Say you have to enter a story to a computer. Now the computer has to identify the philosophical concept on which the story is based, say:

  1. Was it a "self-fulfilling prophecy"?

  2. Was it an example of "Deadlock" or "Pinocchio paradox situation"?

  3. Was it an example of how rumours magnify? or something similar to a chain reaction process?

  4. Was it an example of "cognitive dissonance" of a person?

  5. Was it a story about "altruism"?

  6. Was it a story about a "misunderstanding" when a person did something "innovative" but it accidentally was innovated earlier so the person was "falsely accused" of "plagiarising"?

et cetera,

Given that the story is not only a heavy rephrase of the pre-existing story; not only character names and identities are totally changed, but the context completely changed, the exact tasks they were doing are changed.

Can computers identify such "concepts" from stories? if yes, then what mechanism it uses?

  • $\begingroup$ I gave my best to criticize an existing answer. Was my comment not good enough, so that you have deleted it? $\endgroup$ – Manuel Rodriguez Jul 16 at 22:17

No. This is currently out of the scope for any language processing system. It requires a general understanding of abstract concepts which is not possible for machines at present.

In order to recognise a self-fulfilling prophecy, you first need to identify that something is a prophecy. So it needs to be something that expresses a possible future state, for which you need to identify what possible future states are; and then you need to see whether it is self-fulfilling. Conceptually this is far too complex to do.

You might get away for some of these with formal criteria (eg use of future tense for something describing a future state/event), but this is far too imprecise.

"Altruism" requires knowledge about typical expected behaviour; you would need to be able to identify motives behind people's actions, and then decide whether it was altruistic or not. This is just too complex for now (and the foreseeable future).

  • 1
    $\begingroup$ When one reviews cutting-edge AI narrative output, it's clear that the algorithms understand syntax (they can structure sentences and paragraphs, and write in various formats) and equally clear the algorithms have no semantic understanding of the content itself. $\endgroup$ – DukeZhou Jul 16 at 21:48
  • 2
    $\begingroup$ @DukeZhou Indeed -- structure is pretty much in hand, but the meaning (both semantic and pragmatic) of that structure is still far outside our grasp. $\endgroup$ – Oliver Mason Jul 17 at 8:28

Moral learning and concept understanding with artificial intelligence requires a model. That model takes a story written in natural language as input and detects philosophical events like altruism or a self-fulfilling prophecy. Unfortunately, such a model for story parsing isn't available at github so we have to create it from scratch. The best practice method for creating new models is “learning from demonstration”. That means, the AI has to observe the story understanding capabilities of a human operator.

Let us go into the details. At first we need a short story from the Winnie-the-Pooh universe, who is a fictional teddy bear. A human operator reads the story loud and has to press a button, if he detects a philosophical concept in the plot. Or to explain it shorter, a new case is created which contains of the original story plus the detected altruism event. Now we can take the annotated story as input data for training a neural network and this will produce the required model. To test the model, the first step is to replicate existing annotations (training dataset) and then the model can be used to detect concepts in an unknown dataset, for example in the story “The Three Little Pigs”.


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.