Integrating GPT and NARS, a Cognitive Architecture, by Patrick Hammer, PhD

Generative AI is breaking all records and is as hot as a technology can get. But can it go the distance and give us AGI? One tantalizing prospect is fusing large language models with cognitive architectures. Patrick Hammer has been pursing such a goal.

Dr. Hammer will present:

A NARS implemented as a GPT model prompted to act like a Non-Axiomatic Logic reasoner, with NARS-based memory and control machinery implemented in Python. This is the first reasoning system implementation which uses GPT for reasoning with long-term memory. Also it is following NARS principles regarding evidence tracking, memory management and attentional control which allow the system to operate in a long-term manner while being able to learn and apply new knowledge effectively. In addition, other relevant issues may be discussed.
Features:

• Natural Q&A interaction with the user
• System has no initial knowledge (unless IncludeGPTKnowledge flag is set) but you can simply teach it
• System will make NAL inferences and will raise questions

Diagram

Patrick Hammer holds a PhD in Computer Science (AI track) from Temple University in Philadelphia, doing much work on the NARS system. He is a Postdoctoral Researcher at Digital Futures. The position is hosted by the Division of Biological Psychology and is associated with the project “Intelligence through reasoning” which is aimed at enhancing the autonomy of intelligent agents. Much of the work is being carried out within the Digital Futures research centre that aims to shape an economically, environmental and socially sustainable society through digital transformation. The work is being conducted in a collaboration between the Department of Psychology, Stockholm University and the Division of Computational Science and Technology at The School of Electrical Engineering and Computer Science at KTH Royal Institute of Technology.