Taking Turing to the Theater %: On Imitation Algorithms

12/04/2018 - 12:00

Computer science has grown out of the seed of imitation. From von Neumann's machine to the famous Turing test, which sparked the field of AI, algorithms have always tried to imitate humans and nature. Examples of such ``imitation algorithms'' are simulated annealing which imitates thermodynamics, genetic algorithms which imitate biology, or deep learning which imitates human learning.


    In this talk, I describe an algorithm which imitates human psychology. Specifically, I discuss $M$ algorithms, which serve as a simple example of psychology-based imitation algorithms. The $M$ algorithm is one of the simplest natural language processing (NLP) algorithms.


    Respecting the long tradition of imitation algorithms, the $M$ algorithm is both extremely simple and extremely powerful. Like other imitation algorithms, the $M$ algorithm is able to solve extraordinarily difficult problems efficiently. The $M$ algorithm efficiently pinpoints critical events in films, theater productions, and other scripts, revealing the rhythm of the texts. 


    At first glance, when trying to design an algorithm which pinpoints critical events of a text, it seems necessary for the algorithm to understand the complete text. Additionally, it would be expected that all layers of the narrative, background information, etc., would also be necessary. In short, it would be expected that the algorithm would imitate the human process of comprehending a text.


    Surprisingly, the $M$ algorithm utilizes the structure of the complete text itself without understanding even a \emph{single} word, sentence, or character in order to discover critical events. The content of the narrative is not necessary for the algorithm to work. Other than an awareness of the illusion of time, borrowed from psychology, the $M$ algorithm circumvents the human process of reading.


    This talk is based on a book (in process).