A brief history: from Aristotle to GOFAI to agents

Two philosophical roots, and Kant's synthesis

AI did not start in 1956 — it inherits a 2,000-year argument about where knowledge comes from.

• The rationalist tradition (Plato, Descartes, Leibniz) holds that the world is reconstructed through clear, distinct mathematical ideas. In AI this becomes: describe a domain as predicate-calculus statements and solve problems by proving theorems about that formal world. This is the lineage of symbolic AI.
• The empiricist tradition (Hobbes, Locke, Hume) holds that nothing enters the mind except through the senses, and that knowledge arises by association of perceptual properties through repeated experience. In AI this becomes connectionist networks, associative memory, and statistical learning.
• Kant synthesized the two: knowledge needs both an a priori component (structure the mind brings) and an a posteriori component (experience). The modern echo is hybrid AI — built-in structure (representations, architectures) plus learning from data.

The through-line from antiquity is Aristotle's matter/form distinction: the form (pattern) of a thing can be separated from its matter (medium). That single idea — patterns can be manipulated independently of their physical substrate — is what makes "artificial" intelligence conceivable at all. Boole later showed all of logic reduces to AND, OR, NOT on binary values; Frege built the first-order predicate calculus; Russell & Whitehead treated mathematics as pure formal symbol manipulation. Each step pushed reasoning closer to something a machine could do.

The physical symbol system hypothesis and GOFAI

Newell and Simon distilled the rationalist program into a sharp, testable claim — the one most likely to appear on your exam:

> Physical Symbol System Hypothesis (PSSH): a physical symbol system has the necessary and sufficient means for general intelligent action.

Unpack the two words. Necessary — anything that is intelligent must be a symbol system. Sufficient — being a symbol system is enough to be (capable of) intelligence; nothing extra is required. The PSSH is the archetype of the rationalist approach, and it is exactly what the alternatives below reject. Logic-and-inference AI in this style is nicknamed GOFAI — Good Old-Fashioned AI — knowledge in formal logic, intelligence as logical inference over it.

Three alternatives to symbolic AI

The PSSH has been challenged (Searle, Winograd & Flores, Brooks). Three families of alternatives reject "intelligence = symbol manipulation":

1. Connectionist (neural) networks — model the brain's architecture rather than the rational mind. Knowledge is implicit in patterns of connection weights, learned by adapting those weights. Strength: graceful, noise-tolerant partial matching where brittle symbolic programs fail. (Modules 8.)
2. Genetic / evolutionary algorithms — do not reason about a problem at all; they evolve a population of candidate solutions via selection, crossover, and mutation, letting fitter candidates survive and reproduce. (Module 8.8.)
3. Agent-based / emergent intelligence — many simple, autonomous, situated, interacting agents whose cooperative behavior is greater than the sum of its parts. Intelligence is emergent from a structured society of agents, not located in one reasoner. (Herbert Simon's ant parable: an ant's complicated path mostly reflects the environment's complexity, not the ant's.)

Landmark systems (recall fodder)

Worked example — match each historic system to what it did. These names are classic short-answer bait; learn the one-line gloss for each.

The pattern across DENDRAL, MYCIN, PROSPECTOR, and XCON: domain-specific knowledge beat general-purpose reasoning. SHRDLU is the cautionary twin — a dazzling demo in a tiny world whose techniques would not generalize.