Sketchpad is the first interactive computer-graphics program. Ivan Sutherland built it as his MIT doctoral thesis between 1961 and 1963, submitted January 1963, advised by Claude Shannon with Marvin Minsky and Steven Coons on the committee.
Before Sketchpad, computers received instructions on paper tape and returned answers on paper tape. After Sketchpad, a person could draw on a screen with a light pen, declare that two lines should be parallel, and watch the computer enforce the relationship in real time. The change is not incremental. It is the founding moment of computer-aided design, of object-oriented programming's "instance" concept, and of the graphical user interface itself — twenty years before the term existed.
Sketchpad ran on the TX-2, a one-of-a-kind research computer at MIT Lincoln Laboratory operational by 1958. The TX-2 had sixty-five thousand thirty-six-bit words of magnetic core memory, paper-tape input, magnetic-tape storage, a seven-inch CRT, a light pen, knobs and toggle switches, and a thirty-seven-button "button box" mounted to the console.
It was an "on-line" computer in an era when almost all computers were "batch" computers. A person sat at the console rather than handing a deck of cards to an operator. This was the precondition for everything Sketchpad became — without continuous human-machine dialog the constraint system would have been pointless.
The light pen looked like a pen but was an inverted camera. When held to the CRT it detected the scan-line of the electron beam passing under its tip and reported back the time, which the computer used to compute the pen's position. The pen had no transmitter. The screen was the transmitter; the pen was the receiver.
Sutherland wrote a pen-tracking algorithm that displayed a small dynamic cross under the pen at all times. As the pen moved the cross moved with it, prediction running ahead of motion. This is the direct ancestor of every cursor on every screen ever since.
In this room the mouse impersonates the pen. The crosshair is a faithful tribute to the original tracking cross.
The pen lets you draw. The button box lets you declare. Sketchpad's deepest move was to treat geometric relationships as first-class objects.
A user could draw two rough lines, push the PARALLEL button, and Sketchpad would solve a small system of equations to rotate them into agreement. PERPENDICULAR, HORIZONTAL, EQUAL-LENGTH, COINCIDENT — each was a constraint type the system knew how to satisfy. The user could combine many of them and the program would solve them all together by iterative relaxation.
This is parametric CAD. Forty-eight years before Solidworks called itself parametric, before Grasshopper or Onshape, the idea is already running on the TX-2. The CANVAS mode in this room is a faithful (if greatly simplified) miniature of that solver.
To make all of this work Sutherland invented a data structure that could store arbitrary topological relationships among objects. He called it a "ring" — each entity (line, point, constraint, instance) held doubly-linked pointers to every entity it was connected to, in a circular list per relationship.
It is, in retrospect, a graph database. The thesis describes traversal, insertion, deletion, and garbage collection on these rings. The same ideas re-surface in Smalltalk a decade later, in CAD kernels two decades later, in scene graphs three decades later. Sutherland built the substrate before anyone knew there was a substrate to build.
The other half of the thesis is instances. Any drawing could be saved as a "master." A master could be invoked anywhere on the screen by creating an "instance" — a placement of the master at a transform (position, rotation, scale).
Crucially, a master could contain other instances. Sub-pictures could be recursive. Edit a master and every instance updated, everywhere, at every depth. This is the symbol library in Illustrator. This is the component in Figma. This is the prefab in a game engine. This is the block in AutoCAD. The pattern has not changed in sixty-three years.
To show off recursive sub-pictures Sutherland built a stick-figure face — round head, line nose, two eyes, hair. He called her Nefertite. The trick: each eye was an instance of a master called EYE, and the EYE master had three variants (open, half-closed, closed).
By swapping which variant the right-eye instance pointed to, Nefertite could be made to wink. The same trick made her hair swing and her bicycle pedals turn in later demos. The animation came from changing instance pointers, not from re-drawing — the breakthrough that became key-frame animation.
The NEFERTITE mode in this room reproduces the trick. The eyes are real instances. The variants are real masters. Click an eye to cycle.
Sutherland's final demo treated constraints as physics. A bridge truss was drawn as a set of jointed lines. Each line was constrained to be of fixed length. Then a downward force was added at a joint.
The constraint solver, which had been satisfying parallelism and equal-length, now satisfied force-balance equations. The truss flexed, sagged, transmitted stress. The same machinery that drew geometry was simulating mechanics. This is the seed of every physics-based CAD plugin, every FEA system, every game engine's rigid-body solver.
The TRUSS mode in this room is a tiny version of that demo. APPLY LOAD adds downward force; RELAX runs the solver.
Sutherland left MIT, ran ARPA's IPTO, co-founded Evans & Sutherland (head-mounted display, real-time computer graphics), taught at Utah where his students included Alan Kay, Jim Clark (SGI, Netscape), Ed Catmull (Pixar), John Warnock (Adobe), Henri Gouraud and Bui Tuong Phong (the shading models named after them), and Nolan Bushnell (Atari). Every cartridge in this museum descends from a student in his classroom.
Sutherland received the Turing Award in 1988 and the Kyoto Prize in 2012. The thesis was published as a book in 2003 (Cambridge tech report UCAM-CL-TR-574). Read it. It is still ahead of most software.
Nine and ten years on, two rooms reached analogous structures in three dimensions. SV (cart 013, MAGI SynthaVision 1972) stored a solid as a boolean expression and resolved it per ray. BLD (cart 012, Cambridge BUILD 1973) stored it as a closed surface of vertices, edges, and faces, with Boolean operations defined over the boundary. Neither lineage is a direct descendant of Sutherland's ring, but both inherit the same intuition: keep relationships first-class so the computer can re-solve when geometry shifts. Sketchpad is in the plane. SV and BUILD are the solid.
Twenty-five years on, Sutherland's "draw, then declare a relationship, then watch the machine enforce it" returns as a workflow for the interface itself. IB (cart 015, Interface Builder 1988) lets you drag a button into a window, then wire its action to a target. The light pen became the mouse; the constraint became the action-port; the ring of pointers became the NIB archive. No code written. No code generated. The system holds the structure and re-solves at runtime — the same intuition, applied to the GUI itself.
"The Sketchpad system makes it possible for a man and a computer to converse rapidly through the medium of line drawings."
— Ivan E. Sutherland, January 1963
From THE LITERATURE · CANON / 001B · the theory wing of CLASSICERY. Hear Ivan Sutherland on Sketchpad: A Man-Machine Graphical Communication System read aloud — → track 03 →