AI retrocomputing projects: Eliza [1]

14 Apr 2020


At the beginning of 2019 I decided to get back into coding by building an interpreter for the Lisp programming language, following the excellent MAL guidelines. This successful exercise yielded JKL 1.0, my own version of Lisp. A year later, I decided to give JKL 1.0 a serious workout by re-implementing one of the classic Artificial Intelligence (AI) systems pioneered in the 1960s and 1970s, as described by Peter Norvig in his excellent book Paradigms of Artificial Intelligence Programming (hereafter ‘Paradigms’). This evolving post charts my progress on this task.

As a starting point for this retrocomputing challenge, I decided to try and recreate the Eliza system, which in modern terms is a chatbot that can fake a keyboard-based dialogue with a human. It was originally described in a 1966 issue of the Communications of the Association for Computing Machinery. Although primitive by today’s standards, Eliza was ground-breaking at the time, and an early indicator of the potential (and limitations) of AI. I picked Eliza from the various systems described in Paradigms because it’s basic function - chat with a person - is easy to understand, but at the same time an excellent test of a useful AI mechanism: pattern matching.

Some aspects of Eliza implementation needed functions that weren’t already present in JKL 1.0, which was after all based on a minimalist teaching language (MAL). These changes, which went hand-in-hand with the early stages of Eliza development, involved:

Eliza - the first chatbot

The initial top level of Eliza is a sort of READ-EVAL-PRINT loop:

(def! eliza (fn* ()
   (loop ()
      (prn (read-tokens "eliza> "))