Building Better Minds: Artificial General Intelligence via the CogPrime Architecture

Part 1 A Path to Beneficial Artificial General Intelligence
	Introduction
		AI Returns to Its Roots
		The Secret Sauce
		Extraordinary Proof?
		Potential Approaches to AGI
			Can Digital Computers Really Be Intelligent?
		Five Key Words
			Memory and Cognition in CogPrime
		Virtually and Robotically Embodied AI
		Language Learning
		AGI Ethics
		Structure of the Book
		Key Claims of the Book
	Section I Artificial and Natural General Intelligence
		What Is Human-Like General Intelligence?
			Introduction
				What Is General Intelligence?
				What Is Human-like General Intelligence?
			Commonly Recognized Aspects of Human-like Intelligence
			Further Characterizations of Humanlike Intelligence
				Competencies Characterizing Human-like Intelligence
				Gardner's Theory of Multiple Intelligences
				Newell's Criteria for a Human Cognitive Architecture
			Preschool as a View into Human-like General Intelligence
				Design for an AGI Preschool
			Integrative and Synergetic Approaches to Artificial General Intelligence
				Achieving Humanlike Intelligence via Cognitive Synergy
		A Patternist Philosophy of Mind
			Introduction
			Some Patternist Principles
			Cognitive Synergy
			The General Structure of Cognitive Dynamics: Analysis and Synthesis
				Component-Systems and Self-Generating Systems
				Analysis and Synthesis
				The Dynamic of Iterative Analysis and Synthesis
				Self and Focused Attention as Approximate Attractors of the Dynamic of Iterated Forward/analysis
				Conclusion
			Perspectives on Machine Consciousness
			Postscript: Formalizing Pattern
		Brief Survey of Cognitive Architectures
			Introduction
			Symbolic Cognitive Architectures
				SOAR
				ACT-R
				Cyc and Texai
				NARS
				GLAIR and SNePS
			Emergentist Cognitive Architectures
				DeSTIN: A Deep Reinforcement Learning Approach to AGI
				Developmental Robotics Architectures
			Hybrid Cognitive Architectures
				Neural versus Symbolic; Global versus Local
			Globalist versus Localist Representations
				CLARION
				The Society of Mind and the Emotion Machine
				DUAL
				4D/RCS
				PolyScheme
				Joshua Blue
				LIDA
				The Global Workspace
				The LIDA Cognitive Cycle
				Psi and MicroPsi
				The Emergence of Emotion in the Psi Model
				Knowledge Representation, Action Selection and Planning in Psi
				Psi versus CogPrime
		A Generic Architecture of Human-Like Cognition
			Introduction
			Key Ingredients of the Integrative Human-Like Cognitive Architecture Diagram
			An Architecture Diagram for Human-Like General Intelligence
			Interpretation and Application of the Integrative Diagram
		A Brief Overview of CogPrime
			Introduction
			High-Level Architecture of CogPrime
			Current and Prior Applications of OpenCog
				Transitioning from Virtual Agents to a Physical Robot
			Memory Types and Associated Cognitive Processes in CogPrime
				Cognitive Synergy in PLN
			Goal-Oriented Dynamics in CogPrime
			Analysis and Synthesis Processes in CogPrime
			Conclusion
		Section II Toward a General Theory of General Intelligence
			A Formal Model of Intelligent Agents
				Introduction
				A Simple Formal Agents Model (SRAM)
					Goals
					Memory Stores
					The Cognitive Schematic
				Toward a Formal Characterization of Real-World General Intelligence
					Biased Universal Intelligence
					Connecting Legg and Hutter's Model of Intelligent Agents to the Real World
					Pragmatic General Intelligence
					Incorporating Computational Cost
					Assessing the Intelligence of Real-World Agents
				Intellectual Breadth: Quantifying the Generality of an Agent's Intelligence
				Conclusion
			Cognitive Synergy
				Cognitive Synergy
				Cognitive Synergy
				Cognitive Synergy in CogPrime
					Cognitive Processes in CogPrime
				Some Critical Synergies
				The Cognitive Schematic
				Cognitive Synergy for Procedural and Declarative Learning
					Cognitive Synergy in MOSES
					Cognitive Synergy in PLN
				Is Cognitive Synergy Tricky?
					The Puzzle: Why Is It So Hard to Measure Partial Progress Toward Human-Level AGI?
					A Possible Answer: Cognitive Synergy is Tricky!
					Conclusion
			General Intelligence in the Everyday Human World
				Introduction
				Some Broad Properties of the Everyday World That Help Structure Intelligence
				Embodied Communication
					Generalizing the Embodied Communication Prior
				Naive Physics
					Objects, Natural Units and Natural Kinds
					Events, Processes and Causality
					Stuffs, States of Matter, Qualities
					Surfaces, Limits, Boundaries, Media
					What Kind of Physics Is Needed to Foster Human-like Intelligence?
				Folk Psychology
					Motivation, Requiredness, Value
				Body and Mind
				The Extended Mind and Body
				Conclusion
			A Mind-World Correspondence Principle
				Introduction
				What Might a General Theory of General Intelligence Look Like?
				Steps Toward A (Formal) General Theory of General Intelligence
				The Mind-World Correspondence Principle
				How Might the Mind-World Correspondence Principle Be Useful?
				Conclusion
	Section III Cognitive and Ethical Development
		Stages of Cognitive Development
			Introduction
			Piagetan Stages in the Context of a General Systems Theory of Development
			Piaget's Theory of Cognitive Development
				Perry's Stages
				Keeping Continuity in Mind
			Piaget's Stages in the Context of Uncertain Inference
				The Infantile Stage
				The Concrete Stage
				The Formal Stage
				The Reflexive Stage
		The Engineering and Development of Ethics
			Introduction
			Review of Current Thinking on the Risks of AGI
			The Value of an Explicit Goal System
			Ethical Synergy
				Stages of Development of Declarative Ethics
				Stages of Development of Empathic Ethics
				An Integrative Approach to Ethical Development
				Integrative Ethics and Integrative AGI
			Clarifying the Ethics of Justice: Extending the Golden Rule in to a Multifactorial Ethical Model
				The Golden Rule and the Stages of Ethical Development
				The Need for Context-Sensitivity and Adaptiveness in Deploying Ethical Principles
			The Ethical Treatment of AGIs
				Possible Consequences of Depriving AGIs of Freedom
				AGI Ethics as Boundaries Between Humans and AGIs Become Blurred
			Possible Benefits of Closely Linking AGIs to the Global Brain
				The Importance of Fostering Deep, Consensus-Building Interactions Between People with Divergent Views
			Possible Benefits of Creating Societies of AGIs
			AGI Ethics As Related to Various Future Scenarios
				Capped Intelligence Scenarios
				Superintelligent AI: Soft-Takeoff Scenarios
				Superintelligent AI: Hard-Takeoff Scenarios
				Global Brain Mindplex Scenarios
			Conclusion: Eight Ways to Bias AGI Toward Friendliness
				Encourage Measured Co-Advancement of AGI Software and AGI Ethics Theory
				Develop Advanced AGI Sooner Not Later
	Section IV Networks for Explicit and Implicit Knowledge Representation
		Local, Global and Glocal Knowledge Representation
			Introduction
			Localized Knowledge Representation using Weighted, Labeled Hypergraphs
				Weighted, Labeled Hypergraphs
			Atoms: Their Types and Weights
				Some Basic Atom Types
				Variable Atoms
				Logical Links
				Temporal Links
				Associative Links
				Procedure Nodes
				Links for Special External Data Types
				Truth Values and Attention Values
			Knowledge Representation via Attractor Neural Networks
				The Hopfield neural net model
				Knowledge Representation via Cell Assemblies
			Neural Foundations of Learning
				Hebbian Learning
				Virtual Synapses and Hebbian Learning Between Assemblies
				Neural Darwinism
			Glocal Memory
				A Semi-Formal Model of Glocal Memory
				Glocal Memory in the Brain
				Glocal Hopfield Networks
				Neural-Symbolic Glocality in CogPrime
		Representing Implicit Knowledge via Hypergraphs
			Introduction
			Key Vertex and Edge Types
			Derived Hypergraphs
				SMEPH Vertices
				SMEPH Edges
			Implications of Patternist Philosophy for Derived Hypergraphs of Intelligent Systems
				SMEPH Principles in CogPrime
		Emergent Networks of Intelligence
			Introduction
			Small World Networks
			Dual Network Structure
				Hierarchical Networks
				Associative, Heterarchical Networks
				Dual Networks
	Section V A Path to Human-Level AGI
		AGI Preschool
			Introduction
				Contrast to Standard AI Evaluation Methodologies
			Elements of Preschool Design
			Elements of Preschool Curriculum
				Preschool in the Light of Intelligence Theory
			Task-Based Assessment in AGI Preschool
			Beyond Preschool
			Issues with Virtual Preschool Engineering
				Integrating Virtual Worlds with Robot Simulators
				BlocksNBeads World
		A Preschool-Based Roadmap to Advanced AGI
			Introduction
			Measuring Incremental Progress Toward Human-Level AGI
			Conclusion
		Advanced Self-Modification: A Possible Path to Superhuman AGI
			Introduction
			Cognitive Schema Learning
			Self-Modification via Supercompilation
				Three Aspects of Supercompilation
				Supercompilation for Goal-Directed Program Modification
			Self-Modification via Theorem-Proving
	Part 2 An Architecture for Beneficial Artificial General Intelligence
		Section VI Architectural and Representational Mechanisms
			The OpenCog Framework
				Introduction
				The OpenCog Architecture
					OpenCog and Hardware Models
					The Key Components of the OpenCog Framework
				The AtomSpace
					The Knowledge Unit: Atoms
					AtomSpace Requirements and Properties
					Accessing the Atomspace
					Persistence
					Specialized Knowledge Stores
				MindAgents: Cognitive Processes
					A Conceptual View of CogPrime Cognitive Processes
					Implementation of MindAgents
					Tasks
					Scheduling of MindAgents and Tasks in a Unit
					The Cognitive Cycle
				Distributed AtomSpace and Cognitive Dynamics
					Distributing the AtomSpace
					Distributed Processing
			Knowledge Representation Using the Atomspace
				Introduction
				Denoting Atoms
					Meta-Language
					Denoting Atoms
				Representing Functions and Predicates
					Execution Links
					Denoting Schema and Predicate Variables
					Variable and Combinator Notation
					Inheritance Between Higher-Order Types
					Advanced Schema Manipulation
			Representing Procedural Knowledge
				Introduction
				Representing Programs
				Representational Challenges
				What Makes a Representation Tractable?
				The Combo Language
				Normal Forms Postulated to Provide Tractable Representations
					A Simple Type System
					Boolean Normal Form
					Number Normal Form
					List Normal Form
					Tuple Normal Form
					Enum Normal Form
					Function Normal Form
					Action Result Normal Form
				Program Transformations
					Reductions
					Neutral Transformations
					Non-Neutral Transformations
				Interfacing Between Procedural and Declarative Knowledge
					Programs Manipulating Atoms
				Declarative Representation of Procedures
		Section VII The Cognitive Cycle
			Emotion, Motivation, Attention and Control
				Introduction
				A Quick Look at Action Selection
				Psi in CogPrime
				Implementing Emotion Rules Atop Psi's Emotional Dynamics
					Grounding the Logical Structure of Emotions in the Psi Model
				Goals and Contexts
					Goal Atoms
				Context Atoms
				Ubergoal Dynamics
					Implicit Ubergoal Pool Modification
					Explicit Ubergoal Pool Modification
				Goal Formation
				Goal Fulfillment and Predicate Schematization
				Context Formation
				Execution Management
				Goals and Time
			Attention Allocation
				Introduction
				Semantics of Short and Long Term Importance
					The Precise Semantics of STI and LTI
					STI, STIFund, and Juju
					Formalizing LTI
					Applications of LTIburst versus LTIcont
				Defining Burst LTI in Terms of STI
				Valuing LTI and STI in terms of a Single Currency
				Economic Attention Networks
					Semantics of Hebbian Links
					Explicit and Implicit Hebbian Relations
				Dynamics of STI and LTI Propagation
					ECAN Update Equations
					ECAN as Associative Memory
				Glocal Economic Attention Networks
					Experimental Explorations
				Long-Term Importance and Forgetting
				Attention Allocation via Data Mining on the System Activity Table
				Schema Credit Assignment
				Interaction between ECANs and other CogPrime Components
					Use of PLN and Procedure Learning to Help ECAN
					Use of ECAN to Help Other Cognitive Processes
				MindAgent Importance and Scheduling
				Information Geometry for Attention Allocation
					Brief Review of Information Geometry
					Information-Geometric Learning for Recurrent Networks: Extending the ANGL Algorithm
					Information Geometry for Economic Attention Allocation: A Detailed Example
			Economic Goal and Action Selection
				Introduction
				Transfer of STI ``Requests for Service'' Between Goals
				Feasibility Structures
				Goal Based Schema Selection
					A Game-Theoretic Approach to Action Selection
				SchemaActivation
				GoalBasedSchemaLearning
			Integrative Procedure Evaluation
				Introduction
				Procedure Evaluators
					Simple Procedure Evaluation
					Effort Based Procedure Evaluation
					Procedure Evaluation with Adaptive Evaluation Order
				The Procedure Evaluation Process
					Truth Value Evaluation
					Schema Execution
		Section VIII Perception and Action
			Perceptual and Motor Hierarchies
				Introduction
				The Generic Perception Process
					The ExperienceDB
				Interfacing CogPrime with a Virtual Agent
					Perceiving the Virtual World
					Acting in the Virtual World
				Perceptual Pattern Mining
					Input Data
					Transaction Graphs
					Spatiotemporal Conjunctions
					The Mining Task
				The Perceptual-Motor Hierarchy
				Object Recognition from Polygonal Meshes
					Algorithm Overview
					Recognizing PersistentPolygonNodes (PPNodes) from PolygonNodes
					Creating Adjacency Graphs from PPNodes
					Clustering in the Adjacency Graph.
					Discussion
				Interfacing the Atomspace with a Deep Learning Based Perception-Action Hierarchy
					Hierarchical Perception Action Networks
					Declarative Memory
					Sensory Memory
					Procedural Memory
					Episodic Memory
					Action Selection and Attention Allocation
				Multiple Interaction Channels
			Integrating CogPrime with a Compositional Spatiotemporal Deep Learning Network
				Introduction
				Integrating CSDLNs with Other AI Frameworks
				Semantic CSDLN for Perception Processing
				Semantic CSDLN for Motor and Sensorimotor Processing
				Connecting the Perceptual and Motoric Hierarchies with a Goal Hierarchy
			Making DeSTIN Representationally Transparent
				Introduction
				Review of DeSTIN Architecture and Dynamics
					Beyond Gray-Scale Vision
				Uniform DeSTIN
					Translation-Invariant DeSTIN
					Mapping States of Translation-Invariant DeSTIN into the Atomspace
					Scale-Invariant DeSTIN
					Rotation Invariant DeSTIN
					Temporal Perception
				Interpretation of DeSTIN's Activity
					DeSTIN's Assumption of Hierarchical Decomposability
					Distance and Utility
				Benefits and Costs of Uniform DeSTIN
				Imprecise Probability as a Strategy for Linking CogPrime and DeSTIN
					Visual Attention Focusing
					Using Imprecise Probabilities to Guide Visual Attention Focusing
					Sketch of Application to DeSTIN
			Bridging the Symbolic/Subsymbolic Gap
				Introduction
				Simplified OpenCog Workflow
				Integrating DeSTIN and OpenCog
					Mining Patterns from DeSTIN States
					Probabilistic Inference on Mined Hypergraphs
					Insertion of OpenCog-Learned Predicates into DeSTIN's Pattern Library
				Multisensory Integration, and Perception-Action Integration
					Perception-Action Integration
					Thought-Experiment: Eye-Hand Coordination
				Conclusion
		Section IX Procedure Learning
			Procedure Learning as Program Learning
				Introduction
					Program Learning
				Representation-Building
				Specification Based Procedure Learning
			Learning Procedures via Imitation, Reinforcement and Correction
				Introduction
				IRC Learning
					A Simple Example of Imitation/Reinforcement Learning
					A Simple Example of Corrective Learning
				IRC Learning in the PetBrain
					Introducing Corrective Learning
				Applying A Similar IRC Methodology to Spontaneous Learning
			Procedure Learning via Adaptively Biased Hillclimbing
				Introduction
				Hillclimbing
				Entity and Perception Filters
					Entity filter
					Entropy perception filter
				Using Action Sequences as Building Blocks
				Automatically Parametrizing the Program Size Penalty
					Definition of the complexity penalty
					Parameterizing the complexity penalty
					Definition of the Optimization Problem
				Some Simple Experimental Results
				Conclusion
			Probabilistic Evolutionary Procedure Learning
				Introduction
					Explicit versus Implicit Evolution in CogPrime
				Estimation of Distribution Algorithms
				Competent Program Evolution via MOSES
					Statics
					Dynamics
					Architecture
					Example: Artificial Ant Problem
					Discussion
					Conclusion
				Supplying Evolutionary Learning with Long-Term Memory
				Hierarchical Program Learning
					Hierarchical Modeling of Composite Procedures in the AtomSpace
					Identifying Hierarchical Structure In Combo trees via MetaNodes and Dimensional Embedding
				Fitness Function Estimation via Integrative Intelligence
Section X Declarative Learning
	Probabilistic Logic Networks
		Introduction
		First Order Probabilistic Logic Networks
			Core FOPLN Relationships
			PLN Truth Values
			Auxiliary FOPLN Relationships
			PLN Rules and Formulas
		Higher-Order PLN
			Reducing HOPLN to FOPLN
		Predictive Implication and Attraction
		Confidence Decay
			An Example
		Why is PLN a Good Idea?
	Spatiotemporal Inference
		Introduction
		Related Work on Spatio-temporal Calculi
		Uncertainty with Distributional Fuzzy Values
		Spatio-temporal Inference in PLN
		Examples
			Spatiotemporal Rules
			The Laptop is Safe from the Rain
			Fetching the Toy Inside the Upper Cupboard
		An Integrative Approach to Planning
	Adaptive, Integrative Inference Control
		Introduction
		High-Level Control Mechanisms
			The Need for Adaptive Inference Control
		Inference Control in PLN
			The Evaluator Choice Problem as a Bandit Problem
			Chains of Thought
		Inference Pattern Mining
		Hebbian Inference Control
		Evolution As an Inference Control Scheme
		Incorporating Other Cognitive Processes Into Inference
		PLN and Bayes Nets
	Pattern Mining
		Introduction
		Finding Interesting Patterns via Program Learning
		Pattern Mining via Frequent/Surprising Subgraph Mining
		Fishgram
			Example Patterns
			The Fishgram Algorithm
			Preprocessing
			Search Process
			Comparison to other algorithms
	Speculative Concept Formation
		Introduction
		Evolutionary Concept Formation
		Conceptual Blending
			Outline of a CogPrime Blending Algorithm
			Another Example of Blending
		Clustering
		Concept Formation via Formal Concept Analysis
			Calculating Membership Degrees of New Concepts
			Forming New Attributes
			Iterating the Fuzzy Concept Formation Process
Section XI Integrative Learning
	Dimensional Embedding
		Introduction
		Link Based Dimensional Embedding
		Harel and Koren's Dimensional Embedding Algorithm
			Step 1: Choosing Pivot Points
			Step 2: Similarity Estimation
			Step 3: Embedding
		Embedding Based Inference Control
		Dimensional Embedding and InheritanceLinks
	Mental Simulation and Episodic memory
		Introduction
		Internal Simulations
		Episodic Memory
	Integrative Procedure Learning
		Introduction
			The Diverse Technicalities of Procedure Learning in CogPrime
		Preliminary Comments on Procedure Map Encapsulation and Expansion
		Predicate Schematization
			A Concrete Example
		Concept-Driven Schema and Predicate Creation
			Concept-Driven Predicate Creation
			Concept-Driven Schema Creation
		Inference-Guided Evolution of Pattern-Embodying Predicates
			Rewarding Surprising Predicates
			A More Formal Treatment
		PredicateNode Mining
		Learning Schema Maps
			Goal-Directed Schema Evolution
		Occam's Razor
	Map Formation
		Introduction
		Map Encapsulation
		Atom and Predicate Activity Tables
		Mining the AtomSpace for Maps
			Frequent Itemset Mining for Map Mining
			Evolutionary Map Detection
		Map Dynamics
		Procedure Encapsulation and Expansion
			Procedure Encapsulation in More Detail
			Procedure Encapsulation in the Human Brain
		Maps and Focused Attention
		Recognizing And Creating Self-Referential Structures
			Encouraging the Recognition of Self-Referential Structures in the AtomSpace
Section XII Communication Between Human and Artificial Minds
	Communication Between Artificial Minds
		Introduction
		A Simple Example Using a PsyneseVocabulary Server
			The Psynese Match Schema
		Psynese as a Language
		Psynese Mindplexes
			AGI Mindplexes
		Psynese and Natural Language Processing
			Collective Language Learning
	Natural Language Comprehension
		Introduction
		Linguistic Atom Types
		The Comprehension and Generation Pipelines
		Parsing with Link Grammar
			Link Grammar vs. Phrase Structure Grammar
		The RelEx Framework for Natural Language Comprehension
			RelEx2Frame: Mapping Syntactico-Semantic Relationships into FrameNet Based Logical Relationships
			A Priori Probabilities For Rules
			Exclusions Between Rules
			Handling Multiple Prepositional Relationships
			Comparatives and Phantom Nodes
		Frame2Atom
			Examples of Frame2Atom
			Issues Involving Disambiguation
		Link2Atom: A Semi-Supervised Alternative to RelEx and RelEx2Frame
		Mapping Link Parses into Atom Structures
			Example Training Pair
		Making a Training Corpus
			Leveraging RelEx to Create a Training Corpus
			Making an Experience Based Training Corpus
			Unsupervised, Experience Based Corpus Creation
		Limiting the Degree of Disambiguation Attempted
		Rule Format
			Example Rule
		Rule Learning
		Creating a Cyc-Like Database via Text Mining
		PROWL Grammar
			Brief Review of Word Grammar
			Word Grammar's Logical Network Model
			Link Grammar Parsing vs Word Grammar Parsing
			Contextually Guided Greedy Parsing and Generation Using Word Link Grammar
		Aspects of Language Learning
			Word Sense Creation
			Feature Structure Learning
			Transformation and Semantic Mapping Rule Learning
		Experiential Language Learning
		Which Path(s) Forward?
	Natural Language Generation
		Introduction
		SegSim for Sentence Generation
			NLGen: Example Results
		Experiential Learning of Language Generation
		Atom2Link
		Conclusion
	Embodied Language Processing
		Introduction
		Semiosis
		Teaching Gestural Communication
		Simple Experiments with Embodiment and Anaphor Resolution
		Simple Experiments with Embodiment and Question Answering
			Preparing/Matching Frames
			Frames2RelEx
			Example of the Question Answering Pipeline
			Example of the PetBrain Language Generation Pipeline
		The Prospect of Massively Multiplayer Language Teaching
	Natural Language Dialogue
		Introduction
			Two Phases of Dialogue System Development
		Speech Act Theory and its Elaboration
		Speech Act Schemata and Triggers
			Notes Toward Example SpeechActSchema
		Probabilistic Mining of Trigger contexts
		Conclusion
Section XIII From Here to AGI
	Summary of Argument for the CogPrime Approach
		Introduction
		Multi-Memory Systems
		Perception, Action and Environment
		Developmental Pathways
		Knowledge Representation
		Cognitive Processes
			Uncertain Logic for Declarative Knowledge
			Program Learning for Procedural Knowledge
			Attention Allocation
			Internal Simulation and Episodic Knowledge
			Low-Level Perception and Action
			Goals
		Fulfilling the ``Cognitive Equation''
		Occam's Razor
			Mind Geometry
		Cognitive Synergy
			Synergies that Help Inference
		Synergies that Help MOSES
			Synergies that Help Attention Allocation
			Further Synergies Related to Pattern Mining
			Synergies Related to Map Formation
		Emergent Structures and Dynamics
		Ethical AGI
		Toward Superhuman General Intelligence
			Conclusion
	Build Me Something I Haven't Seen: A CogPrime Thought Experiment
		Introduction
		Roles of Selected Cognitive Processes
		A Semi-Narrative Treatment
		Conclusion
	Glossary
	Steps Toward a Formal Theory of Cognitive Structure and Dynamics
		Introduction
		Modeling Memory Types Using Category Theory
			The Category of Procedural Memory
			The Category of Declarative Memory
			The Category of Episodic Memory
			The Category of Intentional Memory
			The Category of Attentional Memory
		Modeling Memory Type Conversions Using Functors
			Converting Between Declarative and Procedural Knowledge
			Symbol Grounding: Converting Between Episodic and Declarative Knowledge
			Converting Between Episodic and Procedural Knowledge
			Converting Intentional or Attentional Knowledge into Declarative or Procedural Knowledge
			Converting Episodic Knowledge into Intentional or Attentional Knowledge
		Metrics on Memory Spaces
			Information Geometry on Memory Spaces
			Algorithmic Distance on Memory Spaces
		Three Hypotheses About the Geometry of Mind
			Hypothesis 1: Syntax-Semantics Correlation
			Hypothesis 2: Cognitive Geometrodynamics
			Hypothesis 3: Cognitive Synergy
		Next Steps in Refining These Ideas
		Returning to Our Basic Claims about CogPrime
	Emergent Reflexive Mental Structures
		Introduction
		Hypersets and Patterns
			Hypersets as Patterns in Physical or Computational Systems
		A Hyperset Model of Reflective Consciousness
		A Hyperset Model of Will
			In What Sense Is Will Free?
			Connecting Will and Consciousness
		A Hyperset Model of Self
		Validating Hyperset Models of Experience
		Implications for Practical Work on Machine Consciousness
			Attentional Focus in CogPrime
			Maps and Focused Attention in CogPrime
			Reflective Consciousness, Self and Will in CogPrime
			Encouraging the Recognition of Self-Referential Structures in the AtomSpace
		Algebras of the Social Self
		The Intrinsic Sociality of the Self
		Mirror Neurons and Associated Neural Systems
			Mirror Systems
		Quaternions and Octonions
		Modeling Mirrorhouses Using Quaternions and Octonions
		Specific Instances of Mental Mirrorhousing
		Mirroring in Development
		Concluding Remarks
	GOLEM: Toward an AGI Meta-Architecture Enabling Both Goal Preservation and Radical Self-Improvement
		Introduction
		The Goal Oriented Learning Meta-Architecture
			Optimizing the GoalEvaluator
			Conservative Meta-Architecture Preservation
		The Argument For GOLEM's Steadfastness
		A Partial Formalization of the Architecture and Steadfastness Argument
			Toward a Formalization of GOLEM
			Some Conjectures about GOLEM
		Comparison to a Reinforcement Learning Based Formulation
		Specifying the Letter and Spirit of Goal Systems (Are Both Difficult Tasks)
		A More Radically Self-Modifying GOLEM
		Concluding Remarks
	Lojban++: A Novel Linguistic Mechanism for Teaching AGI Systems
		Introduction
		Lojban versus Lojban++
		Some Simple Examples
		The Need for Lojban Software
		Lojban and Inference
			Lojban versus Predicate Logic
		Conclusion
		Postscript: Basic Principles for Using English Words in Lojban++
		Syntax-based Argument Structure Conventions for English Words
		Semantics-based Argument Structure Conventions for English Words
		Lojban gismu of clear use within Lojban++
		Special Lojban++ cmavo
			qui
			it , quu
			quay
	PLN and the Brain
		How Might Probabilistic Logic Networks Emerge from Neural Structures and Dynamics?
		Avoiding Issues with Circular Inference
		Neural Representation of Recursion and Abstraction
	Possible Worlds Semantics and Experiential Semantics
		Introduction
		Inducing a Distribution over Predicates and Concepts
		Grounding Possible Worlds Semantics in Experiential Semantics
		Reinterpreting Indefinite Probabilities
			Reinterpreting Indefinite Quantifiers
		Specifying Complexity for Intensional Inference
		Reinterpreting Implication between Inheritance Relationships
		Conclusion
	Propositions About Environments in Which CogPrime Components are Useful
		Propositions about MOSES
			Proposition: ENF Helps to Guide Syntax-Based Program Space Search
			Demes are Useful if Syntax/Semantics Correlations in Program Space Have a Small Scale
			Probabilistic Program Tree Modeling Helps in the Presence of Cross-Modular Dependencies
			Relating ENF to BOA
			Conclusion Regarding Speculative MOSES Theory
		Propositions About CogPrime
			When PLN Inference Beats BOA
			Conditions for the Usefulness of Hebbian Inference Control
			Clustering-together of Smooth Theorems
			When PLN is Useful Within MOSES
			When MOSES is Useful Within PLN
			On the Smoothness of Some Relevant Theorems
			Recursive Use of ``MOSES+PLN'' to Help With Attention Allocation
			The Value of Conceptual Blending
			A Justification of Map Formation
		Concluding Remarks
		References