What is artificial intelligence?
A simple definition, everyday examples, and why AI is advancing so fast.
本页目录
1. A simple definition
What is AI?
Artificial intelligence is a set of techniques that let a machine imitate abilities associated with intelligence: perceiving, learning, reasoning, communicating or deciding. The term was proposed in 1955 by John McCarthy, then popularized at the Dartmouth workshop in the summer of 1956.
A practical way to remember it: classic software follows rules written by a human; a modern AI works out its own rules from examples.
Who proposed the term "artificial intelligence" in 1955?
Classic software and modern AI work the same way: both follow rules written in advance by a human.
What year was the Dartmouth workshop that popularized the term "AI"?
2. AI, machine learning, deep learning: nested dolls
Four nested levels
These terms nest like Russian dolls, from broadest to most specific: artificial intelligence (the general goal) → machine learning (the machine learns from data) → deep learning (multi-layer neural networks, since 2012) → generative AI (produces new content: text, image, sound, video, code).
Rank these 4 concepts from broadest to most specific
What sets generative AI apart from other forms of deep learning?
Machine learning is a subset of artificial intelligence, not the other way around.
3. What AI is good at… and bad at
Strengths and fragilities
| Generally good at | Generally fragile at |
|---|---|
| Spotting patterns in large amounts of data | Verifying the truth of what it states |
| Rephrasing, summarizing, translating, classifying | Reasoning reliably about unfamiliar situations |
| Quickly producing a first draft | Knowing about events after its training cutoff |
Sort each task: is AI generally good or fragile at it?
Why can an AI confidently state something false?
4. "Narrow" AI vs. "general" AI
Two very different categories
Narrow AI (or "weak" AI): built for specific tasks. Every current system falls into this category, even the most versatile conversational assistants. General AI (AGI): a hypothetical AI able to adapt to any intellectual task with human-like flexibility — it doesn't exist today.
"The AI understands what it's saying." An assistant like ChatGPT produces convincing sentences by statistically predicting the most likely words; this proves neither understanding nor consciousness.
Today's AI systems, including the most advanced conversational assistants, are all "narrow" AI.
What is AGI (general AI)?
5. Why AI has advanced so fast since 2012
Three ingredients came together
Lots of data — The internet, smartphones and social media produced huge amounts of text, images and video. The ImageNet dataset (2009) played a decisive role.
Far greater computing power — GPUs, designed for video games, turned out to be ideal for training neural networks. In 2012, AlexNet was trained on just two consumer graphics cards.
Better algorithms — The Transformer architecture, published in 2017 by Google researchers, made it possible to build much more powerful language models.
Put these events in chronological order
Which component, originally designed for video games, turned out to be ideal for training neural networks?
Boss challenge: AI strength or fragility?
Sort each statement according to whether it describes a typical strength or fragility of today's AI.