Artificial General Intelligence: 15th International Conference, AGI 2022, Seattle, WA, USA, August 19–22, 2022, Proceedings

Preface
Organization
Contents
Accepted for Poster Presentation
A General-Purpose Machine Reasoning Engine
	1 Introduction
	2 Related Work
	3 Machine Reasoning Engine
		3.1 Architecture
		3.2 Algorithms
	4 Results
		4.1 Command Line Interface
		4.2 Experiments
		4.3 Training Scripts and Results
	5 Conclusion
	References
COMFO: Multilingual Corpus for Opinion Mining
	1 Introduction
	2 Related Works
	3 Constitution of the COMFO Corpus
		3.1 Collection and Cleaning of Journalistic Comments
		3.2 Annotation by Lexical Approach
	4 Evaluation of the COMFO Corpus
		4.1 Evaluation of Experts
		4.2 Discussion
	5 Conclusion
	References
Information as Entanglement—A Framework for Artificial General Intelligence
	1 Introduction
		1.1 Computational Models as the Basis for Enabling Intelligent Behavior
		1.2 The Concept of Information is Central for Understanding Intelligence
		1.3 Proposed Unified Definitions of Information and Intelligence
		1.4 Approach and Structure
	2 A Unified Definition of Information
		2.1 Information as Entanglement
		2.2 Systems for Entanglement—Increasing Interactions Predictability
		2.3 Information—Measuring Agent Level of Entanglement
	3 Agents Need to Control Their Entanglement to Achieve Their Objectives
		3.1 Agents Predictability Ensures Their Performance
		3.2 Elements of a Control System
		3.3 Defining an Entanglement Control Signal
		3.4 Defining an Entanglement Controller
		3.5 Agent Architecture for Managing Entanglement Under Uncertainties
	4 The Information Digital Twin (IDT)
	5 Discussion—Information as a General Metric for Intelligence
	6 Conclusion
	References
Causal Analysis of Generic Time Series Data Applied for Market Prediction
	1 Introduction
		1.1 Background for This Work
		1.2 Overview of the Field
	2 Practical Approach
		2.1 Data Acquisition
		2.2 Analytical Framework
	3 Experimental Results
		3.1 Practical Applications
	4 Conclusion
	References
Dynamic and Evolving Neural Network for Event Discrimination
	1 Introduction
	2 Proposed AGI Model
		2.1 AGI Function
		2.2 AGI Characteristics
		2.3 Two Information Types in AGI
	3 Proposed DLM
		3.1 Proposed DLM Function
		3.2 Advantages of the Proposed DLM
		3.3 Task Hierarchy in the Proposed DLM
		3.4 Temporal Dimension of the DLM
		3.5 Related Work
		3.6 Contribution
	References
Hierarchical Temporal DNN and Associative Knowledge Representation
	1 Introduction
	2 The Proposed DLM
		2.1 Proposed DLM Function
	3 Associative Knowledge Representation Model (AKREM)
		3.1 Communication
		3.2 Detailed AKREM
		3.3 Memories in AKREM
	References
MARTI-4: New Model of Human Brain, Considering Neocortex and Basal Ganglia – Learns to Play Atari Game by Reinforcement Learning on a Single CPU
	1 Introduction
	2 Background
	3 Related Work
	4 Reinforcement Learning Environment
	5 Deep Control Architecture (DCA)
		5.1 Main Ideas of DCA
		5.2 DCA Structure
		5.3 Learning
		5.4 Prediction
	6 MARTI
		6.1 First Layer Hypercolumns
		6.2 Action Coder/Decoder
		6.3 Second Layer Hypercolumns
		6.4 Basal Ganglia
		6.5 Surprise Feeling and Inner Rewards
		6.6 Whole Cycle of Analysing/predicting
		6.7 World Model and Prediction Horizon
	7 Experiments
	8 Discussion
	9 Conclusion
	References
General-Purpose Minecraft Agents and Hybrid AGI
	1 Introduction
	2 Minecraft as Testbed for Cross-Paradigm AGI
	3 Brief Analysis of Human Player Behavior
	4 Universal Agents with Reasoning
	5 Minecraft Agent Design
	6 AGENt’s Capabilities
	7 Conclusion
	References
Graph Strategy for Interpretable Visual Question Answering
	1 Introduction
	2 Related Works
	3 Task Statement
	4 GS-VQA Model Overview
	5 Experiments
	6 Conclusion
	A Answering Procedure Algorithm
	B THOR-VQA Question Templates
	C THOR-VQA Question Counts
	References
Analogical Problem Solving in the Causal Cognitive Architecture
	1 Introduction
	2 Functioning of the Causal Cognitive Architecture 3 (CCA3)
		2.1 Input Sensory Vectors Shaping Modules
		2.2 Input Sensory Vectors Associations Modules
		2.3 Navigation Maps
		2.4 Sequential/Error Correcting Module
		2.5 Object Segmentation Gateway Module
		2.6 Causal Memory Module
		2.7 Navigation Module
	3 Analogical Feedback
		3.1 The Problem of Processing the Intermediate Results
		3.2 Analogical Feedback Demonstration Example
	4 Discussion
	References
A Biologically Plausible Graph Structure for AGI
	1 Introduction
		1.1 Flexibility
		1.2 Biological Plausibility
		1.3 Performance
	2 A Brief Introduction to Knowledge in Neurons
		2.1 The Information of Knowledge
		2.2 Biological Plausibility
		2.3 Important Conclusions from the Biological Perspective
	3 The Universal Knowledge Store (UKS)
		3.1 The Link
		3.2 The Thing
		3.3 Thing References
		3.4 Labels and Values
	4 Summary and Current Development
		4.1 Application 1, Perception: Learning by Correlation
		4.2 Application 2: Maze/Learning by Trial and Error
	References
The Delta Normal AGI
	1 Introduction
	2 Definitions
		2.1 The Topic-Region
		2.2 Compresssion and Expansion
		2.3 COFO
		2.4 Drivers
		2.5 Problems and Solutions
		2.6 Normalization
		2.7 Problem Solving Graph (PSD)
		2.8 Data Record Format
		2.9 Learning, Adaptation and Optimization
		2.10 Elements and Quarks
		2.11 Matching
	3 Natural Language Transformed to Sents
	4 Problem Solving
	5 Combinatorial Processes
	6 The Core Process of the AGI
		6.1 Input from the Internet
		6.2 Combinatory Sentence Making
		6.3 Making It Work
	7 Future Work
	8 Summary
	References
Purely Symbolic Induction of Structure
	1 Introduction
	2 From Graphs to Grammar
	3 Symbolic Learning
	4 Chunking/Tokenization
	5 Abstraction and Recursion
		5.1 Common Sense
	References
Accepted for Full Oral Presentation
Extended Subdomains: A Solution to a Problem of Hernández-Orallo and Dowe
	1 Introduction
	2 Background: The Hernández-Orallo and Dowe Problem
		2.1 The Generalized Hernández-Orallo and Dowe Problem
	3 Extending Subdomains to Solve the Hernández-Orallo and Dowe Problem
	4 Conclusion
	References
Versatility-Efficiency Index (VEI): Towards a Comprehensive Definition of Intelligence Quotient (IQ) for Artificial General Intelligence (AGI) Agents
	1 Introduction
	2 Versatility and the Legg-Hutter Definition
	3 Efficiency and the Pennachin-Goertzel Definition
		3.1 Complexity of Environments
		3.2 Wellness of Performance
	4 Versatility-Efficiency Index
	5 Conclusion
	References
Moral Space for Paraconsistent AGI
	1 Paraconsistency 2021
	2 Paraconsistent Ethics; the Gist
		2.1 The Two Takes on Paraconsistent Logic and Ethics
		2.2 Goertzel’s Two Arguments for Paraconsistent Logic of Morals
	3 Axio-Ontology 1 and 2 for Paraconsistent Ethics
		3.1 Paraconsistent Metaethical Ontology 1 (Human Limitations)
		3.2 Paraconsistent Metaethical Ontology 2 (Axiologically Grounded Inconsistency)
		3.3 Mixed Ontology 1/2
		3.4 Propinquities
	4 Positional Moral Paraconsistency
		4.1 Moral Pluralisms, Psychological (Greene/Haidt) and Philosophical (Ross)
		4.2 Dancy’s Moral Particularism
		4.3 Sen’s Socio-Economic Calculus of Agent-Relative Reasons
		4.4 Non-homogenous Moral Space: Sidgwick and Pargetter
	5 Conclusion: The Existential Twist to Paraconsistent Ethics
	References
PERI.2 Goes to PreSchool and Beyond, in Search of AGI
	1 Introduction
	2 Our Two Theoretical Pillars
		2.1 Pillar 1: Logic-Based AI and Cognitive Science
		2.2 Pillar 2: Psychometric AI
	3 The Goertzelian (et al.) Academic Road to AGI
	4 PERI.2 in Kindergarten
		4.1 Automated Reasoning of a Meta-forms Problem/Solution Pair
	5 PERI.2, Concretely: A Glimpse
	6 Related Work
		6.1 Remarks on NARS w.r.t Our Theoretical Pillars
	7 Are Harder Problems Computationally Feasible?
	8 Future Work: What About Compromised Perception?
	References
Reinforcement Learning with Information-Theoretic Actuation
	1 Introduction
	2 Preliminaries
	3 Information-Theoretic Actuation – Internal Actions
	4 Connecting Internal with External
	5 Discussion
	References
Homomorphisms Between Transfer, Multi-task, and Meta-learning Systems
	1 Introduction
	2 Abstract Learning Systems
	3 Transfer Learning Systems
	4 Multi-task and Meta-learning Systems
		4.1 Multi-task Learning
		4.2 Meta-learning
	5 Homomorphisms Between Learning Systems
		5.1 Discussion
	6 Conclusion
	References
Core and Periphery as Closed-System Precepts for Engineering General Intelligence
	1 Introduction
	2 Related Work
	3 Existing Precepts and Their Limits
	4 Outcomes and Requisite Variety
	5 Core and Periphery
		5.1 Definition
		5.2 Core and Periphery as Precepts
		5.3 Relevance
	6 Conclusion
	References
Toward Generating Natural-Language Explanations of Modal-Logic Proofs
	1 Introduction
	2 Cognitive Calculi
		2.1 A Micro Calculus: C
	3 NLG via Transformer Language Models
		3.1 Pegasus
		3.2 C and the Proof Domain
		3.3 Model Fine-Tuning
	4 Evaluation
		4.1 Example #1
		4.2 Example #2
		4.3 Example #3
		4.4 Example #4
		4.5 Overall
	5 Related Work
	6 Conclusion
	A  Fine-Tuning and Evaluation Implementation
	References
ONA for Autonomous ROS-Based Robots
	1 Introduction: A Reasoner Which Learns and Decides
	2 Non Axiomatic Logic (NAL)
	3 Practical Reasoning and Learning
	4 Demonstrated Capabilities
	5 Conclusion
	References
Generalized Identity Matching in NARS
	1 Introduction
	2 Methods
		2.1 OpenNARS for Applications
		2.2 Identity Match-to-Sample Task in NARS
		2.3 Experimental Setup
		2.4 NARS Examples from the Training Phase
	3 Results
		3.1 NARS Explanation of the Results
	4 Discussion
	References
Adaptive Multi-strategy Market-Making Agent for Volatile Markets
	1 Introduction
	2 Adaptive Multi-Strategy Agent
		2.1 Key Principles
		2.2 Implementation Details
		2.3 Agent/Strategy Assessment and Selection
		2.4 Algorithm
		2.5 Inventory Sharing Policy
	3 Experimental Setup
		3.1 Evaluation Environment
		3.2 Three Types of Historical Market Intervals
		3.3 Three Sets of Market Making Agents and Hodler
		3.4 Experimental Configurations
	4 Experimental results
		4.1 Performance Comparison by Interval
		4.2 Performance Comparison by Market Making Agent
		4.3 Possible Experimental Problems
	5 Further Improvements
	6 Conclusion and Future Work
	References
Toward a Comprehensive List of Necessary Abilities for Human Intelligence, Part 1: Constructing Knowledge
	1 Introduction
	2 Necessary Abilities for Human Cognition
	3 Are All Necessary for Intelligence and More Than Obviously so?
	4 Conclusion
	References
Toward a Comprehensive List of Necessary Abilities for Human Intelligence, Part 2: Using Knowledge
	1 Introduction
	2 Necessary Abilities for Human Cognition
	3 For Artificial General Intelligence, Are All Really Necessary?
	4 Conclusion
	References
What Can Nonhuman Animals, Children, and g Tell Us About Human-Level Artificial General Intelligence (AGI)?
	1 Introduction
	2 What Human-Level AGI Can Learn from Nonhuman Animals: Uniquely Human Cognitive Abilities
	3 From Human Intelligence Research
		3.1 What to Test
		3.2 Underlying Structure of Intelligence Test Scores
		3.3 What is g?
		3.4 Genetic and Brain Evidence for Intelligence and g
		3.5 What can AGI Learn from g?
	4 What We Can Learn from Children: Progression with Exception
	5 Conclusions
	References
Cognitive Architecture for Co-evolutionary Hybrid Intelligence
	1 Data-Centric AI Crisis
	2 Co-evolutionary Hybrid Intelligence
	3 Cognitive Architectures: State of the Art
	4 Co-evolutionary Hybrid Intelligence Cognitive Architecture
	5 Conclusion
	References
An Approach to Generation Triggers for Parrying Backdoor in Neural Networks
	1 Introduction
	2 Description of the Approach
	3 Experiment
	4 Discussion
	5 Conclusion
	References
The Learning Agent Triangle: Towards a Unified Disambiguation of the AGI Challenge
	1 Introduction
	2 Importance of a Complete Design Description
	3 The Learning Agent Triangle
		3.1 The Architecture
		3.2 The Objective Goal
		3.3 The Optimization Method
		3.4 Conditioned by the Computational Limitations
	4 Discussion
	A Study Case: Y. Bengio Discusses His Ideas with L. Fridman
	References
Maze Learning Using a Hyperdimensional Predictive Processing Cognitive Architecture
	1 Introduction
	2 Neural Building Blocks
		2.1 Neural Generative Coding (NGC)
		2.2 Memory
	3 The CogNGen Cognitive Architecture
		3.1 Perceptual Modules
		3.2 Procedural Memory and Motor Control
		3.3 Long-Term Memory
	4 Experimental Results
		4.1 The Mini GridWorld Problem
		4.2 Baseline Models
		4.3 Experimental Results
	5 Conclusions
	References
Market Prediction as a Task for AGI Agents
	1 Introduction
	2 Related Work
	3 Task
	4 Data Set
	5 Methods
	6 Results
	7 Conclusion
	References
Monte Carlo Bias Correction in Q-Learning
	1 Introduction
	2 Preliminaries
	3 Related Works
	4 Monte Carlo Bias Correction
	5 Experiments
		5.1 Roulette
		5.2 Grid World
		5.3 Taxi
	6 Summary
	References
Free Will Belief as a Consequence of Model-Based Reinforcement Learning
	1 Introduction
	2 Reinforcement Learning
	3 What is Freedom?
	4 Why Do We Believe We are Free?
	5 Discussion
	References
Thoughts on Architecture
	1 Introduction
	2 Historical Context
	3 Defining Architecture
	4 Multilayered Architectures
	5 Theoretical Computational Architectures
	6 Architectural Exploration
	7 Conclusion
	References
On the Possibility of Regulation of Human Emotions via Multimodal Social Interaction with an Embodied Agent Controlled by eBICA-Based Emotional Interaction Model
	1 Introduction
	2 The Concept and Implementation Plan
		2.1 Cognitive Architecture
		2.2 Interaction Model
		2.3 Behavioral Paradigms
	3 Outcomes of Preliminary Studies
	4 Discussion
	References
QKSA: Quantum Knowledge Seeking Agent
	1 Introduction
	2 Framework Features
		2.1 Representations of General Quantum Environments
		2.2 Process Tomography Algorithms for Modeling
		2.3 Computational Resource-Bounded Algorithmic Cost
		2.4 Mutating Meta-learning Hyper-parameter Embedded in a Quine
	3 QKSA Formalism
	4 Conclusion
	References
Elements of Active Continuous Learning and Uncertainty Self-awareness: A Narrow Implementation for Face and Facial Expression Recognition
	1 Introduction
	2 Proposed Solution
		2.1 Uncertainty Meta-learning
		2.2 Active CNN Ensemble Learning and Life-Time SNN Learning
	3 Data Set
	4 Experiments
		4.1 Trusted Accuracy Metrics
	5 Results
	6 Discussion, Conclusions, and Future Work
	References
Thrill-K Architecture: Towards a Solution to the Problem of Knowledge Based Understanding
	1 Introduction: The Rise of Cognitive AI
	2 Dimensions of Knowledge
	3 From Knowledge to Understanding
	4 Information-Centric Classification of AI Systems
		4.1 Key Elements of a Class 3 System
	5 Thrill-K: A Blueprint for Hybrid Machine Intelligence
		5.1 Thrill-K’s Three Levels of Knowledge
		5.2 Scalability of a Thrill-K System
	6 Conclusion: Thrill-K’s Contribution to Robustness, Adaptation and Higher Intelligence
	References
Grammar Induction - Experimental Results
	1 Introduction
		1.1 Software Infrastructure
	2 Pair Counting
	3 MST Parsing
	4 Grammatically Similar Words
	5 Clustering
	6 Conclusion
	References
Brain Principles Programming
	1 Introduction
	2 Part I. Basic Theories and Formal Models
		2.1 Basic Elements of Perception and the World
		2.2 Probabilistic Formal Concepts
		2.3 “Intelligent Object” and “Intelligent Function”
	3 Part II. Brain Principles Programming Formalization
		3.1 The Principle of the Complexity Generation
		3.2 The Principle of the Relationship
		3.3 The Principle of Approximation to the Essence
		3.4 The Principle of Locality-Distribution
		3.5 The Principle of Heaviness
		3.6 Conclusions
	References
A Meta-Probabilistic-Programming Language for Bisimulation of Probabilistic and Non-Well-Founded Type Systems
	1 Introduction
	2 Labeled Metagraphs as a Guarded Recursive Datatype
	3 M as the Final Coalgebra of a Labeled Transition System
		3.1 Labeled Transition System Based on Metagraph Rewriting
		3.2 M-Interpretation as Metagraph Dynamics
	4 Bisimulation of Type Systems in M
		4.1 Simply Typed Lambda Calculus
		4.2 Pure Type Systems
		4.3 Probabilistic Dependent Types
	5 Implementation of Bisimulation Proof in a Guarded Cubical Type Theory Type Checker
	6 Discussion
	A  Proof of Bisimulation for Small-Scale Type System in a Guarded Cubical Type Theory Type Checker
	References
Artificial Open World for Evaluating AGI: A Conceptual Design
	1 Introduction
	2 What Intelligence is
	3 Evaluation
		3.1 The Trap of Developer's Experience
		3.2 Overall Principles
		3.3 Conceptual Design of Artificial Open World
	4 Discussion
	References
Ownability of AGI
	1 Introduction
	2 Proposals for Establishing Ownership
	3 Obstacles to Ownership
	4 Conclusions
	References
Author Index