the past when people claimed "AI will change everything!", but each time the technology hit barriers and expectations faded. What sets this time apart are two advances.
Earlier AI was operated by experts using commands or programs. Today's AI, however, answers when you speak to it in ordinary language. No manuals or code needed.
Text, images, code, music, and video. AI can now generate new content. This technology is called "generative AI," and it is the core of the current boom.
A Quick Guide to AI Types – Generative, Predictive, and Recognition AI
"AI" covers many kinds of technology. You don't need to memorize them all, but dividing them into three groups by what the AI does makes the AI world much easier to understand. Generative AI is only one part of AI as a whole.
💡 One service can play several roles. Recent AI can look at a photo and describe it (recognition), then write text based on it (generation), handling several roles in one service. AI that handles several types of information, such as text, images, and audio, is called "multimodal AI" (Google's glossary). For details, see Chapter 6.
The most talked-about generative AI today works like this: if you ask ChatGPT, "Come up with an outline for next week's presentation," it creates the whole outline; if you write "a cat by the sea at sunset" in Midjourney, it generates an image within seconds. What was considered "work only humans can do" just a few years ago is now something AI can handle. That is the essence of the current boom.
What's Inside ChatGPT – An LLM Is a "Super-Powerful Predictive Text"
When you use ChatGPT, Claude, or Gemini, it can look as if it is "thinking" before it answers. What supports this text generation is the LLM (large language model). In one sentence, the mechanism is a supercharged version of the predictive text on your smartphone.
Type "good m" and it suggests "good morning." It predicts words likely to follow the last few characters, based on your input history and similar data.
It does the same thing, predicting what comes next, but at an enormous scale. It uses patterns learned from a vast amount of text and predicts what comes next by looking at the context of the whole conversation.
The unit in which an LLM builds text is called a token, which corresponds to words, parts of words, characters, and so on (Google's introduction to language models). A smartphone looks only at the last few characters, but an LLM looks at the context of the whole conversation to predict the next likely token, and piles these up to build sentences. This "ability to grasp context" is what makes LLMs seem smart.
How Did It Get "Smart"? – 3 Steps
It learns word patterns from vast amounts of text such as web pages, books, and papers (pre-training). This is the most labor-intensive stage, running large computing systems for a long time.
It learns additionally from high-quality conversation examples that show how to answer when asked a question (fine-tuning).
Humans rate responses as "good" or "dangerous" to improve their quality (such as RLHF). This brings responses closer to being polite and not harmful.
This learning takes place when the model is built. Even if you rephrase your requests in a chat, the model itself is not rewritten (Google's explanation). Also, the amount the model can read at one time (the context window) has a limit, so in long conversations, conditions given at the beginning may be less reflected. Restate important conditions when needed.
⚠️ It doesn't "know"; it "predicts." When you ask "What is the capital of Japan?" and get "Tokyo," the AI is not looking up Tokyo as knowledge. It has learned that "Tokyo" tends to follow "the capital of Japan is". Because of this mechanism, it can confidently present wrong information that sounds plausible. This is what is called "hallucination" below. When you use the search feature, the AI also uses information it looks up on the spot to answer.
What AI Can and Cannot Do – Avoiding Overtrust and Underestimation
There are two views of AI: that it is a magical tool that can do anything, and that it is "just a machine" and underestimated. In reality, it lies somewhere in between, with clear strengths and weaknesses.
- Writing and editing text (drafting emails, summarizing, proofreading)
- Translation (reading and writing English emails becomes much faster)
- A sounding board for brainstorming ideas
- Having it write code or math
- Guaranteeing facts (plausible-sounding, but not necessarily correct)
- Judging its own confidence (it writes confidently even when it's wrong)
- Final decisions (don't hand over medical, legal, or other important decisions)
- Latest information (things after its training cutoff are hard to handle without search)
📊 Reliability varies by field and task. Even with the same AI, everyday writing is a strength, but errors increase in specialized fields such as law and medicine, in precise figures, and in recent events. A neat table or a source-like link does not by itself mean something is correct. It is safest to think of AI as different from one field to another.
Practical Tips for Using AI Well
The key is to use AI for what it is good at, and always have a human check it where it is weak. Remember these three principles and you won't go far wrong.
Have AI create a draft, check the content, and then send it.
Have it propose several options, and decide for yourself which one to choose.
Use the AI's answer as a starting point, and check important facts against official sources.
Try It Yourself – 3 Free AI Tools You Can Start Using Today
Seeing is believing. Most major AI tools can be tried for free. You don't need to subscribe to a paid plan to practice.
The spark behind the AI boom. If you're unsure, start with this one. Sign up at chatgpt.com with an email address or a Google account.
Try: "Make a packing list for a 2-night, 3-day trip to Kyoto" / paste your own email and say "Rewrite this more politely"
Made by Anthropic. Known for polite, logical answers. Sign up at claude.ai. New sign-ups require verification with a phone number that can receive SMS. Available on Web, iOS, and Android.
Try: Upload a PDF and ask "Give me the 5 key points" / "What are the weaknesses of this proposal, and how could they be improved?"
Integrates with Search, Gmail, and Docs. Also good at retrieving the latest information. Log in at gemini.google.com with a Google account (there are conditions based on age, region, and account type).
Try: "Today's weather in Tokyo and what to wear" / upload a photo and ask "What is this?"
Which one should you choose? Honestly, if you're just starting, any of them is fine. All three are free, so trying them all and finding the one that suits you is the best approach. You can check the registration requirements in Claude's phone number verification and Gemini's terms of use (verified October 14, 2026). For more on how to use each tool, see Chapter 2, "Choosing AI Tools".
Your First Practice: Asking for a Reading Club Announcement
Once you've signed up, go ahead and make a first request. Here, you'll ask for an announcement that mixes things that are already decided with things that aren't yet. It's a chance to check two things: whether AI polishes the wording, and whether it refrains from filling in what hasn't been decided. (The content here is just an example.)
"Please write a reading club announcement in a friendly tone for people attending for the first time. Decided: the book is The Great Gatsby; the date and time are November 14, 14:00–16:00; the venue is the 2nd-floor meeting room at the Central City Library; people can attend even if they haven't finished the book. Not yet decided: the admission fee. Don't add anything that isn't decided; write 'TBD' instead."
📚 Reading club notice
Let's read The Great Gatsby together. Newcomers are very welcome. You're welcome to join even if you haven't finished the book.
Date: November 14, 14:00–16:00
Venue: Central City Library, 2nd-floor meeting room
Admission fee: TBD (we'll let you know once it's decided)
- The admission fee says "Free" or "500 yen"
- The date or venue differs from what you provided
- A promise you didn't request is included, such as "We'll provide tea and snacks."
The two things to check are whether all four facts you provided appear as given, and whether the admission fee is still "TBD." If the fee has become "Free," tell the AI specifically, "The admission fee is not decided yet. Please change it back to 'TBD,'" and then check the revised text again. Just because the writing reads naturally doesn't mean it has kept the conditions. This is the most important sense to develop when working with AI. You can learn more about how to structure requests in Chapter 3, "How to Write Prompts".
What You Should Know Before Using AI
Before you start using AI, there are three things worth keeping in mind. Each is an important point that you can't afford to say "I didn't know" about.
AI can generate false information in a plausible way. It may even invent nonexistent papers or sources.
What to do: Don't take it at face value. Verify important facts against the original sources or official information.
Depending on the service and settings, what you enter may be stored or used for training.
What to do: Don't enter company confidential information, other people's personal information, or passwords. If you use AI for work, check your company's usage rules.
The rights to AI-generated output are being debated around the world. The U.S. Copyright Office has indicated that simply entering a prompt is not enough for copyright protection, and that a human's creative expression is required.
What to do: Check whether the output resembles someone else's work. Labeling something "AI-generated" does not make the rights issues disappear.
You can read the U.S. Copyright Office's view in its January 29, 2025 announcement. For a deeper look at AI risks and ethical issues, see Chapter 5.
- AI is booming now because, since ChatGPT launched in November 2022, AI that can converse and create has become available to anyone.
- There are three types: generative, predictive, and recognition. One service often plays several roles.
- What's inside ChatGPT is a "super-powerful predictive text." It predicts what comes next, token by token.
- AI can make plausible-sounding mistakes. The rule of thumb: AI drafts, and humans do the final check.
- Start by trying one free tool. Separate what's decided from what isn't when you make a request, and compare the answers.
Having read this far, you now understand the basic workings of AI and how to work with it. In the next chapter, Chapter 2, "Choosing AI Tools", we compare the features of major AI tools and explain how to find the one that suits you.
References
- Google: Machine Learning Glossary – definitions of AI and multimodal
- Google: Introduction to Language Models – tokens and context
- Google: How Large Language Models Work – generation and hallucinations
- Google: Tuning and Prompting – model adjustment versus how requests are phrased
- OpenAI: Conversation State and Input Scope – context window
- Anthropic: Verifying Your Phone Number for Claude – sign-up requirements
- Google: Gemini Terms of Use – age, region, and account requirements
- U.S. Copyright Office: Report on AI and Copyright (January 29, 2025) – distinguishing human creative expression from prompts