Artificial Intelligence is the study and construction of agents that act intelligently — systems that perceive their situation, reason about it and choose actions that achieve goals. Machine learning is one (now dominant) way to build them; search, logic and probability are the others.

Artificial Intelligence → Machine Learning → Deep Learning → Generative AI / LLMs

Every arrow is a subset: all deep learning is machine learning, and all machine learning is AI — but a chess engine using search, a route planner using A* and a medical expert system using rules are AI without any learning. Those classical techniques are what this subject covers; learning-based AI has its own notes (Machine Learning, Deep Learning, Generative AI).

Four ways to define AI

Russell & Norvig's classic textbook sorts definitions along two axes — thinking vs acting, and matching humans vs being rational (doing the right thing):

	Human-like	Rational
Thinking	Thinking humanly — cognitive modelling: build programs that reason the way people do (cognitive science)	Thinking rationally — the “laws of thought”: correct inference with formal logic
Acting	Acting humanly — the Turing test: behave indistinguishably from a person	Acting rationally — the rational agent: act to achieve the best expected outcome

The Turing test and the Chinese room

A 1966 program already shows why acting human is a weak test of being intelligent. ELIZA imitated a therapist with a handful of pattern → template rules and no understanding at all:

Python

import re

RULES = [                                  # tried in order; the last one always matches
    (r"I need (.*)", "Why do you need {0}?"),
    (r"I am (.*)", "How long have you been {0}?"),
    (r".*\bmother\b.*", "Tell me more about your family."),
    (r"(.*)", "Please, go on."),
]
REFLECT = {"my": "your", "me": "you", "i": "you", "am": "are"}

def eliza(text):
    for pattern, reply in RULES:
        m = re.fullmatch(pattern, text.rstrip(".!"), re.IGNORECASE)
        if m:
            parts = [" ".join(REFLECT.get(w.lower(), w) for w in g.split()) for g in m.groups()]
            return reply.format(*parts)

print(eliza("I need a holiday"))                  # Why do you need a holiday?
print(eliza("I am worried about my exams."))      # How long have you been worried about your exams?
print(eliza("My mother calls every day"))         # Tell me more about your family.
print(eliza("The weather is nice"))               # Please, go on.

Weak vs strong, narrow vs general

Term	Meaning	Status
Weak AI	Machines can act as if they were intelligent	Philosophical claim — widely accepted
Strong AI	Machines that act intelligently really think / have minds	Philosophical claim — open (Chinese room debate)
Narrow AI (ANI)	Excels at one task or domain: chess, face recognition, spam filtering, translation	Every deployed system today
General AI (AGI)	Human-level competence across essentially all cognitive tasks, able to transfer and learn new ones	Research goal; LLMs are broader than older systems but debated
Superintelligence (ASI)	Far exceeds humans in every domain	Hypothetical — central to AI-safety discussions

A short history — booms and winters

Year	Milestone
1943	McCulloch & Pitts model a neuron as a logic gate
1950	Turing, “Computing Machinery and Intelligence” — the imitation game
1956	Dartmouth workshop (McCarthy, Minsky, Shannon, Rochester) — the name “artificial intelligence”; the field is born
1958–66	Perceptron (Rosenblatt), Samuel's learning checkers player, the General Problem Solver, LISP, ELIZA
1969–73	Minsky & Papert show single perceptrons can't learn XOR; the Lighthill report criticises AI's “combinatorial explosion”
1974–80	First AI winter — promises unmet, funding cut
1980–87	Expert systems boom (XCON at DEC, MYCIN-style rule bases); Japan's Fifth Generation project; backprop popularised (1986)
1987–93	Second AI winter — Lisp-machine market collapses, expert systems prove brittle and costly to maintain
1990s–2000s	Probabilistic AI (Bayesian networks, HMMs), statistical ML; Deep Blue beats Kasparov (1997)
2011–12	IBM Watson wins Jeopardy!; AlexNet wins ImageNet — the deep-learning era begins
2016	AlphaGo beats Lee Sedol — deep networks + Monte Carlo tree search + reinforcement learning
2017–now	The Transformer (2017), large language models, ChatGPT (2022), multimodal models and LLM agents

What is the difference between AI, machine learning and deep learning?

What is the rational agent approach, and why is it preferred?

What were the AI winters?