Engineering General Intelligence, Part 1: A Path to Advanced AGI via Embodied Learning and Cognitive Synergy (2014) .. by Ben Goertzel, etc
Contents
1 Introduction … 1
1.1 AI Returns to Its Roots … 1
1.2 AGI Versus Narrow AI … 2
1.3 CogPrime… 3
1.4 The Secret Sauce … 4
1.5 Extraordinary Proof? … 5
1.6 Potential Approaches to AGI … 6
1.6.1 Build AGI from Narrow AI … 6
1.6.2 Enhancing Chatbots… 7
1.6.3 Emulating the Brain … 7
1.6.4 Evolve an AGI … 8
1.6.5 Derive an AGI Design Mathematically … 8
1.6.6 Use Heuristic Computer Science Methods … 9
1.6.7 Integrative Cognitive Architecture… 9
1.6.8 Can Digital Computers Really Be Intelligent? … 10
1.7 Five Key Words … 11
1.7.1 Memory and Cognition in CogPrime … 11
1.8 Virtually and Robotically Embodied AI… 13
1.9 Language Learning … 13
1.10 AGI Ethics… 14
1.11 Structure of the Book … 15
1.12 Key Claims of the Book … 15
Part I Overview of the CogPrime Architecture
2 A Brief Overview of CogPrime … 21
2.1 Introduction … 21
2.2 High-Level Architecture of CogPrime … 21
2.3 Current and Prior Applications of OpenCog… 23
2.3.1 Transitioning from Virtual Agents to a Physical Robot … 25
2.4 Memory Types and Associated Cognitive Processes in CogPrime… 30
2.4.1 Cognitive Synergy in PLN … 32
2.5 Goal-Oriented Dynamics in CogPrime… 35
2.6 Analysis and Synthesis Processes in CogPrime… 36
2.7 Conclusion… 40
3 Build Me Something I Haven’t Seen: A CogPrime Thought Experiment… 41
3.1 Introduction … 41
3.2 Roles of Selected Cognitive Processes… 42
3.3 A Semi-Narrative Treatment … 54
3.4 Conclusion… 57
Part II Artificial and Natural General Intelligence
4 What is Human-Like General Intelligence? … 61
4.1 Introduction … 61
4.1.1 What is General Intelligence?… 62
4.1.2 What is Human-Like General Intelligence? … 62
4.2 Commonly Recognized Aspects of Human-Like Intelligence … 63
4.3 Further Characterizations of Human-Like Intelligence … 66
4.3.1 Competencies Characterizing Human-Like Intelligence … 66
4.3.2 Gardner’s Theory of Multiple Intelligences … 67
4.3.3 Newell’s Criteria for a Human Cognitive Architecture … 68
4.3.4 Intelligence and Creativity … 69
4.4 Preschool as a View into Human-Like General Intelligence … 70
4.4.1 Design for an AGI Preschool … 72
4.5 Integrative and Synergetic Approaches to Artificial General Intelligence … 74
4.5.1 Achieving Human-Like Intelligence via Cognitive Synergy … 75
5 A Patternist Philosophy of Mind … 77
5.1 Introduction … 77
5.2 Some Patternist Principles … 78
5.3 Cognitive Synergy … 83
5.4 The General Structure of Cognitive Dynamics: Analysis and Synthesis … 85
5.4.1 Component-Systems and Self-Generating Systems … 85
5.4.2 Analysis and Synthesis … 87
5.4.3 The Dynamic of Iterative Analysis and Synthesis … 90
5.4.4 Self and Focused Attention as Approximate Attractors of the Dynamic of Iterated Forward-Analysis … 91
5.4.5 Conclusion … 94
5.5 Perspectives on Machine Consciousness … 95
5.6 Postscript: Formalizing Pattern … 96
6 Brief Survey of Cognitive Architectures … 101
6.1 Introduction … 101
6.2 Symbolic Cognitive Architectures… 103
6.2.1 SOAR … 104
6.2.2 ACT-R … 105
6.2.3 Cyc and Texai … 107
6.2.4 NARS … 107
6.2.5 GLAIR and SNePS … 108
6.3 Emergentist Cognitive Architectures … 109
6.3.1 DeSTIN: A Deep Reinforcement Learning Approach to AGI … 111
6.3.2 Developmental Robotics Architectures… 117
6.4 Hybrid Cognitive Architectures … 119
6.4.1 Neural Versus Symbolic; Global Versus Local … 121
6.5 Globalist Versus Localist Representations … 124
6.5.1 CLARION … 125
6.5.2 The Society of Mind and the Emotion Machine … 126
6.5.3 DUAL … 126
6.5.4 4D/RCS … 128
6.5.5 PolyScheme … 130
6.5.6 Joshua Blue … 130
6.5.7 LIDA… 131
6.5.8 The Global Workspace … 131
6.5.9 The LIDA Cognitive Cycle … 132
6.5.10 Psi and MicroPsi… 136
6.5.11 The Emergence of Emotion in the Psi Model … 139
6.5.12 Knowledge Representation, Action Selection and Planning in Psi … 140
6.5.13 Psi Versus CogPrime… 142
7 A Generic Architecture of Human-Like Cognition… 143
7.1 Introduction … 143
7.2 Key Ingredients of the Integrative Human-Like Cognitive Architecture Diagram … 144
7.3 An Architecture Diagram for Human-Like General Intelligence … 146
7.4 Interpretation and Application of the Integrative Diagram … 152
Part III Toward a General Theory of General Intelligence
8 A Formal Model of Intelligent Agents … 157
8.1 Introduction … 157
8.2 A Simple Formal Agents Model (SRAM) … 158
8.2.1 Goals… 159
8.2.2 Memory Stores … 160
8.2.3 The Cognitive Schematic … 162
8.3 Toward a Formal Characterization of Real-World General Intelligence … 163
8.3.1 Biased Universal Intelligence … 165
8.3.2 Connecting Legg and Hutter’s Model of Intelligent Agents to the Real World … 166
8.3.3 Pragmatic General Intelligence … 167
8.3.4 Incorporating Computational Cost … 168
8.3.5 Assessing the Intelligence of Real-World Agents… 169
8.4 Intellectual Breadth: Quantifying the Generality of an Agent’s Intelligence … 171
8.5 Conclusion… 172
9 Cognitive Synergy … 173
9.1 Introduction … 173
9.2 Cognitive Synergy … 174
9.3 Cognitive Synergy in CogPrime … 177
9.3.1 Cognitive Processes in CogPrime … 177
9.4 Some Critical Synergies … 182
9.5 Cognitive Synergy for Procedural and Declarative Learning … 182
9.5.1 Cognitive Synergy in MOSES … 187
9.5.2 Cognitive Synergy in PLN … 190
9.6 Is Cognitive Synergy Tricky?… 191
9.6.1 The Puzzle: Why Is It So Hard to Measure Partial Progress Toward Human-Level AGI? … 192
9.6.2 A Possible Answer: Cognitive Synergy Is Tricky! . . 193
9.6.3 Conclusion … 194
10 General Intelligence in the Everyday Human World … 197
10.1 Introduction … 197
10.2 Some Broad Properties of the Everyday World that Help Structure Intelligence … 198
10.3 Embodied Communication … 199
10.3.1 Generalizing the Embodied Communication Prior… 202
10.4 Naive Physics… 203
10.4.1 Objects, Natural Units and Natural Kinds … 204
10.4.2 Events, Processes and Causality … 204
10.4.3 Stuffs, States of Matter, Qualities … 205
10.4.4 Surfaces, Limits, Boundaries, Media … 205
10.4.5 What Kind of Physics is Needed to Foster Human-Like Intelligence? … 206
10.5 Folk Psychology … 208
10.5.1 Motivation, Requiredness, Value… 208
10.6 Body and Mind … 208
10.6.1 The Human Sensorium … 209
10.6.2 The Human Body’s Multiple Intelligences … 210
10.7 The Extended Mind and Body … 213
10.8 Conclusion… 213
11 A Mind-World Correspondence Principle … 215
11.1 Introduction … 215
11.2 What Might a General Theory of General Intelligence Look Like? … 216
11.3 Steps Toward A (Formal) General Theory of General Intelligence … 217
11.4 The Mind-World Correspondence Principle … 219
11.5 How Might the Mind-World Correspondence Principle Be Useful? … 220
11.6 Conclusion… 221
Part IV Cognitive and Ethical Development
12 Stages of Cognitive Development… 225
12.1 Introduction … 225
12.2 Piagetan Stages in the Context of a General Systems Theory of Development… 226
12.3 Piaget’s Theory of Cognitive Development … 227
12.3.1 Perry’s Stages… 231
12.3.2 Keeping Continuity in Mind… 232
12.4 Piaget’s Stages in the Context of Uncertain Inference … 232
12.4.1 The Infantile Stage … 234
12.4.2 The Concrete Stage … 236
12.4.3 The Formal Stage … 240
12.4.4 The Reflexive Stage … 242
13 The Engineering and Development of Ethics … 245
13.1 Introduction … 245
13.2 Review of Current Thinking on the Risks of AGI… 246
13.3 The Value of an Explicit Goal System … 249
13.4 Ethical Synergy … 251
13.4.1 Stages of Development of Declarative Ethics … 252
13.4.2 Stages of Development of Empathic Ethics … 255
13.4.3 An Integrative Approach to Ethical Development . . 257
13.4.4 Integrative Ethics and Integrative AGI… 259
13.5 Clarifying the Ethics of Justice: Extending the Golden Rule in to a Multifactorial Ethical Model … 261
13.5.1 The Golden Rule and the Stages of Ethical Development … 264
13.5.2 The Need for Context-Sensitivity and Adaptiveness in Deploying Ethical Principles… 266
13.6 The Ethical Treatment of AGIs … 269
13.6.1 Possible Consequences of Depriving AGIs of Freedom… 271
13.6.2 AGI Ethics as Boundaries Between Humans and AGIs Become Blurred … 272
13.7 Possible Benefits of Closely Linking AGIs to the Global Brain … 273
13.7.1 The Importance of Fostering Deep, Consensus-Building Interactions Between People with Divergent Views … 275
13.8 Possible Benefits of Creating Societies of AGIs … 277
13.9 AGI Ethics as Related to Various Future Scenarios… 278
13.9.1 Capped Intelligence Scenarios … 278
13.9.2 Superintelligent AI: Soft-Takeoff Scenarios … 279
13.9.3 Superintelligent AI: Hard-Takeoff Scenarios … 280
13.9.4 Global Brain Mindplex Scenarios … 281
13.10 Conclusion: Eight Ways to Bias AGI Toward Friendliness . . 284
13.10.1 Encourage Measured Co-Advancement of AGI Software and AGI Ethics Theory … 286
13.10.2 Develop Advanced AGI Sooner Not Later … 286
Part V Networks for Explicit and Implicit Knowledge Representation
14 Local, Global and Glocal Knowledge Representation … 291
14.1 Introduction … 291
14.2 Localized Knowledge Representation Using Weighted, Labeled Hypergraphs… 292
14.2.1 Weighted, Labeled Hypergraphs … 292
14.3 Atoms: Their Types and Weights … 293
14.3.1 Some Basic Atom Types … 294
14.3.2 Variable Atoms… 296
14.3.3 Logical Links … 297
14.3.4 Temporal Links … 299
14.3.5 Associative Links … 300
14.3.6 Procedure Nodes … 301
14.3.7 Links for Special External Data Types … 302
14.3.8 Truth Values and Attention Values … 302
14.4 Knowledge Representation via Attractor Neural Networks … 303
14.4.1 The Hopfield Neural Net Model … 303
14.4.2 Knowledge Representation via Cell Assemblies … 304
14.5 Neural Foundations of Learning … 305
14.5.1 Hebbian Learning … 305
14.5.2 Virtual Synapses and Hebbian Learning Between Assemblies … 306
14.5.3 Neural Darwinism … 307
14.6 Glocal Memory … 308
14.6.1 A Semi-Formal Model of Glocal Memory … 310
14.6.2 Glocal Memory in the Brain… 312
14.6.3 Glocal Hopfield Networks … 315
14.6.4 Neural-Symbolic Glocality in CogPrime … 317
15 Representing Implicit Knowledge via Hypergraphs … 319
15.1 Introduction … 319
15.2 Key Vertex and Edge Types … 320
15.3 Derived Hypergraphs… 320
15.3.1 SMEPH Vertices… 321
15.3.2 SMEPH Edges … 322
15.4 Implications of Patternist Philosophy for Derived Hypergraphs of Intelligent Systems… 323
15.4.1 SMEPH Principles in CogPrime … 324
16 Emergent Networks of Intelligence … 327
16.1 Introduction … 327
16.2 Small World Networks … 328
16.3 Dual Network Structure… 329
16.3.1 Hierarchical Networks … 329
16.3.2 Associative, Heterarchical Networks … 332
16.3.3 Dual Networks … 333
Part VI A Path to Human-Level AGI
17 AGI Preschool … 337
17.1 Introduction … 337
17.1.1 Contrast to Standard AI Evaluation Methodologies … 338
17.2 Elements of Preschool Design … 340
17.3 Elements of Preschool Curriculum … 341
17.3.1 Preschool in the Light of Intelligence Theory… 341
17.4 Task-Based Assessment in AGI Preschool … 343
17.5 Beyond Preschool … 346
17.6 Issues with Virtual Preschool Engineering … 347
17.6.1 Integrating Virtual Worlds with Robot Simulators … 349
17.6.2 BlocksNBeads World … 349
18 A Preschool-Based Roadmap to Advanced AGI… 355
18.1 Introduction … 355
18.2 Measuring Incremental Progress Toward Human-Level AGI … 356
18.3 Conclusion… 364
19 Advanced Self-Modification: A Possible Path to Superhuman AGI… 365
19.1 Introduction … 365
19.2 Cognitive Schema Learning … 367
19.3 Self-Modification via Supercompilation… 368
19.3.1 Three Aspects of Supercompilation … 370
19.3.2 Supercompilation for Goal-Directed Program Modification … 371
19.4 Self-Modification via Theorem-Proving … 372
Appendix A: Glossary … 375
Index … 405