Not a day goes by without hearing the word "AI." News outlets, social media — AI stories flood in every single day. Yet AI as a technology has been researched since the 1950s. So why has it become such a huge deal only now? In this chapter, we'll cover what AI is, why now, what it can and can't do, and how to start using it — all in one go, without jargon.
The big picture in 3 minutes
What's really behind the AI boom — why all the fuss now
The answer is simple. AI has finally reached a level ordinary people can actually use.
The turning point was November 2022, when OpenAI released ChatGPT. Until then, "AI" meant difficult tools for programmers and researchers. But with ChatGPT, you just open a browser and type in a question. Write "Make me a packing list for next week's business trip," and an answer comes back in seconds. That ease of use changed the world.
The explosive spread of AI, in numbers
Let's look at concrete numbers to see just how widely it has spread (as of March 2026).
SimilarWeb 2025
McKinsey 2025
Up sharply from 11% in 2024
roughly ¥45–80 trillion
People who, just three years ago, would have said "AI? What's that?" now use it to draft work emails and reports. This is a shift on the scale of the smartphone's rise.
How is this different from past AI booms?
In fact, this is the third AI boom. There have been earlier periods when people cried "AI will change everything!" — but each time it hit a technical wall and expectations deflated. What makes people say this time is the real thing comes down to two advances.
Earlier AI was operated by experts through commands and programs. Today's AI, though, answers you if you just talk to it in plain English. No manuals, no code required.
Text, images, code, music, video. AI can now generate brand-new content. This is the technology called "generative AI," and it's the core of the current boom.
The types of AI, roughly sorted — generative, predictive, recognition
"AI" is a single word, but it actually comes in many forms. You don't need to memorize them all, but sorting them into three broad categories makes the world of AI far easier to grasp.
💡 The boundaries keep blurring. Recent AI can generate text (generation), understand what's in an image (recognition), and predict what comes next from context (prediction) — all at once. AI that combines several of these abilities is called "multimodal AI." We cover it in detail in Chapter 6.
Today's hottest category, generative AI, will build an entire outline if you ask ChatGPT to "draft the structure for next week's presentation," and Midjourney will generate an image in seconds if you type "a cat on a beach at sunset." Things that until a few years ago were considered "work only humans can do" — AI can now handle them. That is the essence of the current boom.
What's inside ChatGPT — an LLM is "super-powered predictive text"
When you use ChatGPT, Claude, or Gemini, it looks as if they answer by "thinking." But the actual mechanism is surprisingly simple. In a word, it's a super-charged version of your phone's predictive text.
Type "Good" and it suggests "Good morning." From your own typing history, it predicts the word most likely to come next (data size: a few MB, only the last few characters).
It does the same thing — "predict the next word" — but on a vastly different scale. It predicts from the enormous body of text on the internet (data size: multiple TB, considering the full context of the conversation).
Your phone only looks at the last few characters, but an LLM considers the entire context of the conversation and generates "the single best next word," one word at a time. This "ability to understand context" is what makes an LLM look smart.
How it got "smart" — three steps
It reads the web, books, papers, and more, storing language patterns in hundreds of billions of parameters (pre-training). Thousands of GPUs, over several months.
Using high-quality example conversations, it learns "when asked this, answer like this" through additional training (fine-tuning).
Humans rate responses as "good/dangerous" to improve answer quality (RLHF), making replies polite and non-harmful.
⚠️ It isn't "knowing" — it's "predicting." When "What's the capital of Japan?" gets "Tokyo," it isn't because the AI holds Tokyo as a fact — it's because it has learned the pattern that "Tokyo" tends to follow "the capital of Japan is." Because of this design, it can confidently produce plausible-sounding but wrong information — that's the "hallucination" discussed below.
What AI can and can't do — avoid both overtrust and underestimation
AI attracts both the hope of "a magic tool that can do anything" and the dismissal of "just a machine after all." The reality is in between: what it's good at and what it's bad at are clearly divided.
- Writing and editing text (drafting emails, summarizing, proofreading) = where it shines most
- Translation (an English email goes from 30 minutes to 1)
- A sounding board for brainstorming ideas
- Getting it to write code or formulas
- Guaranteeing facts (plausible doesn't mean correct)
- Admitting mistakes (the more wrong the answer, the more confident it sounds / MIT research)
- Final judgment (don't hand it medical, legal, or major decisions)
- The latest information (it can struggle with anything after its training cutoff)
📊 Reliability changes wildly by field. Even the best model hallucinates about 0.7% of the time on basic tasks (Suprmind 2026). Yet research finds that in law over 75%, and in medicine over 23% of the time, misinformation gets mixed in (Stanford RegLab). It's safest to think "AI is a different animal depending on the field."
Practical tips for dividing the work
The key is to use AI in its strong areas, and always have a human check its weak ones. Remember these three principles and you won't go far wrong.
Have AI create the draft, review the content, then send.
Have it produce several options, and decide which to pick yourself.
Use AI's answer as a starting point, and confirm important facts against official sources.
Let's actually try it — 3 free AI tools you can try today
Seeing is believing. Every major AI tool can be tried for free. No credit card required. With just an email address or a Google account, you can be chatting with an AI five minutes from now.
The spark that lit the AI boom. When in doubt, start with this. Sign up with email or a Google account at chat.openai.com.
Try: "Make a packing list for a 3-day trip to Kyoto" / paste one of your own emails and say "rewrite this more politely."
Made by Anthropic. Known for careful, logical answers. Sign up with email at claude.ai; available on web, iOS, and Android.
Try: upload a PDF and say "give me the 5 key points" / "what are this proposal's weaknesses and how could it be improved?"
Integrates with Search, Gmail, and Docs. Also great at fetching the latest information. Just log in with your Google account at gemini.google.com.
Try: "Today's weather in Tokyo and what to wear" / upload a photo and ask "what is this?"
Which should you choose? Honestly, if it's your first time, any of them is fine. All three are free, so the best move is to try them all and find the one that suits you. We break down when to use each tool in detail in Chapter 2, "How to choose an AI tool."
The bare minimum you should know before using AI
Before you get started, there are just three things to keep in mind. Each is an important point where "I didn't know" won't cut it.
AI plausibly generates information that isn't true. It may even fabricate papers that don't exist. Across all businesses, one estimate puts the losses at around $67 billion a year (AllAboutAI 2024).
Countermeasure: Don't take it at face value; verify important facts against a separate source.
What you type in may be used for training. One survey found that about 8.5% of prompts contain confidential data.
Countermeasure: Don't enter company secrets, other people's personal information, or passwords. Check your company's usage policy.
The rights to AI-generated works are being debated worldwide. In the US, there have been rulings that "works made by AI alone get no copyright."
Countermeasure: Don't use the output as-is; arrange and add to it yourself. For work, either state that it's AI-generated or edit it substantially.
We dig much deeper into AI's risks and ethical challenges in Chapter 5.
- AI is booming now because it can finally "converse and create" (generative AI).
- There are three types: generative, predictive, recognition. The lines are blurring (multimodal).
- Inside ChatGPT is "super-powered predictive text" — it's predicting, not knowing.
- AI makes mistakes without blinking. The iron rule: AI drafts, humans do the final check.
- Start by trying one for free. ChatGPT, Claude, or Gemini — any is fine.
Having read this far, you already understand the basic workings of AI and how to work with it. In the next chapter, Chapter 2, "How to choose an AI tool," we compare the features of the major AI tools and explain how to find the one that fits you.
References
- McKinsey, "The State of AI" (2025 survey) — 88% corporate AI adoption
- Suprmind, "AI Hallucination Rates & Benchmarks" (2026) — comparative hallucination-rate data
- AllAboutAI, "AI Hallucination Report" (2024) — $67.4 billion in economic losses from hallucinations
- MedhaCloud, "67 AI Adoption Statistics for 2026" — statistics on AI adoption
- Netguru, "AI Adoption Statistics in 2026" — 38% of knowledge workers use AI daily