Artificial intelligence (AI)

"Artificial intelligence" is a moving label. Unlike "model" or "token", the term doesn’t name a fixed thing — it points at whatever computers can newly, impressively do. Each time a technique becomes reliable it stops being called "AI" and gets a more specific name, and the label slides forward to the next thing still unsolved.

This is the AI effect: once a technique works, it gets renamed — "just" search, "just" statistics — and "AI" moves on. Bertram Raphael put it in 1971: "AI is a collective name for problems which we do not yet know how to solve properly by computer." Larry Tesler’s variant, from around 1979: "Intelligence is whatever machines haven’t done yet."

The label’s drift is visible across eras:

Era

What "AI" meant

1950s

Symbolic reasoning — theorem provers, checkers

1960s–70s

Rule-based symbolic programs — ELIZA, SHRDLU

1980s

Expert systems — hand-written if-then rules

1990s

Game-tree search — Deep Blue beats Kasparov (1997); researchers avoided the word "AI"

2000s

Statistical machine learning — spam filters, recommenders; sold as "ML", not "AI"

2010s

Deep learning — AlexNet (2012), AlphaGo (2016)

2020s

Large language models — ChatGPT (2022) made "AI" mean chatbots

This is why conversations about AI talk past each other. A claim like "AI can’t reason" carries a hidden timestamp — it depends entirely on which era’s technique the speaker has in mind. When a discussion stalls, swap "AI" for the precise term: the model, the harness, the agent, the context.

At its core, though, the term is about agency — the capacity of a machine to act, to make decisions, and to pursue goals with some degree of autonomy.

Related topics include:

References