---
title: The Foundations of AI
module: Foundations
moduleNumber: 1
lessonNumber: 2
order: 102
summary: >
  Where the rational-agent idea came from and what surrounds it. AI inherited its
  core tools from eight older disciplines — philosophy, mathematics, economics,
  neuroscience, psychology, computer engineering, control theory, and linguistics.
  Its history runs in cycles of boom and winter, from the 1956 Dartmouth workshop
  through expert systems to the statistical turn. And the deep-learning era —
  AlexNet, the Transformer, GPT-3, AlphaGo — is a new way of computing the agent
  function at scale, not a new definition of AI.
topics: [Foundations]
sources:
  - book: AIMA
    ref: "§1.2 The Foundations of AI; §1.3 The History of AI; §1.4 The State of the Art"
  - book: AIMA
    ref: "Ch. 1 — Introduction"
---

This builds on [What Is Artificial Intelligence?](/artificial-intelligence/foundations/what-is-ai),
which settled on a single definition: AI is the study and design of rational agents.
A definition is only half the answer; where the idea came from and how it has fared
are the other half. This lesson traces the disciplines AI draws on, the cycles of
boom and winter it has passed through, and the deep-learning turn that now dominates
its public face, and shows that none of it disturbs the rational-agent frame fixed
in the previous lesson.

## The disciplines AI draws on

AI did not arise in isolation. It inherited ideas, and sometimes whole formalisms,
from eight disciplines that had each spent centuries on a piece of the problem.[^rn-found]

$$
% caption: AI as a confluence of older disciplines. Each contributed a piece of
% the puzzle: what to compute (philosophy, economics), how to compute it
% (mathematics, engineering), and what a thinking system looks like (neuroscience,
% psychology, linguistics, control theory).
\begin{tikzpicture}[>=stealth, font=\footnotesize,
  src/.style={draw, minimum width=27mm, minimum height=9mm, align=center},
  hub/.style={draw=acc, text=acc, very thick, minimum width=22mm, minimum height=12mm, align=center}]
  \definecolor{acc}{HTML}{2348F2}
  \node[hub] (ai) at (0,0) {AI};
  % ring of sources
  \node[src] (phi) at (-4.6,2.3)  {philosophy};
  \node[src] (mat) at (0,3.0)     {mathematics};
  \node[src] (eco) at (4.6,2.3)   {economics};
  \node[src] (neu) at (5.5,0)     {neuroscience};
  \node[src] (psy) at (4.6,-2.3)  {psychology};
  \node[src] (eng) at (0,-3.0)    {computer engineering};
  \node[src] (con) at (-4.6,-2.3) {control theory};
  \node[src] (lin) at (-5.5,0)    {linguistics};
  \foreach \n in {phi,mat,eco,neu,psy,eng,con,lin}
    \draw[->, black] (\n) -- (ai);
\end{tikzpicture}
$$

| Discipline | Question it answered | Tools bequeathed to AI |
| --- | --- | --- |
| **Philosophy** | Can formal rules yield valid conclusions? Is the mind a physical machine? | syllogisms; **materialism** — brain under physical law _is_ mind, the premise that makes a built mind conceivable |
| **Mathematics** | What is provable, computable, and computable in reasonable time? | **logic** (Boole, Frege); **computation**, **tractability**, **NP-completeness** (Turing); **probability** (Cardano–Bayes); Gödel's incompleteness as the outer limit |
| **Economics** | How should a rational agent choose to maximize expected outcomes? | **utility** (von Neumann–Morgenstern); **decision theory** = utility + probability; **game theory**; Bellman's **Markov decision processes** for sequential choice |
| **Neuroscience** | How do brains process information? | the existence proof that simple cells (neurons) give rise to thought — "brains cause minds" |
| **Psychology** | How does a mind turn stimulus into action? | Craik's information-processing view — stimulus → internal representation → manipulation → action, the knowledge-based-agent loop |
| **Computer engineering** | What artifact runs the intelligence? | the machine itself; AI repaid the debt with time sharing, interpreters, the linked list, automatic storage management |
| **Control theory** | How does an artifact regulate itself toward a goal? | Wiener's feedback minimizing error between current and desired state; maximizing an **objective function** over time |
| **Linguistics** | What does understanding language require? | Chomsky's result that language needs knowledge and context, not just syntax — binding AI to knowledge representation |

## A short history

The field advances in cycles: optimism, a wall, retrenchment, a new idea that
clears the wall. Six eras cover the arc.[^rn-hist]

$$
% caption: A timeline of AI. Bursts of optimism (early enthusiasm, the expert-system
% boom) alternate with walls and funding cuts (the AI winters), each followed by a
% new paradigm: knowledge, then statistics, then the agent view.
\begin{tikzpicture}[>=stealth, font=\footnotesize]
  \definecolor{acc}{HTML}{2348F2}
  \definecolor{red}{HTML}{C0392B}
  % time axis
  \draw[->, thick] (0,0) -- (13.4,0);
  \foreach \x/\yr in {0/1943, 1.9/1956, 4.3/1969, 6.7/1980, 9.1/1987, 11.5/2001} {
    \draw[thick] (\x,0.12) -- (\x,-0.12);
    \node[font=\scriptsize, anchor=north] at (\x,-0.16) {\yr};
  }
  % era labels above, alternating heights to avoid collisions
  \node[acc, align=center, anchor=south, font=\scriptsize] at (0.95,0.35)
    {gestation\\ McCulloch-Pitts,\\ Turing};
  \node[align=center, anchor=south, font=\scriptsize] at (3.1,1.55)
    {birth (Dartmouth)\\ early enthusiasm};
  \node[red, align=center, anchor=south, font=\scriptsize] at (4.3,0.35)
    {f\/irst winter\\ (Lighthill 1973)};
  \node[acc, align=center, anchor=south, font=\scriptsize] at (5.6,1.55)
    {knowledge-based\\ expert systems};
  \node[red, align=center, anchor=south, font=\scriptsize] at (8.4,0.35)
    {AI Winter\\ (late 1980s)};
  \node[acc, align=center, anchor=south, font=\scriptsize] at (10.3,1.55)
    {statistical AI\\ neural nets return,\\ scienti\/f\/ic method};
  \node[acc, align=center, anchor=south, font=\scriptsize] at (12.4,0.35)
    {big data,\\ agents};
\end{tikzpicture}
$$

**Gestation (1943–1955).** McCulloch and Pitts modelled an artificial neuron as an
on/off switch and showed that networks of them could compute any logical function;
Hebb gave a rule for updating connection strengths. Turing, in "Computing Machinery
and Intelligence" (1950), already sketched machine learning, genetic algorithms,
and reinforcement learning, and proposed teaching a machine as one teaches a child.

**Birth (1956).** John McCarthy convened a two-month workshop at Dartmouth, coining
the name **artificial intelligence**. The proposal's conjecture set the field's
agenda: that "every aspect of learning or any other feature of intelligence can in
principle be so precisely described that a machine can be made to simulate it."
Newell and Simon arrived with the Logic Theorist, a running program that proved
theorems, and for the next twenty years the Dartmouth attendees and their students
dominated the field.

**Early enthusiasm (1952–1969).** A run of successes — Newell and Simon's General
Problem Solver, Samuel's self-improving checkers player, McCarthy's invention of
**Lisp** and of time sharing, Minsky's **microworlds** — fed extravagant
predictions. Simon claimed in 1957 that machines that "think, learn and create"
already existed and would soon match the human mind across the board.

**A dose of reality (1966–1973).** The predictions collapsed on contact with scale.
Machine translation foundered because translation needs world knowledge, not just
syntax ("the spirit is willing but the flesh is weak" came back as "the vodka is
good but the meat is rotten"). Early programs solved microworlds by trying all
combinations of steps, a strategy that dies on the **combinatorial explosion** of
real problems. Minsky and Papert's _Perceptrons_ exposed the representational
limits of simple neural nets, and funding evaporated — the first **AI winter**,
formalized in Britain by the Lighthill report (1973).

**Knowledge-based systems (1969–1979).** The lesson drawn was that general-purpose
search — the "weak methods" — does not scale, and that power comes from
domain-specific **knowledge**. DENDRAL inferred molecular structure from mass
spectra; MYCIN diagnosed blood infections with about 450 rules and outperformed
junior doctors. **Expert systems** cleanly separated a knowledge base from the
reasoning engine, and briefly turned AI into an industry — before overpromising
brought a second AI winter in the late 1980s.

**Statistics, agents, and the scientific method (1986–present).** Three shifts
converged. **Back-propagation** was reinvented in the mid-1980s, reviving neural
nets. AI adopted rigorous experimental methodology, building on existing theory
rather than replacing it: **hidden Markov models** took over speech recognition,
**Bayesian networks** gave a principled formalism for uncertain reasoning, and
machine learning reintegrated with statistics and information theory. In parallel,
researchers returned to the whole **agent** and its environment, and the recent
availability of very large data sets shifted emphasis from the cleverness of the
algorithm to the scale of the data.

## The state of the art

AI stopped being a promise and became infrastructure. A short sample of what
already worked when _AIMA_ was written suggests the range:[^rn-soa]

| Domain | Working system |
| --- | --- |
| Robotic vehicles | STANLEY and BOSS, driverless cars that won DARPA's Grand and Urban Challenges |
| Speech recognition | end-to-end automated dialog for airline booking |
| Planning and scheduling | NASA's Remote Agent controlling a spacecraft autonomously |
| Game playing | Deep Blue defeating world chess champion Kasparov |
| Spam fighting | learning classifiers sorting a billion messages a day |
| Logistics | DART, whose Gulf War planning DARPA said repaid its 30-year AI investment |
| Machine translation | statistical translation from two trillion words of text |

None of it is science fiction; it is the ordinary output of engineering,
mathematics, and statistics. As one summary of the period put it, thousands of AI
applications are now embedded in the infrastructure of every industry.

## The deep-learning era

The third edition of _AIMA_ was published in 2009, a year before the shift that
now defines the field's public face. That shift was not a new definition of AI —
the rational-agent frame absorbs it without strain — but a change in _which
technique_ dominates the state of the art. It is worth tracing, both because it is
the AI most people now meet and because it is the concrete subject of the sibling
[deep-learning](/deep-learning/foundations/what-is-deep-learning) course.

**The 2012 turn.** The catalyst was image classification. Krizhevsky, Sutskever,
and Hinton entered a deep convolutional network, later called **AlexNet**, in the
ImageNet Large Scale Visual Recognition Challenge and cut the top-5 error from
about $26\%$ to $16\%$ — a margin no incremental method had approached (Krizhevsky
et al., NeurIPS 2012).[^alexnet] Three ingredients that _AIMA_ names individually as loose
threads came together at once: back-propagation (the algorithm the statistical
turn revived), very large labeled data sets (ImageNet's ~1.2 million images), and
enough parallel compute (two GPUs) to train a large network in days. Nothing in
the underlying learning rule was new; the scale was.

**The 2017 architecture.** For sequence data — language above all — the decisive
step was the **Transformer**, which replaced recurrence with an attention
mechanism that lets every position attend directly to every other (Vaswani et al.,
"Attention Is All You Need," NeurIPS 2017).[^transformer] Attention parallelizes cleanly across
a sequence, so Transformers scale to far larger models and corpora than the
recurrent networks before them. This is the architecture underneath essentially
every large language model, and it is developed in full in the
[NLP](/natural-language-processing/transformers/transformers-and-attention) course.

**Scaling and the language-model agents.** Trained on very large text corpora,
Transformers improved smoothly as data, parameters, and compute grew together —
loss falling as a power law $L(N) \propto N^{-\alpha}$ in each resource (Kaplan et
al., 2020, the **scaling laws**). GPT-3 (Brown et al., "Language Models are
Few-Shot Learners," NeurIPS 2020) showed that at scale a single model performs many
tasks it was never explicitly trained for, specified only by a natural-language
prompt.[^gpt3] The connection back to this lesson is direct: such a model, wrapped
in a loop that calls tools and acts on an environment, _is_ an agent in the exact
sense of the next lesson — it perceives (prompt and tool outputs) and acts (the text
and tool calls it emits). The rational-agent frame needed no revision; the model is
a new way of computing the agent function $f: \mathcal{P}^\ast \to \mathcal{A}$, not a
new kind of thing.

$$
% caption: A language-model agent as an instance of the perceive-act loop. The
% model reads a prompt plus tool results (percepts), emits text and tool calls
% (actions) into an environment of tools and users, and the results feed back as
% the next percept. Structurally identical to the agent diagram above.
\begin{tikzpicture}[>=stealth, font=\small,
  box/.style={draw, minimum width=30mm, minimum height=15mm, align=center, font=\footnotesize}]
  \definecolor{acc}{HTML}{2348F2}
  \node[box, draw=acc, text=acc, very thick] (m) at (0,0) {language model\\(agent)};
  \node[box] (t) at (6.4,0) {tools, users,\\environment};
  \draw[->, thick] (t.north west) ++(0,-0.15) .. controls (3.2,1.6) .. (m.north east) ++(0,-0.15);
  \node[font=\footnotesize, anchor=south] at (3.2,1.15) {prompt, tool results (percepts)};
  \draw[->, acc, thick] (m.south east) ++(0,0.15) .. controls (3.2,-1.6) .. (t.south west) ++(0,0.15);
  \node[acc, font=\footnotesize, anchor=north] at (3.2,-1.15) {text, tool calls (actions)};
\end{tikzpicture}
$$

**Learning to act, not just to classify.** In parallel, reinforcement learning
scaled the same way. AlphaGo combined deep networks with Monte-Carlo tree search
and defeated a top human Go professional (Silver et al., "Mastering the Game of
Go with Deep Neural Networks and Tree Search," _Nature_ 2016) — a result long
thought a decade or more away, because Go's branching factor defeats the brute
search that cracked chess. Its successor AlphaGo Zero reached superhuman play from
self-play alone, given only the rules (Silver et al., _Nature_ 2017).[^alphago] These are
squarely the utility-maximizing agents this course builds toward, developed in the
[reinforcement-learning](/reinforcement-learning/foundations/what-is-reinforcement-learning)
subject.

$$
% caption: The deep-learning era layered on AIMA's timeline. Each milestone is
% a new way of computing the agent function at larger scale, not a new definition of
% AI: perception (AlexNet 2012), sequence models (Transformer 2017), general
% language models (GPT-3 2020), and superhuman decision-making (AlphaGo 2016).
\begin{tikzpicture}[>=stealth, font=\footnotesize]
  \definecolor{acc}{HTML}{2348F2}
  \draw[->, thick] (0,0) -- (12.6,0);
  \foreach \x/\yr in {0.5/2009, 3.0/2012, 5.5/2016, 8.0/2017, 10.8/2020} {
    \draw[thick] (\x,0.12) -- (\x,-0.12);
    \node[font=\scriptsize, anchor=north] at (\x,-0.16) {\yr};
  }
  \node[align=center, anchor=south, font=\scriptsize] at (0.6,0.35) {AIMA\\3rd ed.};
  \node[acc, align=center, anchor=south, font=\scriptsize] at (3.0,1.45) {AlexNet\\deep vision};
  \node[acc, align=center, anchor=south, font=\scriptsize] at (5.5,0.35) {AlphaGo\\deep RL};
  \node[acc, align=center, anchor=south, font=\scriptsize] at (8.0,1.45) {Transformer\\attention};
  \node[acc, align=center, anchor=south, font=\scriptsize] at (10.8,0.35) {GPT-3\\LLMs};
\end{tikzpicture}
$$

Two honest qualifications keep this in proportion. First, these systems inherit
the field's oldest problem in a new form: they are trained to imitate or to score
well on a proxy measure, and a rational agent maximizes _the letter_ of its
measure, so misspecified objectives produce confident, fluent errors — the
alignment concern the course revisits in
[decision theory](/artificial-intelligence/uncertainty/making-decisions). Second,
the deep-learning turn did not retire the rest of the field. Search still plans,
logic still represents structured knowledge, probability still quantifies belief;
the strongest current systems (AlphaGo is the clean example) _combine_ learned
components with search and probabilistic reasoning. The chapters that follow are
not superseded by the deep-learning era — they are the other half of it.

## The frame for everything that follows

Strip away the history and the disciplines and one commitment remains: **AI is the
study and design of rational agents.** An agent perceives an environment and acts
on it; a rational agent chooses, for each percept sequence, the action expected to
maximize an external performance measure. Every remaining topic in this course is a
tool for making that choice well under some new difficulty.

- When the environment is known and the agent must find a sequence of actions to a
  goal, the tool is [search](/artificial-intelligence/search/uninformed-search) —
  uninformed, [informed](/artificial-intelligence/search/informed-search),
  [local](/artificial-intelligence/search/local-search), and, against an opponent,
  [adversarial](/artificial-intelligence/search/adversarial-search).
- When the agent must represent what it knows and derive new facts, the tools are
  [propositional](/artificial-intelligence/logic-and-planning/propositional-logic)
  and [first-order logic](/artificial-intelligence/logic-and-planning/first-order-logic),
  [inference](/artificial-intelligence/logic-and-planning/inference-and-resolution),
  and [planning](/artificial-intelligence/logic-and-planning/classical-planning).
- When the environment is uncertain, the tools are
  [probability](/artificial-intelligence/uncertainty/probability-and-bayes),
  [Bayesian networks](/artificial-intelligence/uncertainty/bayesian-networks),
  [reasoning over time](/artificial-intelligence/uncertainty/reasoning-over-time),
  and [decision theory](/artificial-intelligence/uncertainty/making-decisions).
- When the agent must improve from experience, the tools are
  [learning from examples](/artificial-intelligence/learning/learning-from-examples),
  [probabilistic learning](/artificial-intelligence/learning/probabilistic-learning),
  and [reinforcement learning](/artificial-intelligence/learning/reinforcement-learning).

The next lesson,
[intelligent agents](/artificial-intelligence/foundations/intelligent-agents),
makes the agent, its environment, and the performance measure precise, and shows
how the structure of the environment dictates the structure of a rational agent.
Everything after that is a way to fill in the one line at the center of the frame:
given the percepts, choose the best action.

[^rn-found]: **Russell & Norvig**, _AIMA_ §1.2 — the contributing disciplines: philosophy, mathematics, economics, neuroscience, psychology, computer engineering, control theory, and linguistics.
[^rn-hist]: **Russell & Norvig**, _AIMA_ §1.3 — the history of AI, from the McCulloch-Pitts gestation and the 1956 Dartmouth workshop through early enthusiasm, the AI winters, knowledge-based systems, and the statistical and agent-based turns.
[^rn-soa]: **Russell & Norvig**, _AIMA_ §1.4 — the state of the art: robotic vehicles, speech recognition, planning, game playing, spam fighting, logistics, and machine translation as deployed AI systems.
[^alexnet]: **A. Krizhevsky, I. Sutskever, G. Hinton**, "ImageNet Classification with Deep Convolutional Neural Networks," NeurIPS 2012 — the deep CNN (AlexNet) that cut ImageNet top-5 error from roughly 26% to about 16% and set off the deep-learning era.
[^transformer]: **A. Vaswani et al.**, "Attention Is All You Need," NeurIPS 2017 — the Transformer, replacing recurrence with self-attention; the architecture underneath modern large language models.
[^gpt3]: **T. Brown et al.**, "Language Models are Few-Shot Learners," NeurIPS 2020 — GPT-3, a single large language model performing many tasks specified only by a prompt; and **J. Kaplan et al.**, "Scaling Laws for Neural Language Models" (2020), for the empirical scaling relationships.
[^alphago]: **D. Silver et al.**, "Mastering the Game of Go with Deep Neural Networks and Tree Search," _Nature_ 529 (2016), and "Mastering the Game of Go without Human Knowledge," _Nature_ 550 (2017) — deep reinforcement learning with tree search defeating a top human Go player, then reaching superhuman play from self-play alone.