6. History — Early Enthusiasm (1952–1969)
Source: AIMA 4th Ed, §1.3.2
The Optimism Era
After Dartmouth, AI entered a period of remarkable optimism. Several programs seemed to demonstrate that machines could reason, learn, and even play games better than their creators.
Key Milestones
General Problem Solver (GPS) — 1957
Simon and Newell built the General Problem Solver: - Modeled human problem-solving rather than just getting correct answers. - Used means-ends analysis: identify the difference between current state and goal, find an operator to reduce that difference, apply it, repeat. - First program designed to work the same way humans do — testing against human protocols. - Simon and Newell claimed: “We have invented a thinking machine.”
Physical Symbol System Hypothesis (PSSH): > “A physical symbol system has the necessary and sufficient means for general intelligent action.” Any system exhibiting intelligence must manipulate structured symbols — humans, computers, or any other physical substrate.
Arthur Samuel’s Checkers Program — 1952–1959
- IBM researcher Arthur Samuel built a checkers (draughts) program that used reinforcement learning — learning from outcomes to improve play.
- Learned to play at a strong amateur level.
- Demonstrated: computers can learn to do things better than they were programmed to do, refuting the idea that “computers can only do what they’re told.”
- Precursor to TD-Gammon (backgammon) and AlphaGo.
Lisp and the Advice Taker — 1958
John McCarthy made two landmark contributions: 1. Lisp — a high-level programming language that became the dominant AI language for 30 years. Based on lambda calculus; excellent for symbolic manipulation. 2. Advice Taker (conceptual proposal): a hypothetical system that: - Stores general world knowledge as logical axioms - Uses deduction to derive plans of action - Accepts new axioms without being reprogrammed - Embodied the principles of knowledge representation and reasoning
Geometry Theorem Prover — 1959
Nathaniel Rochester’s group at IBM, particularly Herbert Gelernter, built a program that could prove geometry theorems that many mathematics students would find difficult.
Microworlds (MIT, Minsky’s Group)
Minsky supervised a series of students who worked on microworlds — limited, self-contained domains:
| Program | Author | What It Did |
|---|---|---|
| SAINT (1963) | Slagle | Solved first-year calculus integration problems |
| ANALOGY (1968) | Evans | Solved geometric analogy problems from IQ tests |
| STUDENT (1967) | Bobrow | Solved algebra story problems in English |
| SHRDLU (1972) | Winograd | Natural language understanding in the blocks world |
Blocks world: Simulation of a tabletop with toy blocks. SHRDLU could accept commands like “Find a block taller than the one you are holding and put it in the box.” This was impressive but fragile — it only worked in that tiny domain.
Perceptrons — 1960s
Building on McCulloch-Pitts: - Widrow & Hoff (1960): Adalines — used gradient descent to train linear threshold units. - Frank Rosenblatt (1962): Perceptron — single-layer trainable neural network. - Perceptron convergence theorem: if a solution exists, the learning algorithm will find it.
This was hugely exciting — but the limitation (only linear decision boundaries) would soon become apparent.
The Prevailing Mood
Herbert Simon in 1957: > “There are now in the world machines that think, that learn and that create. Moreover, their ability to do these things is going to increase rapidly until — in a visible future — the range of problems they can handle will be coextensive with the range to which the human mind has been applied.”
He also predicted that within 10 years: - A computer would be chess champion ✓ (took 40 years, not 10) - A significant mathematical theorem would be proved by machine ✓
The promises were not wrong in direction — only wildly optimistic about the timeline.