A working definition
Artificial intelligence is the engineering of systems that perform tasks we would call intelligent if a person did them: perceiving, reasoning, planning, and using language. In practice, almost all deployed AI today is narrow — extremely capable at one family of tasks, with no general understanding of the world.
- Narrow AI: one task family (translation, ranking, detection).
- General AI: hypothetical, transfers across arbitrary domains.
- Machine learning: the subset of AI that learns behaviour from data.
The genesis: 1950–1980
Turing's 1950 paper asked whether machines can think and proposed an imitation test. The 1956 Dartmouth workshop named the field. The following decades were dominated by symbolic AI: hand-written rules, search algorithms and expert systems that encoded a specialist's knowledge as if-then statements.
Why the paradigm shifted
Rule systems were brittle — real-world inputs are noisy and the number of rules explodes. Two AI winters followed the funding hype. From the late 1980s the statistical approach took over: rather than writing rules, fit parameters to data and let the model infer the rule. Cheap compute, large datasets and backpropagation turned that idea into the modern field.