Artificial General Intelligence ebook download

Artificial General Intelligence

Bengoertzel - Cassio Pennachin

“Only a small community has concentrated on general intelligence. No one has tried to make a thinking machine . . . The bottom line is that we really haven’t progressed too far toward a truly intelligent machine. We have collections of dumb specialists in small domains; the true majesty of general intelligence still awaits our attack. . . .

We have got to get back to the deepest questions of AI and general intelligence...” –MarvinMinsky as interviewed in Hal’s Legacy, edited by David Stork, 2000.

Our goal in creating this edited volume has been to fill an apparent gap in the scientific literature, by providing a coherent presentation of a body of contemporary research that, in spite of its integral importance, has hitherto kept a very low profile within the scientific and intellectual community. This body of work has not been given a name before; in this book we christen it “Artificial General Intelligence” (AGI). What distinguishes AGI work from run-of-the-mill “artificial intelligence” research is that it is explicitly focused on engineering general intelligence in the short term. We have been active researchers in the AGI field for many years, and it has been a pleasure to gather together papers from our colleagues working on related ideas from their own perspectives. In the Introduction we give a conceptual overview of the AGI field, and also summarize and interrelate the key ideas of the papers in the subsequent chapters.

Of course, “general intelligence” does not mean exactly the same thing to all researchers. In fact it is not a fully well-defined term, and one of the issues raised in the papers contained here is how to define general intelligence in a way that provides maximally useful guidance to practical AI work. But,nevertheless, there is a clear qualitative meaning to the term. What is meant by AGI is, loosely speaking, AI systems that possess a reasonable degree of self-understanding and autonomous self-control, and have the ability to solve a variety of complex problems in a variety of contexts, and to learn to solve new problems that they didnt know about at the time of their creation. A marked distinction exists between practical AGI work and, on the other hand:

• Pragmatic but specialized “narrow AI” research which is aimed at cre- ating programs carrying out specific tasks like playing chess, diagnosing diseases, driving cars and so forth (most contemporary AI work falls into this category.) • Purely theoretical AI research, which is aimed at clarifying issues regarding the nature of intelligence and cognition, but doesnt involve technical details regarding actually realizing artificially intelligent software.

Some of the papers presented here come close to the latter (purely theo- retical) category, but we have selected them because the theoretical notions they contain seem likely to lead to such technical details in the medium-term future, and/or resonate very closely with the technical details of AGI designs proposed by other authors.

The audience we intend to reach includes the AI community, and also the broader community of scientists and students in related fields such as philoso- phy, neuroscience, linguistics, psychology, biology, sociology, anthropology and engineering. Significantly more so than narrow AI, AGI is interdisciplinary in nature, and a full appreciation of the general intelligence problem and its various potential solutions requires one to take a wide variety of different perspectives.

Not all significant AGI researchers are represented in these pages, but we have sought to bring together a multiplicity of perspectives, including many that disagree with our own. Bringing a diverse body of AGI research together in a single volume reveals the common themes among various researchers work, and makes clear what the big open questions are in this vital and critical area of research. It is our hope that this book will interest more researchers and students in pursuing AGI research themselves, thus aiding in the progress of science.

In the three years that this book has been in the making, we have noticed a significant increase in interest in AGI-related research within the academic AI community, including a number of small conference workshops with titles related to “Human-Level Intelligence.” We consider this challenge to the over- whelming dominance of narrow-AI an extremely positive move; however, we submit that “Artificial General Intelligence” is a more sensible way to concep- tualize the problem than “Human-Level Intelligence.” The AGI systems and approaches described in these pages are not necessarily oriented towards emu- lating the human brain; and given the heterogeneity of the human mind/brain and its highly various levels of competence at various sorts of tasks, it seems very difficult to define “Human-Level Intelligence” in any way that is generally applicable to AI systems that are fundamentally non-human-like in concep- tion. On the other hand, the work of Hutter and Schmidhuber reported here provides a reasonable, abstract mathematical characterization of general intel- ligence which, while not in itself providing a practical approach to AGI design and engineering, at least provides a conceptually meaningful formalization of the ultimate goal of AGI work.

The grand goal of AGI remains mostly unrealized, and how long it will be until this situation is remedied remains uncertain. Among scientists who believe in the fundamental possibility of strong AI, the most optimistic se- rious estimates we have heard are in the range of 5-10 years, and the most pessimistic are in the range of centuries. While none of the articles contained here purports to present a complete solution to the AGI problem, we believe that they collectively embody meaningful conceptual progress, and indicate clearly that the direct pursuit of AGI is an endeavor worthy of significant research attention.

Content

Contemporary Approaches to Artificial General Intelligence

Cassio Pennachin, Ben Goertzel

1 ABriefHistoryofAGI 1

1.1 SomeHistoricalAGI-RelatedProjects 2

2 What Is Intelligence? . 6

2.1 The Psychology of Intelligence . 6

2.2 TheTuringTest . 8

2.3 A Control Theory Approach to Defining Intelligence . 8

2.4 Efficient Intelligence

3 The Abstract Theory of General Intelligence .

4 TowardaPragmaticLogic.

5 EmulatingtheHumanBrain.

6 EmulatingtheHumanMind .

7 Creating Intelligence by Creating Life .

8 The Social Nature of Intelligence .

9 IntegrativeApproaches

 TheOutlookforAGI .

Acknowledgments

References

The Logic of Intelligence

Pei Wang

1 Intelligence and Logic .

1.1 To Define Intelligence

1.2 A Working Definition of Intelligence

1.3 ComparisonWithOtherDefinitions

1.4 LogicandReasoningSystems .

2 TheComponentsofNARS

2.1 Experience-Grounded Semantics .

2.2 InheritanceStatement

2.3 CategoricalLanguage

2.4 SyllogisticInferenceRules

2.5 ControlledConcurrencyinDynamicMemory

3 ThePropertiesofNARS

3.1 ReasonableSolutions.

3.2 UnifiedUncertaintyProcessing

3.3 NARSasaParallelandDistributedNetwork

3.4 ResourcesCompetition .

3.5 FlexibleBehaviors .

3.6 AutonomyandCreativity.

4 Conclusions

References

The Novamente Artificial Intelligence Engine

Ben Goertzel, Cassio Pennachin

1 Introduction.

1.1 TheNovamenteAGISystem

1.2 Novamente for Knowledge Management and Data Analysis

2 EnablingSoftwareTechnologies

2.1 A Distributed Software Architecture for Integrative AI .

2.2 Database Integration and Knowledge Integration

3 What Is Artificial General Intelligence? .

3.1 What Is General Intelligence? .

3.2 TheIntegrativeApproachtoAGI

3.3 Experiential Interactive Learning and Adaptive Self-modification

4 ThePsynetModelofMind

5 TheNovamenteAGIDesign .

5.1 AnIntegrativeKnowledgeRepresentation.

5.2 TheMindOS .

5.3 AtomTypes

5.4 NovamenteMaps

5.5 MindAgents

5.6 MapDynamics

5.7 FunctionalSpecialization .

5.8 NovamenteandtheHumanBrain

5.9 EmergentStructures .

6 InteractingwithHumansandDataStores .

6.1 DataSources

6.2 KnowledgeEncoding.

6.3 Querying .

6.4 FormalLanguageQueries.

6.5 ConversationalInteraction

6.6 ReportGeneration.

6.7 Active Collaborative Filtering and User Modeling

7 ExampleNovamenteAIProcesses

7.1 Probabilistic Inference

7.2 Nonlinear-DynamicalAttentionAllocation

7.3 ImportanceUpdating

7.4 SchemaandPredicateLearning

7.5 PatternMining

7.6 NaturalLanguageProcessing

8 Conclusion

Appendix: Novamente Applied to Bioinformatic Pattern Mining .

References

Essentials of General Intelligence:

The Direct Path to Artificial General Intelligence

Peter Voss

1 Introduction.

2 General Intelligence .

2.1 Core Requirements for General Intelligence

2.2 Advantages of Intelligence Being General .

3 Shortcuts toAGI.

4 Foundational Cognitive Capabilities

5 AnAGI intheMaking

5.1 AGIEngineArchitectureandDesignFeatures.

6 From Algorithms to General Intelligence

6.1 Sample Test Domains for Initial Performance Criteria

6.2 Towards Increased Intelligence .

7 OtherResearch

8 Fast-trackAGI:WhySoRare?.

9 Conclusion

References

Artificial Brains

Hugo de Garis

1 Introduction.

2 EvolvableHardware

2.1 NeuralNetworkModels.

3 TheCAM-BrainMachine(CBM).

3.1 EvolvedModules

3.2 TheKittenRobot“Robokitty”

4 Short-andLong-TermFuture

5 Postscript – July 2

References

The New AI: General & Sound & Relevant for Physics

J¨ urgen Schmidhuber

1 Introduction.

2 MoreFormally.

3 Prediction Using a Universal Algorithmic Prior Based on the

ShortestWayofDescribingObjects.

4 Super Omegas and Generalizations of Kolmogorov Complexity &

Algorithmic Probability .

5 Computable Predictions Through the Speed Prior Based on the

FastestWayofDescribingObjects

6 SpeedPrior-BasedPredictions forOurUniverse

7 OptimalRationalDecisionMakers

8 OptimalUniversalSearchAlgorithms .

9 OptimalOrderedProblemSolver(OOPS) .

 OOPS-BasedReinforcementLearning.

 The G¨ odelMachine.

 Conclusion

 Acknowledgments

References

G¨ odel Machines: Fully Self-Referential Optimal Universal

Self-improvers

J¨ urgen Schmidhuber

1 IntroductionandOutline

2 Basic Overview, Relation to Previous Work, and Limitations .

2.1 NotationandSet-up .

2.2 Basic Idea of G¨ odelMachine

2.3 Proof Techniques and an O()-optimal Initial Proof Searcher .

2.4 RelationtoHutter’sPreviousWork

2.5 Limitations of G¨ odelMachines

3 Essential Details of One Representative G¨ odelMachine .

3.1 ProofTechniques

4 GlobalOptimalityTheorem .

4.1 AlternativeRelaxedTargetTheorem.

5 Bias-OptimalProofSearch(BIOPS)

5.1 How a Surviving Proof Searcher May Use Biops to Solve

RemainingProofSearchTasks

6 Discussion&AdditionalRelations toPreviousWork

6.1 Possible Types of G¨ odel Machine Self-improvements

6.2 ExampleApplications

6.3 Probabilistic G¨ odelMachineHardware .

6.4 More Relations to Previous Work on Less General

Self-improvingMachines

6.5 Are Humans Probabilistic G¨ odelMachines? .

6.6 G¨ odelMachinesandConsciousness.

6.7 FrequentlyAskedQuestions.

7 Conclusion

8 Acknowledgments

References

Universal Algorithmic Intelligence: A Mathematical

Top→Down Approach

Marcus Hutter

1 Introduction.

2 Agents in Known Probabilistic Environments

2.1 TheCyberneticAgentModel

2.2 Strings .

2.3 AIModel forKnownDeterministicEnvironment.

2.4 AI Model for Known Prior Probability .

2.5 Probability Distributions .

2.6 Explicit Form of the AIµ Model .

2.7 FactorizableEnvironments

2.8 ConstantsandLimits

2.9 SequentialDecisionTheory .

3 UniversalSequencePrediction .

3.1 Introduction

3.2 AlgorithmicInformationTheory.

3.3 Uncertainty & Probabilities .

3.4 Algorithmic Probability & Universal Induction

3.5 LossBounds&ParetoOptimality .

4 TheUniversalAlgorithmicAgentAIXI

4.1 The Universal AIξ Model .

4.2 OntheOptimalityofAIXI

4.3 Value Bounds and Separability Concepts

4.4 Pareto Optimality of AIξ .

4.5 TheChoiceof theHorizon

4.6 Outlook

4.7 Conclusions .

5 ImportantProblemClasses

5.1 SequencePrediction(SP) .

5.2 StrategicGames (SG)

5.3 FunctionMinimization(FM)

5.4 SupervisedLearningfromExamples (EX).

5.5 Other Aspects of Intelligence

6 Time-BoundedAIXIModel

6.1 Time-Limited Probability Distributions .

6.2 TheIdeaof theBestVoteAlgorithm.

6.3 ExtendedChronologicalPrograms .

6.4 ValidApproximations

6.5 Effective Intelligence Order Relation .

6.6 The Universal Time-Bounded AIXItl Agent .

6.7 LimitationsandOpenQuestions.

6.8 Remarks

7 Discussion.

7.1 GeneralRemarks

7.2 Outlook&OpenQuestions .

7.3 TheBigQuestions .

7.4 Conclusions .

AnnotatedBibliography

References

Program Search as a Path to Artificial General Intelligence

 Lukasz Kaiser

1 Intelligence and the Search for Programs

2 TheoreticalResults .

2.1 ProgramSearchintheStandardAIModel

2.2 Self-improvingProgramSearch

2.3 DiscussionofEfficiencyDefinitions.

3 ConvenientModelofComputation

3.1 ExtendedProgramNotation

3.2 Compiling Typed Rewriting Systems .

4 ReasoningUsingGames.

4.1 ReasonandSearchGameforTerms

5 Conclusions

References

The Natural Way to Artificial Intelligence

Vladimir G. Red’ko

1 Introduction.

2 TheEpistemologicalProblem

3 Approaches to the Theory of Evolutionary Origin of Human Intelligence

3.1 “Intelligent Inventions” of Biological Evolution

3.2 MethodologicalApproaches .

3.3 Role of Investigations of “Artificial Life” and “Simulation of

AdaptiveBehavior”

4 TwoModels .

4.1 Alife Model of Evolutionary Emergence of Purposeful

AdaptiveBehavior.

4.2 ModelofEvolutionofWebAgents.

5 Towards the Implementation of Higher Cognitive Abilities

6 Conclusion

7 Acknowledgements .

References

3D Simulation: the Key to A.I.

Keith A. Hoyes

1 Introduction.

2 Pillars of Intelligence .

2.1 DeepBlue

2.2 VirtualReality

2.3 TheHumbleEarthworm

3 Consciousness

3.1 FeelingandQualia.

4 General Intelligence .

4.1 Human Intelligence

5 3DSimulationandLanguage

6 Epistemology

7 Instantiation: theHeartofConsciousness

8 InaNutshell

9 Real-WorldAI .

9.1 ExamplesandMetaphors .

9.2 MathandSoftware.

9.3 BarcodeExample

9.4 SoftwareDesign .

 Conclusion

References

Levels of Organization in General Intelligence

Eliezer Yudkowsky

1 Foundations of General Intelligence .

2 Levels of Organization in Deliberative General Intelligence .

2.1 Concepts:AnIllustrationofPrinciples

2.2 LevelsofOrganizationinDeliberation

2.3 TheCodeLevel .

2.4 TheModalityLevel

2.5 TheConceptLevel.

2.6 TheThoughtLevel

2.7 TheDeliberationLevel .

3 SeedAI .

3.1 AdvantagesofMinds-in-General .

3.2 RecursiveSelf-enhancement .

3.3 Infrahumanity and Transhumanity: “Human-Equivalence” as

Anthropocentrism .

4 Conclusions

References

Index

------------------------------

1 A Brief History of AGI

The vast bulk of the AI field today is concerned with what might be called “narrow AI” – creating programs that demonstrate intelligence in one or an- other specialized area, such as chess-playing, medical diagnosis, automobile- driving, algebraic calculation or mathematical theorem-proving. Some of these narrow AI programs are extremely successful at what they do. The AI projects discussed in this book, however, are quite different: they are explicitly aimed at artificial general intelligence, at the construction of a software program that can solve a variety of complex problems in a variety of different domains, and that controls itself autonomously, with its own thoughts, worries, feelings, strengths, weaknesses and predispositions.

Artificial General Intelligence (AGI) was the original focus of the AI field, but due to the demonstrated difficulty of the problem, not many AI researchers are directly concerned with it anymore. Work on AGI has gotten a bit of a bad reputation, as if creating digital general intelligence were analogous to building a perpetual motion machine. Yet, while the latter is strongly implied to be impossible by well-established physical laws, AGI appears by all known science to be quite possible. Like nanotechnology, it is “merely an engineering problem”, though certainly a very difficult one.

The presupposition of much of the contemporary work on “narrow AI” is that solving narrowly defined subproblems, in isolation, contributes signifi- cantly toward solving the overall problem of creating real AI. While this is of course true to a certain extent, both cognitive theory and practical experience suggest that it is not so true as is commonly believed. In many cases, the best approach to implementing an aspect of mind in isolation is very different from the best way to implement this same aspect of mind in the framework of an integrated AGI-oriented software system.

The chapters of this book present a series of approaches to AGI. None of these approaches has been terribly successful yet, in AGI terms, although several of them have demonstrated practical value in various specialized do- mains (narrow-AI style). Most of the projects described are at an early stage of engineering development, and some are still in the design phase. Our aim is not to present AGI as a mature field of computer science – that would be

-------------------Click Icon file to download this ebook------------------------