Machine Trainable Software Models Towards a Cognitive Thinking AI with the Natural Language Processing Platform NLX

EasyChair Preprint no. 7848, version 2

Abstract

Since the last decade, machine learning, especially with artificial neural networks, has triggered a new quantum leap in computer science. Despite the considerable achievements, these applications still lack a general purpose approach for artificial intelligence (AI). The main reason is the absence of the ability for cognitive reflection or self-awareness. They are mainly highly specialized trained patterns that can solve intricate problems but cannot describe themselves. I would like to contrast this with a new method of trainable software models that shall be capable for self-awareness. Implemented in the project Natural Language Platform NLX. With this project it shall be demonstrated that self-aware AI is key for human-like cognitive tasks. The hypothesis claims that to reach this goal machines require to describe its system context semantically by a formal model. Neuronal networks are good at specific tasks, but the trained patterns cannot derive a reasoning for the trained solution only that it satisfies its intended functionality but not why. Creating formal model instead of patterns has turned out, that the formal nature of natural language is the best to reach that goal of a self-aware AI. Certainly there are other AI’s that do natural language processing with neuronal networks. But most of the models try to resolve the content with too rigid constraints and with little attention to the context. For this project context plays a key role to resolve the meaning of natural language. If the context is resolved correct it can be used for general purpose tasks resolving anything imaginable

Keyphrases: AI, cognitive thinking, machine learning, Software Models

@Booklet{EasyChair:7848,
  author = {Felix Schaller},
  title = {Machine Trainable Software Models Towards a Cognitive Thinking AI with the Natural Language Processing Platform NLX},
  howpublished = {EasyChair Preprint no. 7848},

  year = {EasyChair, 2022}}