What is AI?
AI is software that learns patterns from data instead of following rules a programmer typed by hand. US law (15 U.S.C. § 9401) defines it as a machine-based system that makes predictions, recommendations, or decisions for human-defined objectives. The AI people use daily — ChatGPT, Gemini, Claude — are large language models.
Why — the first-principles explanation
Normal software is a list of instructions. A human decides every rule: if the price is over $100, apply a discount. That works when a human can state the rule. It falls apart for tasks nobody can write down — recognizing a cat, translating Korean, telling a joke. You know a cat when you see one, but you cannot write the if-statement for "cat."
AI flips this around. Instead of writing the rule, you show the machine millions of examples and let it find the rule itself. Inside is a huge grid of numbers called weights — modern systems have hundreds of billions of them. Training nudges those numbers, over and over, until the system's guesses match the examples. Nobody hand-writes the weights; they're discovered by trial and error at enormous scale. This is why AI can do things its builders can't explain: the knowledge lives in numbers, not in readable code.
That mechanism explains both the magic and the failures. AI is fundamentally a pattern-matcher over what it has seen. It has no notion of true and false, only of what typically follows what. So it can write a fluent paragraph about a court case that never happened, because fluent-sounding text is the pattern it learned. It is guessing well, not knowing.
The word "AI" also keeps moving. Spam filters and chess programs were called AI in the 1990s; now they're just software. What people mean by AI in 2026 is mostly generative AI — systems that produce new text, images, code, or audio — which broke into the mainstream when ChatGPT launched in November 2022.
An example that makes it click
Think about how a kid learns to spot dogs. You never hand them a definition — "four legs, fur, wet nose, tail." That definition would flag a cat. Instead you point at dogs on the sidewalk for two years. Eventually the kid nails it, including a hairless dog they've never seen, and they can't tell you what rule they used. They just know.
AI learns the same way, only the child sees 20 dogs a week and the machine sees 20 million pictures. And that's also where the cracks show: show that kid only golden retrievers, and they'll call a chihuahua a cat. AI inherits the blind spots of whatever it was shown — which is why the training data matters as much as the code.
Key facts
- US federal law, 15 U.S.C. § 9401(3), defines artificial intelligence as "a machine-based system that can, for a given set of human-defined objectives, make predictions, recommendations or decisions influencing real or virtual environments."
- The EU AI Act (Regulation 2024/1689, Article 3(1)) defines an AI system as one that "infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions."
- Both the US and EU legal definitions center on inference — deriving outputs from input — not on robots or consciousness.
- NIST published its AI Risk Management Framework 1.0 on January 26, 2023, the main US voluntary standard for managing AI risk.
- Modern large language models contain hundreds of billions of learned parameters; Meta's Llama 3.1 405B has 405 billion.
- AI has no built-in fact-checking step. It predicts likely output, which is why fabricated but fluent answers ('hallucinations') happen.
▶ The 60-second explainer (script)
AI is software that learns patterns from data instead of following rules a person typed by hand. That's it. That's the whole idea. Regular software is a list of instructions a human wrote: if the price is over a hundred dollars, apply a discount. Clean, predictable, and useless for anything a human can't put into words. Nobody can write the if-statement for what makes a cat a cat. So AI flips it. You show the machine millions of examples and let it work out the pattern itself. Inside there's a giant grid of numbers — modern systems have hundreds of billions of them. Training nudges those numbers until the machine's guesses match reality. Nobody hand-writes them. They're discovered. That's why AI can do things its own builders can't fully explain. US law puts it plainly: a machine-based system that makes predictions, recommendations, or decisions toward human-defined objectives. The EU AI Act says nearly the same thing. Notice what's missing from both — no robots, no consciousness. Just inference. And that's also the catch. AI is a pattern-matcher. It doesn't know true from false, only what usually follows what. So it can confidently describe a court case that never happened, because sounding right is the pattern it learned. It's guessing extremely well. That's not the same as knowing.
What authoritative sources say
People also ask
Is AI the same as a robot?
No. AI is the software that makes decisions; a robot is a physical body. Most AI has no body at all — it runs in a data center and sends back text or images.
Does AI actually understand what it says?
There's no scientific consensus, but mechanically it predicts likely outputs from patterns in training data. It has no verification step, which is why it can state false things fluently.
Is AI conscious or alive?
No. No legal or scientific definition of AI includes consciousness. Both the US statute and the EU AI Act define AI purely by what it does — infer outputs from inputs.
What's the difference between AI and machine learning?
Machine learning is the main method used to build AI today — learning patterns from data. AI is the broader goal; machine learning is how nearly all of it is currently achieved.
When did AI become mainstream?
The term dates to 1956, but public adoption exploded after ChatGPT's launch in November 2022, which put a usable generative AI into ordinary people's hands.
The same question, asked other ways
- What is an AI?3,600/mo
- What is AI (artificial intelligence)?2,900/mo
- What is artificial intelligence (AI)?2,900/mo
- What exactly is AI?1,600/mo
- What is the definition of AI?1,600/mo
- What is AI technology?1,600/mo
- How is AI defined?1,300/mo