With notable progress in recent years, the AGI-oriented NARS (Non-Axiomatic Reasoning System) has been attracting increasing interest from researchers and students from around the world. Built on top of its open-source implementation OpenNARS, several NARS-based or NARS-related projects have been undertaken and are currently under development. To encourage communication and collaboration among the researchers, as well as to introduce these projects to the AGI community, a NARS Workshop will be held during AGI-19 as an integrated part of the conference.
The workshop will be a one-day event (Aug 6th). It will start with a tutorial that introduces the conceptual design and current implementations of NARS, followed by presentations and discussions of the existing projects. The workshop concludes with a panel discussion about future research. The workshop will be open to all AGI-19 attendees. Additional demonstrations and tutorials may be arranged on a case-by-case manner for those wishing to learn more about the project.
Abstract: Using a proprietary visual scene object tracker and the OpenNARS reasoning system we demonstrate how to predict and detect various anomaly classes. The approach combines an object tracker with a base ontology and the OpenNARS reasoning system to learn to classify scene regions based on accumulating evidence from typical entity class (tracked object) behaviours. The system can autonomously satisfy goals related to anomaly detection and respond to user Q&A in real time. The system learns directly from experience with no initial training required (one-shot). The solution is a fusion of sub-symbolic (object tracker) and symbolic (ontology and reasoning).
Abstract:Reproducing the human ability to cooperate and collaborate in a dynamic environment is a significant challenge in the field of human-robot teaming interaction. Generally, in this context, a robot has to adapt itself to handle unforeseen situations. This problem is typical for runtime planning in human-machine interaction, where some factors are not known before the execution starts. The aim of this work is to show and discuss our method to handle this kind of situation, to improve human-robot cooperation in a human-robot team. Our idea is to use the Belief-Desire-Intention agent paradigm and its implementation through the Jason reasoning cycle as a starting point to include a Non-Axiomatic Reasoning System. The result is a new reasoning cycle that gives the robot the ability to select the best plan, while retracting unsuccessful plans and learning new ones on the fly.
Abstract: AGI systems need to operate without interruption. Hence, a vision system for an AGI system has to be active, adaptive and wor under the Assumption of Insufficient Knowledge and Resources (AIKR). This paper describes details of a prototype for a domain independent vision system which learns online and incrementally over the lifetime of the agent under AIKR. This prototype is intended to become the basis of a more capable vision system for NARS based AGI systems.
Abstract: Adaptive Logic and Neural network (ALANN): A neuro-symbolic approach to, event driven, attentional control of a NARS system. A spiking neural network (SNN) model is used as the control mechanism in conjunction with the Non-Axiomatic Logic (NAL). An inference engine is used to create and adjust links and associated link strengths and provide activation spreading under contextual control.
Abstract: In a previous publication, we argued why the behavioral psychology theory Relational Frame Theory (RFT) might be interesting for AGI researchers. This paper explores details of RFT in NARS. More specifically, we investigate different response patterns, such as equivalence, opposition, and comparison in NARS. An additional core feature of RFT is transformation of stimulus function which explains how arbitrary symbols can acquire various functions depending on the relational networks in operation. We demonstrate how NARS handles this process. Finally, future applied and basic research opportunities are discussed.
Abstract: The term “epistemic shifts” refers to a widely recognized phenomenon that knowledge ascribers would ascribe different epistemic statuses to the same beliefs held by the subjects under different internal/external conditions. Contextualists and contrastivists tend to explain this phenomenon by making the semantic connotation of “knowing” different from one context to another, while stake-based invariantists and intellectual invariantists tend to explain the same phenomenon by appealing to external factors like stakes or internal factors like “need-for-closure”. I tend to argue that all the preceding theories lack either the minimal integration of the dimensions of involving parameters or the minimal elegance facilitating cognitive modelling or the minimal range of the scope of applicability. My competing model for explaining epistemic shifts is based on the notion of “time budget deficit”, which refers to the deficit of the operating resources related to time that a subject incurs when she is facing a certain cognitive task. My theory has a wide scope of applicability from the bank cases to varieties of skeptic cases. Furthermore, due to the algorithmic characterization of my model, it also has the potential to be serving for the goal of AI.
Abstract: It is an important research topic in AI to construct a formal and explicit decision-making model for human's cognitive process and realize the computer implementation of the model. Some cognitive psychologists have proposed heuristic methods to describe human decision-making processes. ISSUE: however they have not received the attention of mainstream artificial intelligence. Therefore, classical decision-making theory is still often used to construct computational models for decision-making. On the other hand, although some researchers have proposed a heuristic-based decision-making model, they do not reflect the mechanism of the heuristic method, thus hindering the realization of its computational method. METHODS: In this paper, first we build a decision model based on the first six layers of NAL under the assumption of the insufficient knowledge and resources. Then we outline the preliminary implementation process the proposed model in the computer. CONCLUSION: This paper can provide theoretical and technical means for the cognitive process and computationalization of the decision-making in the real environment.
Abstract: On the theoretical level, the paper points out the great value of cognitive linguistics theory to natural language understanding and how to implement it in AGI system. And using the non-axiom reasoning system (NARS), a series of Chinese natural language understanding research was carried out. I hope to promote the common development of AGI and linguistics.
The program will consist of invited talks, contributed talks, and group discussions. The order of the contributed talks may be subject to change.
|10:15-12:00||NARS Tutorial (Coordinator: Tony Lofthouse)|
NARS Workshop1 (Chair: Kristinn Thorisson)
NARS Workshop 2 (Chair: Kai Liu)
|17:00-18:00||Panel on the Roadmap of NARS (Coordinator: Pei Wang)|
Panelists: Patrick Hammer, Robert Johansson, Kai Liu, Kristinn Thorisson, Yingjin Xu
Temple University, U.S.
Reykjavik University & Icelandic Institute for Intelligent Machines, Iceland
Bohai University, China
Temple University, United States
Temple University, United States