Dynamic autonomous joint judgement and adaptive learning

Abstract

People who have been researching in the field of artificial intelligence (AI) have long known about the concept of an artificial general intelligence (AGI) that has been a key motivation for such computer scientists, which aims to foretell of a kind of digital entity that is to be invented in the future, sooner or later, capable of having human-level intelligence spanning novel scenarios and multiple domains of intelligence. Unlike the traditional forms of AI that are narrow-scoped systems that are only capable of operating within the limits of their intended applications and datasets, an AGI agent is aspired to be able to understand and generate its own thoughts, ideas and learn from its experiences just like any normal person. Alongside that, an AGI should be able to interpret human expressions and automatically spontaneously form its own sense of reasoning; in a way, an AGI should be able to perform and behave exactly (if not better than) an average human being, without solely relying on a specific input-output modality such as text, audio, images, videos, etc. This research aims to present a new kind of framework for developing such an AGI system based on a method we call Dynamic Autonomous Joint Judgement and Adaptive Learning. The approach proposed is going to be multidimensional, integrating state-of-the-art machine learning algorithms and paradigms. This is in order to unify several useful and powerful methods together to let the proto- AGI system evaluate and adapt to new information in real-time, developing a form of situational awareness, continuously refine its own decision-making capabilities, interact meaningfully in complex and dynamic virtual environments with a possible potential to scale further beyond into the physical or physical-virtual-augmented reality hybrid contexts. We believe this new approach offers a better alternative than what is currently used in the industry for building large language models (LLM). Most industrial models (like LLMs) are ”data-hungry” and usually require lots of retraining to understand new information. Our Dynamic Intelligent System That Intuitively Learns (DISTIL) model offers a hybrid approach by combining both parametric and non-parametric knowledge structures.