Systems Engineering and Artificial Intelligence

This book provides a broad overview of the benefits from a Systems Engineering design philosophy in architecting complex systems composed of artificial intelligence (AI), machine learning (ML) and humans situated in chaotic environments. The major topics include emergence, verification and validation of systems using AI/ML and human systems integration to develop robust and effective human-machine teams—where the machines may have varying degrees of autonomy due to the sophistication of their embedded AI/ML. The chapters not only describe what has been learned, but also raise questions that must be answered to further advance the general Science of Autonomy. The science of how humans and machines operate as a team requires insights from, among others, disciplines such as the social sciences, national and international jurisprudence, ethics and policy, and sociology and psychology. The social sciences inform how context is constructed, how trust is affected when humans and machines depend upon each other and how human-machine teams need a shared language of explanation. National and international jurisprudence determine legal responsibilities of non-trivial human-machine failures, ethical standards shape global policy, and sociology provides a basis for understanding team norms across cultures. Insights from psychology may help us to understand the negative impact on humans if AI/ML based machines begin to outperform their human teammates and consequently diminish their value or importance. This book invites professionals and the curious alike to witness a new frontier open as the Science of Autonomy emerges. Preface Contents 1 Introduction to “Systems Engineering and Artificial Intelligence” and the Chapters 1.1 Introduction. The Disruptive Nature of AI 1.1.1 Justifying Speedy Decisions 1.1.2 Systems Engineering (SE) 1.1.3 Common Ground: AI, Interdependence, and SE 1.1.4 Social Science 1.1.5 The Science of Human Teams 1.1.6 Human–Machine Teams 1.2 Introduction to the Chapters 1.3 Summary References 2 Recognizing Artificial Intelligence: The Key to Unlocking Human AI Teams 2.1 Introduction 2.1.1 Motivation and Goals 2.1.2 Types of Human-AI Collaboration 2.1.3 Ground Rules 2.2 System Engineering 2.2.1 Design and Embodiment 2.2.2 Generative Language Models 2.2.3 System Architecture 2.2.4 Agile Development 2.3 Applications 2.3.1 Ideation Discussions 2.3.2 Collaborative Writing 2.4 Innovative Brainstorm Workshop 2.4.1 Protocol 2.4.2 Analysis 2.4.3 Preliminary Results 2.5 Related Work 2.6 Future Applications 2.7 Conclusion References 3 Artificial Intelligence and Future of Systems Engineering 3.1 Introduction 3.2 SERC AI4SE and SE4AI Roadmap 3.3 Digital Engineering 3.4 AI/ML Technology Evolution 3.5 Augmented Engineering 3.6 Workforce and Culture 3.7 Summary—The AI imperative for Systems Engineering References 4 Effective Human–Artificial Intelligence Teaming 4.1 Introduction 4.2 Synthetic Teammates 4.3 HAT Findings and Their Implications for Human Teams 4.4 Conclusions and Future Work References 5 Toward System Theoretical Foundations for Human–Autonomy Teams 5.1 Introduction 5.2 Organizational Structure and Role/Function Allocation 5.3 Working Together on Tasks 5.4 Teaming Over Longer Durations 5.5 Formally Modeling and Composing Complex Human–Machine Systems 5.6 Conclusions and Future Directions References 6 Systems Engineering for Artificial Intelligence-based Systems: A Review in Time 6.1 Perspectives on AI and Systems Engineering 6.2 The Dynamics of This Space 6.2.1 Evolving an SE Framework: Ontologies of AI/ML—Dealing with the Breadth of the Fields 6.2.2 Systems Engineering as a Moving Target 6.2.3 The First to Market Motivation 6.2.4 Technical Debt 6.2.5 Summary 6.3 Stepping Through Some Systems Engineering Issues 6.3.1 Capability Maturity Model Integration [CMMI] and SE for R&D 6.3.2 Requirements Engineering 6.3.3 Software Engineering for AI/ML Systems 6.3.4 Test and Evaluation 6.4 Sampling of Technical Issues and Challenges 6.4.1 Emergence and Emergent Behavior 6.4.2 Safety in AI/ML 6.4.3 The Issue of Explanation/Explainability 6.5 Summary References 7 Human-Autonomy Teaming for the Tactical Edge: The Importance of Humans in Artificial Intelligence Research and Development 7.1 Introduction 7.2 The Fundamental Nature of Human-Autonomy Teaming 7.2.1 Complementarity of Human and AI Characteristics 7.2.2 Tracking the Important Roles of the Human Across AI History 7.3 Artificial Intelligence for Human-Autonomy Teams 7.3.1 Quantifying Soldier Understanding for AI 7.3.2 Soldier-Guided AI Adaptations 7.3.3 Characterizing Soldier-Autonomy Performance 7.4 Conclusions References 8 Re-orienting Toward the Science of the Artificial: Engineering AI Systems 8.1 Introduction 8.2 AI Software Engineering 8.3 AI-enabled Complex Systems-of-Systems and Emergent Behaviors 8.4 The Importance of Interoperability 8.5 The Role of Uncertainty in ML 8.6 The Challenge of Data and ML: An NLP Example 8.6.1 System Architecture 8.6.2 Results 8.6.3 Discussion 8.7 Design Science: Toward the Science of AI System Engineering 8.8 Conclusion References 9 The Department of Navy’s Digital Transformation with the Digital System Architecture, Strangler Patterns, Machine Learning, and Autonomous Human–Machine Teaming 9.1 Introduction 9.2 Autonomous Human–Machine Teaming Lifecycle Difficulties 9.3 Unique Challenges Facing the Department of Navy and Autonomous Human–Machine Teaming 9.3.1 Department of Navy Non-technical Challenges 9.3.2 Department of Navy Technical Challenges 9.4 Attacking the Technical Debt and Inflation to Enable AHMT Solutions 9.4.1 AHMT Solutions and New Target Platforms 9.4.2 AHMT Solutions and Legacy Target Platforms 9.5 Conclusion and Path Forward References 10 Digital Twin Industrial Immune System: AI-driven Cybersecurity for Critical Infrastructures 10.1 Introduction 10.1.1 Overview 10.1.2 Cybersecurity Technology Gaps for Advanced Detection, Protection and Monitoring Solutions 10.1.3 Digital Ghost: A Next-Generation Response to Close Critical Energy Infrastructure Gaps 10.2 People, Process and Technology Applicability Gap Analysis 10.2.1 Attack Detection 10.2.2 Attack Localization 10.2.3 Attack Neutralization 10.2.4 Man Versus Machine Anomaly Forecasting and Detection 10.3 Digital Ghost Research Findings and Future Research 10.3.1 Invariant Learning 10.3.2 Autonomous Defense: Critical Sensors Identification and Trust 10.3.3 Humble AI 10.3.4 Explainable AI (XAI) 10.4 Conclusion References 11 A Fractional Brownian Motion Approach to Psychological and Team Diffusion Problems 11.1 Introduction 11.2 Random Walk 11.2.1 Wiener Process from the Fair Simple Random Walk 11.2.2 Wiener Process (standard Brownian Motion) Defined 11.2.3 Simulation of the Wiener Process via G0,1n 11.2.4 Continuity of Sample Paths 11.2.5 Non-differentiability of Wiener Process Sample Paths 11.3 Brownian Motion 11.3.1 Simulation of Brownian Motion 11.4 Stopping Times and Absorbing Boundaries 11.4.1 Two Absorbing Boundaries—The Situation for Ratcliff Drift Diffusion 11.5 Fractional Brownian Motion 11.5.1 Covariance of Brownian Motion 11.5.2 Definition of the Fractional Wiener Process 11.5.3 Existence and Properties of the Fractional Wiener Process 11.5.4 Ratcliff Diffusion Revisited 11.6 Determining H, a Problem in AI 11.6.1 Our Hybrid Approach 11.7 Team Science and Future Work References 12 Human–Machine Understanding: The Utility of Causal Models and Counterfactuals 12.1 Introduction 12.2 Information-Theoretic Framework for SCM Construction 12.3 Assessing and Correcting for Bias in Information-Theoretic SCM Construction 12.4 Construction of SCM for Counterfactuals 12.5 Notes on Related Work 12.6 Summary References 13 An Executive for Autonomous Systems, Inspired by Fear Memory Extinction 13.1 The Problem 13.2 Moondoodya, a Novel Electronic Warfare System 13.3 PTSD Fear Extinction 13.4 A Mathematical Approach to Executive Abstraction 13.5 ‘Effect First’ Modelling 13.6 A Closure Embedding Strategy 13.7 The Tookoonooka Vortex Collaborative 13.8 Conclusions References 14 Contextual Evaluation of Human–Machine Team Effectiveness 14.1 Introduction 14.2 Related Works 14.3 Background 14.3.1 Interference 14.3.2 Inverse Reinforcement Learning (IRL) 14.3.3 Preferential Trajectory-Based IRL (PT-IRL) 14.4 Approach 14.4.1 Experimental Setup 14.4.2 Training Classifier 14.4.3 Human and Human–Machine Teams 14.4.4 Evaluation of Human–Machine Team Effectiveness 14.5 Conclusion and Future Work References 15 Humanity in the Era of Autonomous Human–machine Teams 15.1 Introduction: AHMTs in the Form of the Trio 15.1.1 The Trio: Data, the Internet, and Algorithms 15.1.2 AHMTs Manifested by the Trio 15.1.3 Scitovsky’s Caveat 15.2 Human–Machine Teams 15.2.1 Shelley Model: Frankenstein and His Creature 15.2.2 Lovelock Model: GAIA and Novacene 15.2.3 Margulis Model: Symbiogenesis and Super Cooperators 15.2.4 Polanyi Model: Tension Between Habitation and Improvement 15.2.5 Laloux Model: Soulful Organizations 15.3 Meaning of the Trios for Humanity 15.3.1 Co-evolutions of Humans and Machines 15.3.2 Individuality 15.3.3 Democratization of Individuality 15.4 Meaning of the Trio for the Humanities 15.4.1 Distant Reading 15.4.2 Extended Reading 15.4.3 Participatory Reading 15.5 Concluding Remarks References 16 Transforming the System of Military Medical Research: An Institutional History of the Department of Defense’s (DoD) First Electronic Institutional Review Board Enterprise IT System 16.1 Introduction. A Tale of Two Histories 16.1.1 Goal 1: The eIRB Transformed the MEDCENs 16.1.2 Goal 2: The Initial Meeting on Collaboration 16.1.3 Our Two Goals Merged into One 16.2 The Next Steps in the Transformation from a Paper to Electronic System 16.3 Boundary Maintenance 16.4 Future Steps to Determine Impacts. Preliminary Results in 2010 16.5 Summary 16.6 Postscript References 17 Collaborative Communication and Intelligent Interruption Systems 17.1 Introduction 17.2 Interruptions in Multi-user Multitasking Interactions 17.2.1 Low Cognitive Interruption Timings 17.2.2 High Cognitive Interruption Timings 17.3 Methods 17.3.1 Data Collection 17.3.2 Conditions 17.4 Results and Discussion 17.4.1 Team Performance Analyses 17.4.2 Individual Subjective Analyses 17.4.3 Individual Interruption Task Measures 17.5 Discussion 17.6 Conclusion References 18 Shifting Paradigms in Verification and Validation of AI-Enabled Systems: A Systems-Theoretic Perspective 18.1 Introduction 18.2 A Need for a Paradigm Shift in V&V 18.3 A Systems-Theoretic Interpretation of Intelligence 18.4 Challenges to the V&V of AI-Enabled Systems 18.4.1 Differential Learning in V&V Versus Operational Environment 18.4.2 Endogenous Evolution of Systems 18.4.3 Verification of Learning to Learn 18.4.4 Encapsulation of Intelligent Properties 18.5 Conclusion References 19 Toward Safe Decision-Making via Uncertainty Quantification in Machine Learning 19.1 Introduction 19.2 Decision-Making and Machine Learning 19.2.1 Summary of a ML-augmented Decision-Making Process 19.2.2 Uncertainty Quantification as Part of the Decision-Making Process 19.3 Bayesian Inference 19.3.1 Bayesian Neural Networks 19.3.2 The Predictive Distribution 19.4 Making Decisions in the Presence of Uncertainty: Bayesian Decision Theory 19.5 A Case Study: Vehicle Classification from Acoustic Sensors 19.5.1 The Data Set 19.5.2 The Neural Network Architecture 19.5.3 Inference Approach 19.5.4 The Decision-Making Task: Avoiding Catastrophic Failure 19.5.5 Overall Results 19.5.6 Calibration of the Cost Function 19.6 Resource Requirements of Bayesian Inference 19.7 Conclusion References 20 Engineering Context from the Ground Up 20.1 Introduction 20.2 Information Processing Architecture 20.3 Memory Import and Storage 20.4 Natural Language Processing 20.5 Confidence Aggregator 20.6 Perspective Transformation 20.7 Planning Unit 20.8 Communication Unit 20.9 Command Unit 20.10 Conclusions: Engineering Context References 21 Meta-reasoning in Assembly Robots 21.1 Introduction and Background 21.1.1 Related Work 21.2 Illustrative Examples 21.3 Assembly as a Reasoning Problem 21.4 Failure Mode, Effect, and Repair Analysis 21.4.1 Skill-level Failures 21.4.2 Task-level Failures 21.4.3 Mission-level Failures 21.4.4 A Preliminary Taxonomy of Failure Modes 21.5 Assembly Plan Repair as a Meta-reasoning Problem 21.5.1 Meta-reasoning Architecture for Robots 21.5.2 Skill Failure Detection and Repair 21.5.3 Task Failures and Repairs 21.5.4 Toward Mission Repair 21.6 Discussion and Conclusions References 22 From Informal Sketches to Systems Engineering Models Using AI Plan Recognition 22.1 Motivation 22.2 Related Work 22.2.1 Natural Sketching 22.2.2 Artificial Intelligence 22.3 Plan Recognition Approach 22.3.1 Approach Overview 22.3.2 Tolerance to Drawing Imperfections 22.4 Implementation 22.4.1 Automated Deterministic Planning 22.4.2 Modeling Sketches 22.5 Experiment 22.6 Conclusion References 23 An Analogy of Sentence Mood and Use 23.1 Helpful Misalignment 23.2 Theorizing the Relation 23.3 Setters and Indicators 23.4 Schemes and Force 23.4.1 Practical Reasoning 23.4.2 Analogy References 24 Effective Decision Rules for Systems of Public Engagement in Radioactive Waste Disposal: Evidence from the United States, the United Kingdom, and Japan 24.1 Introduction 24.2 Literature Review: Decision Rules that Encourage Public Engagement 24.3 Empirical Analysis of Public Engagement and Decision Rules 24.3.1 United States 24.3.2 United Kingdom: The “Participatory Turn” and Its Consequences 24.3.3 Japan: Public Interest in Participatory Approach to GDF Siting 24.4 Conclusion References 25 Outside the Lines: Visualizing Influence Across Heterogeneous Contexts in PTSD 25.1 Introduction 25.2 Defining and Representing Context 25.2.1 Defining Narrative 25.2.2 Surrounding Literature 25.3 PTSD Example 25.4 Representations 25.4.1 Visual Grammar: Taxonomy 25.4.2 Visual Grammar: Dynamic Operations 25.4.3 Technical Foundations 25.5 Examples 25.5.1 Four Multi-disciplinary Models 25.5.2 Integration of Multiple Models: Model of Models 25.5.3 Zooming 25.5.4 Sandwich Layers 25.5.5 Signature Structures 25.5.6 Media 25.6 Discussion 25.6.1 Readability 25.6.2 Dynamism 25.6.3 Future Applications 25.7 Conclusion References This book provides a broad overview of the benefits from a Systems Engineering design philosophy in architecting complex systems composed of artificial intelligence (AI), machine learning (ML) and humans situated in chaotic environments. The major topics include emergence, verification and validation of systems using AI/ML and human systems integration to develop robust and effective human-machine teams' here the machines may have varying degrees of autonomy due to the sophistication of their embedded AI/ML. The chapters not only describe what has been learned, but also raise questions that must be answered to further advance the general Science of Autonomy. The science of how humans and machines operate as a team requires insights from, among others, disciplines such as the social sciences, national and international jurisprudence, ethics and policy, and sociology and psychology. The social sciences inform how context is constructed, how trust is affected when humans and machines depend upon each other and how human-machine teams need a shared language of explanation. National and international jurisprudence determine legal responsibilities of non-trivial human-machine failures, ethical standards shape global policy, and sociology provides a basis for understanding team norms across cultures. Insights from psychology may help us to understand the negative impact on humans if AI/ML based machines begin to outperform their human teammates and consequently diminish their value or importance. This book invites professionals and the curious alike to witness a new frontier open as the Science of Autonomy emerges