I know Eliza is considered a Natural Language Processing application, but the application of NLP in this context is “Oracular”.

What I mean by Oracular is that the systems was designed to produce ambiguous output to facilitate the instinct of the user to read meaning into the answer. (My experience with Eliza was as a child on a 64KB system and the program could fool the user for a little while based on sheer novelty, although the limitations were quickly revealed by repetition of output. For kids, this actually became a game of tricking the program into saying funny things;)

This method has a long history in oracles, the most famous certainly being the early binary symbolic system of the I-Ching. (Times being simpler in ancient days, the idea was that a workable amalgam of the universe could be constructed (2)+(4)+(8)+(64) symbols. Each set of symbols is defined by the meanings of the set of the previous order and modified by sequence, which is the key for explaining a given symbol.) The output is ambiguous enough that it may be applied to any input, and rather than the system understanding the input or output, it requires the user to provide the analysis. This may be said to be an engine for generating human insight about a problem. (Monte Carlo may even be utilized, although the sage, working to attain an understanding of each of the symbols, may use intuition to match input with output.)

The reason I ask is I believe this demonstrates a very ancient, algorithmic method of engaging the human mind without the requirement that the algorithm understand the input or output--merely that it produce output to which meaning can be ascribed.

(This almost certainly relates to the relative success of “pornbots” beating the “Turing test” in that the user is chemically induced to read meaning into a given output or string of outputs.)

Aspects of the grounding problem are what got me thinking about this. Not sure if it's relevant that the broken and unbroken lines in the I-Ching represent on and off bits and can be extended to circuits as open and closed.


Tariq's comment hints at this, but this is still in some sense a very mainstream idea.

Check out the transcripts for this year's Loebner Prize. The winner (called Tutor) is again making use of Eliza-like deflections. Some of the other candidates try to use seduction (Columbia), or speaking in a flighty or insistent way (like 2017's Rose).

In all cases, the bots are relying on humans to read meaning into ambiguous or vague statements, which we seem to be hardwired to do. Kind of neat!


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.