Coduction: An actor based implementation and evaluation
Date
2022-10-03
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
ORCID
0000-0003-2439-8378
Type
Thesis
Degree Level
Masters
Abstract
There are various approaches for making a computer system intelligent. However, statistical approaches
such as machine learning are at times said to be more about perception rather than intelligence. The work
presented in the thesis showcases coduction, a symbolic approach for knowledge maintenance and hypothesis
formation, along with its implemented prototype and the experiments conducted to evaluate it. Coduction
tries to address the aspects of intelligence that remain after perception has been addressed.
Coduction involves a teacher and a learning agent. The teacher may be a human or a computer system,
while the learning agent is the intelligent component in the system. The learning agent, simply referred to as
the learner, starts as a blank slate of knowledge and receives natural language statements considered to facts
about the world, from the teacher. Rather than trying to decipher the meaning of the words in a sentence,
the learner only views the statements as groups of symbols in context with the statements it has previously
received. The essence of coduction lies in compressing and efficiently representing textual knowledge, which
is how hypothesis formation takes place. As an additional outcome in the process of hypothesis formation,
the learner develops curiosity about facts related to the knowledge that it has been provided, and asks the
teacher questions regarding the same, potentially requiring the teacher to conduct experiments in the world
to get an answer to those questions.
Coduction is designed as a concurrent actor-based system, in which a dedicated actor is responsible for
every aspect of the system including the knowledge stored symbolically in the knowledge base and its relationships with other aspects. Every time a new statement is provided to the learner, it triggers a concurrent
flow of multiple asynchronous messages through the interconnected network of actors, thus forming a symbolic connectionist network of actors. This process of messages passing through the network of actors leads
to the formation of hypotheses and developing curiosity. The experiments presented in the thesis conducted
to evaluate the effectiveness and performance of various aspects of coduction show promising results.
Description
Keywords
coduction, actors, hypothesis formation, curiosity
Citation
Degree
Master of Science (M.Sc.)
Department
Computer Science
Program
Computer Science