SHRDLU

• 1. The Linotype Machine

  ›

  Terry Winograd began work on SHRDLU at MIT between 1968 and 1970. He wrote the program using Lisp and Micro Planner languages. The software ran on a DEC P-6 computer connected to a graphics terminal. Winograd chose the name SHRDLU from the top row of keys on a Linotype machine. This arrangement listed letters in order of their frequency in English text. The sequence E-T-A-O-I-N-S-H-R-D-L-U represented the most common characters used by typesetters.

  1

• 2. A World Of Blocks

  ›

  The program operated inside a simplified virtual environment called the blocks world. This space contained basic geometric objects like blocks, cones, and balls. Users could move these items around within a virtual box. The entire set of objects and locations required only about fifty words to describe. Nouns included block and cone while verbs covered actions like place on or move to. Adjectives such as big and blue defined the physical properties of each item. The simplicity allowed the system to handle complex interactions with limited vocabulary.

  23

• 3. Parsing English Commands

  ›

  SHRDLU functioned as an early language parser that interpreted English commands. A user might instruct the computer to pick up a big red block. The system would respond with OK if it understood the request. If the command was unclear, the computer replied I DON'T UNDERSTAND WHICH PYRAMID YOU MEAN. It could search back through previous interactions to find context for pronouns. Asking to take the cone off meant taking the green cone mentioned earlier. The software deduced relationships between objects based on rules provided during development.

  1

• 4. Memory And Deduction

  ›

  The software utilized memory structures to maintain context across multiple interactions. One could ask questions about history such as did you pick up anything before the cone. SHRDLU remembered names given to specific objects or arrangements. Defining a steeple as a small triangle on top of a tall rectangle allowed new queries about steeples. The program also tracked what was possible in the world versus impossible. It realized triangles could not be stacked after attempting the action repeatedly. Basic physics made blocks fall over independent of the language parser itself.

  1

• 5. Optimism And Limits

  ›

  SHRDLU served as a landmark demonstration of artificial intelligence capabilities in the late 1960s. This success led other AI researchers toward excessive optimism about future systems. Later attempts to handle realistic ambiguity and complexity failed where SHRDLU had succeeded. Subsequent efforts like Cyc focused on providing programs with considerably more information. Winograd later noted that the project inadvertently fostered unrealistic expectations among peers. The field eventually shifted focus away from simple simulated worlds toward broader knowledge bases.

  3

• 6. Interactive Fiction Roots

  ›

  Winograd identified SHRDLU as a formal precursor to interactive fiction despite lacking traditional narrative elements. He stated in a 1991 interview that the user interacts with simple commands to move objects around a virtual environment. The 1976-1977 game Colossal Cave Adventure is broadly considered the first true work of interactive fiction. No distinct story-telling existed within the original SHRDLU system during its operation. The interaction model influenced how users engaged with text-based games in subsequent decades.

  4