Intelligence
The word underneath "artificial intelligence" — used constantly, defined by nobody, and the reason the field's biggest arguments never resolve.
When not to use it
- As a product claim. "Intelligent" is an adjective doing the work a capability statement should do. Say what it does and how often it's right.
- As a threshold. Nothing becomes true when a system crosses into "intelligent," because there's no line and no instrument.
- As a settled premise in an argument. If you and the person you're arguing with haven't defined it, you'll disagree for an hour and discover you agreed.
Reach for something else instead
- Capability statements — "solves X at Y accuracy on Z inputs." Testable, falsifiable, and it's what you actually needed.
- Generalisation measures — how does it handle inputs unlike training data? That's closer to what people mean and it's measurable.
- Sample efficiency — how much data did it need to learn this? The most defensible proxy anyone has proposed.
Sources & further reading
- Legg & Hutter (2007), A Collection of Definitions of Intelligence — over seventy published definitions, which is itself the finding.
- Chollet (2019), On the Measure of Intelligence — the argument that benchmarks measure skill, not intelligence, and that efficiency of acquisition is the better target.
- Turing (1950), Computing Machinery and Intelligence — where the field chose to sidestep the definition and ask about behaviour instead. Still the most influential dodge in computer science.
Primary sources, listed so you can check the claims on this page rather than take them on trust.
Where people go wrong
- Treating benchmark performance as evidence of intelligence. It's evidence of performance on the benchmark, and contamination makes even that shakier than it looks.
- Assuming abilities correlate. In humans they do, so we infer broadly from narrow evidence. Models break that inference — strong at one thing, incoherent at the neighbouring thing.
- Arguing about whether a system is intelligent without stating a definition. The argument is then about vocabulary, and it can't be won.
At a glance
FieldFoundations
Core ideathe undefined word under the whole field
Definitions in circulation70+
Behavioural viewit's what it does
Generalisation viewit's handling the unfamiliar
DifficultyBeginner
Flashcards for this concept
Question
—
Answer
—
1 / 4
Often compared with
Intelligence vs. capability — an undefined property vs. a measurable claim. Only one of them can be tested.