Interdependent Human-Machine Teams

The Path to Autonomy

1st Edition - December 5, 2024
Editors: William Lawless, Ranjeev Mittu, Donald Sofge, Hesham Fouad
Language: English
Paperback ISBN:
9 7 8 - 0 - 4 4 3 - 2 9 2 4 6 - 0
eBook ISBN:
9 7 8 - 0 - 4 4 3 - 2 9 2 4 7 - 7

Interdependent Human-Machine Teams: The Path to Autonomy examines the foundations, metrics, and applications of human-machine systems, the legal ramifications of autonomy,… Read more

Purchase options

World Book Day Celebration!

Save up to 30% on books and eBooks!

Shop now

Institutional subscription on ScienceDirect

Request a sales quote

Interdependent Human-Machine Teams: The Path to Autonomy examines the foundations, metrics, and applications of human-machine systems, the legal ramifications of autonomy, trust by the public, and trust by the users and AI systems of their users, integrating concepts from various disciplines such as AI, machine learning, social sciences, quantum mechanics, and systems engineering. In this book, world-class researchers, engineers, ethicists, and social scientists discuss what machines, humans, and systems should discuss with each other, to policymakers, and to the public.

It establishes the meaning and operation of “shared contexts” between humans and machines, policy makers, and the public and explores how human-machine systems affect targeted audiences (researchers, machines, robots, users, regulators, etc.) and society, as well as future ecosystems composed of humans, machines, and systems.

Title of Book
Cover image
Title page
Table of Contents
Copyright
List of contributors
Preface
1. Introduction to “autonomous human–machine teams”
1.1 Introduction
1.1.1 Background
1.2 Tradeoffs between structure and performance of teams
1.3 Introduction to the chapters
1.4 Conclusion
2. Toward a new foundation for AI
2.1 Introduction
2.1.1 NNs are not new
2.2 Current position of neural networks in the ACM1 Computer Science Curriculum
2.3 Anticipated Impact of NNs
2.4 Engineered intelligence
2.4.1 IBM Watson
2.4.2 Cyc
2.4.3 Deep Blue
2.5 Learned Intelligence
2.5.1 AlphaGoZero
2.5.2 AlphaZero
2.5.3 ChatGPT
2.6 Early criticism of symbol-manipulating AI
2.7 Reasoning and problem-solving with NN-based systems
2.7.1 LEGO
2.7.2 Simple logic puzzles
2.8 Simple common-sense reasoning
2.8.1 Reasoning with LLMs
2.9 Pattern recognition
2.9.1 Convolutional NNs (CNNs)
2.9.2 Transformers
2.9.3 Evaluation
2.10 Explainable AI
2.11 Autonomy
2.12 Conclusions
3. Human–machine teaming using large language models∗
3.1 Language: A most natural interface
3.2 The dream of talking machines
3.3 The importance of being human
3.3.1 The understanding of understanding
3.3.2 The Chinese room in the age of large language models
3.3.3 Living in the mental states
3.3.4 Who needs mental states
3.4 Language for human–machine teaming
3.5 Natural language and machines
3.6 Design of conversational machines
3.6.1 System language understanding
3.6.2 System vernacular
3.7 Risks of large language models
3.7.1 LLM application accuracy through better dialog
3.8 Concepts of LLM application design
3.8.1 Choosing an LLM
3.8.2 LLMs as components in systems
3.8.3 Principles of prompt engineering
3.9 Retrieval-augmented generation
3.10 Conclusion: The future of LLMs for HMI
4. Development of a team cohesion scale for use in human-autonomy team research
4.1 Introduction
4.1.1 Team cohesion in human autonomy teams background? Do we need background?
4.2 Scale development process
4.2.1 Phase 1: Item development
4.2.1.1 Function-based task cohesion
4.2.1.2 Structural cohesion (four dimensions)
4.2.1.3 Interpersonal cohesion (four dimensions)
4.2.1.4 Perceived team complementarity
4.2.1.5 Team resilience (three dimensions)
4.3 Phase 2: Scale development of initial item pool
4.4 Phase 3: Scale evaluation
4.5 Method
4.5.1 Instrumentation and facilities
4.5.2 Materials, tests, task, and stimuli
4.5.2.1 Questionnaire
4.5.2.2 Experimental task
4.5.3 Subjects
4.5.3.1 Sample size justification
4.5.4 Procedure
4.5.5 Experimental design
4.5.6 Data analysis
4.6 Results
4.6.1 Cohesion scale evaluation
4.6.2 Partial invariance
4.6.2.1 GEQ
4.6.2.2 Function
4.6.2.3 Exclusivity
4.6.2.4 Complementarity
4.6.2.5 Pride
4.6.2.6 Morale
4.6.2.7 Belongingness
4.6.2.8 Attraction to the group
4.6.2.9 Social
4.6.2.10 Leadership direction
4.6.2.11 Resilience: Team learning orientation
4.6.2.12 Resilience: Shared language
4.6.2.13 Resilience: Team flexibility
4.6.2.14 Resilience: Perceived efficacy of collective team action (PECTA)
4.6.2.15 Results summary
4.7 Discussion
4.8 Limitations and future directions
4.9 Conclusions and path forward
Appendix A: Original item pool with item retention recommendations
Function-based cohesion
a Task cohesion
Structural cohesion
a Exclusivity
b Individual attraction to team
c Leadership direction (vertical cohesion)
Interpersonal cohesion
a Team pride
b Social cohesion
c Belongingness
d Morale
Perceived complementarity
Resilience
a Mastery approaches
b Social capital: Shared language
c Collective efficacy: Perceived efficacy for collective team action
5. Enabling human–machine symbiosis: Automated establishment of common ground and estimates of the topological structures of Commander’s Intent
5.1 Introduction
5.2 Related work
5.2.1 Symbiotic artificial intelligence
5.2.2 Interdependence and joint activity theory
5.3 Common ground
5.3.1 Worked example of common ground
5.3.2 Estimating the underlying topology
5.3.3 Ontologies and ontology logs
5.3.4 Formally establishing common ground
5.4 Results and conclusions
6. Measuring consequential changes in human-autonomous system interactions
6.1 Introduction
6.2 Method
6.2.1 The baseline interface and tasks
6.2.2 Initial HMM development
6.2.3 Model comparison metric
6.3 Results
6.3.1 Same people, same interface
6.3.2 Different people and different interfaces
6.3.3 Same versus different people
6.3.4 Same people, modified interface
6.3.4.1 Observation alignment
6.3.5 Different people, similar interface, different tasks
6.4 Discussion
6.5 Limitations
6.5.1 Methodological
6.5.2 Practical
6.6 Conclusions
7. User affordances to engineer open-world enterprise dynamics
7.1 Background
7.2 An illustrative example
7.3 Categorical, situated reasoning
7.4 Challenges
7.4.1 A goal, a possibilistic network presentation
7.5 Three mathematical foundations
7.5.1 Situation and channel theories
7.5.2 Ordered valuation algebras
7.5.3 Constructivist systems and synthetic differential geometry
7.6 A tentative relationship of concepts with candidate visual grammars
7.6.1 CaNeTA, an intermediate user-facing capture model
7.6.2 Assigning dynamic ontology influence
7.6.3 Three levels of analysis
7.6.4 Self-organizing enterprises: Entropy and influence
8. Truth-O-Meter: Collaborating with LLM in fighting its hallucinations
8.1 Introduction
8.1.1 Why LLMs hallucinate
8.1.1.1 Introductory example
8.1.2 Related work
8.1.3 Iterative mode
8.1.4 Handling multiple mutually inconsistent facts obtained from authoritative sources
8.1.5 Correcting factual errors in syntactic and semantic spaces
8.1.5.1 Fact-checking by question answering against sources and syntax-semantic alignment
8.1.5.2 Example of alignment
8.1.6 Evaluation
8.1.7 Hallucination types
8.1.7.1 Token-level hallucination correction
8.1.7.2 Evaluation against fact extraction and verification annotation platform
8.1.8 Automated evaluation on QA datasets
8.1.8.1 Information-seeking dialogues
8.1.8.2 Personalized drug intake recommendation domain
8.1.8.3 Error analysis
8.1.8.4 Examples of repairs for hallucination
8.1.9 Discussions
8.1.10 Conclusions
9. Natural versus artificial intelligence: AI insights from the cognitive sciences
9.1 Introduction: Natural versus artificial intelligence
9.1.1 A path to machine intelligence
9.2 Insights from the cognitive sciences
9.2.1 Insight 0: Inductive learning strategies are brittle
9.2.2 Insight 1: “Learning” and “intelligence” are semantically overloaded terms
9.2.3 Insight 2: Removing technological barriers is necessary but not sufficient
9.2.4 Insight 3: Top-down processing offers a way forward
9.3 The argument for a cognitive concept of AI
9.4 Conclusions
10. Intention when humans team with AI
10.1 Introduction
10.2 State of human–AI teaming
10.3 Integration
10.3.1 Debate over intention
10.3.2 Challenge of collective intent
10.4 Conclusion
11. Autonomy: A family resemblance concept? An exploration of human–robot teams
11.1 Introduction
11.2 Teams and teammates
11.2.1 A small autonomous robot team
11.2.2 Human teams employ inference: Theory of mind
11.2.3 Gameplay between teams of humans and AI
11.2.4 Team conclusion
11.3 Autonomy
11.3.1 Autonomy in humans
11.3.2 Autonomy in groups
11.3.3 Autonomy in robots
11.3.4 Human versus robot autonomy
11.3.4.1 An illustration—Hanabi revisited
11.4 Summary and concluding remarks
11.4.1 Family resemblance
11.4.2 Conclusions
Funding
12. A theoretical approach to management of limited attentional resources to support the m:N operation in advanced air mobility ecosystem
12.1 Introduction
12.2 AAM technology and ecosystem
12.3 Theories of attention allocation
12.4 Trust in human–automation interaction and human–autonomy teaming
12.5 A model of human–AAM technology interaction centering on human attentional limits
12.6 Situation assessment
12.7 Trust
12.8 Overall description of a model of human–AAM technology interaction
12.9 Conclusion
13. Predicting workload of dispatchers supervising autonomous systems
13.1 Introduction
13.2 Simulator of Humans and Automation in Dispatch Operations (SHADO)
13.3 Rail case study
13.3.1 Validation
13.3.1.1 Data validation
13.3.1.2 Black-box validation
13.3.2 Predicting changes in rail dispatchers' workload due to automation
13.4 Modeling surface transportation dispatchers that supervise autonomous vehicles
13.4.1 Single dispatcher operations
13.4.2 Two dispatcher operations
13.5 Limitations
13.6 Conclusion
14. The generative AI weapon of mass destruction: Evolving disinformation threats, vulnerabilities, and mitigation frameworks
14.1 Introduction
14.1.1 Overview of GenAI
14.1.2 Significance of AI in modern disinformation
14.1.3 Definitions and scope
14.1.4 Purpose and structure of the chapter
14.2 Understanding GenAI
14.2.1 GenAI integrity manipulation
14.2.2 Capabilities and applications in content generation
14.2.3 Deepfakes and synthetic media
14.3 The role of GenAI in propaganda
14.3.1 Historical context of disinformation techniques
14.3.2 Intersection of AI and propaganda: Use case
14.3.3 GenAI disinformation threats and vulnerabilities
14.3.3.1 AI hallucinations, model poisoning and deepfakes in disinformation campaigns
14.4 GenAI efficacy assessment frameworks
14.4.1 GenAI model testing and monitoring
14.4.1.1 Prompt integrity testing
14.4.1.2 GenAI Integrity Testing Methodology
14.4.1.3 AI Trust Framework and Maturity Model
14.5 Identifying GenAI risks and vulnerabilities
14.5.1 Identifying risks associated with GenAI in propaganda
14.5.2 Analysis of technological and societal vulnerabilities
14.5.3 Threat modeling: Assessing likelihood and severity of threats
14.6 Mitigation strategies
14.6.1 Governance, regulatory and ethical measures
14.6.2 Neuro-symbolic reasoning
14.6.3 Cyber Kill Chain
14.6.4 MITRE ATLAS
14.6.4.1 GenAI security best practices & frameworks
14.6.4.2 NIST AI Risk Management Framework
14.6.4.3 Use case: Applying MITRE ATLAS to an adversarial AI incident
14.7 Conclusion
14.7.1 Summary of key points
14.7.2 Outlook and challenges
14.7.3 Final remarks on managing GenAI in propaganda
15. Ethics for artificial agents: Toward commensurate capability and self-regulation
15.1 Introduction
15.2 Approach
15.3 Implementing ethics sensitivity in machines
15.3.1 Nonharmful (nonmaleficent) machines
15.3.2 Helpful (beneficent) machines
15.3.3 Dutiful (responsible) machines
15.3.4 Trustworthy and transparent machines
15.4 Scientific accounts of mJDM
15.4.1 Evolutionary accounts
15.4.2 Psychological accounts
15.4.3 Neuroscientific accounts
15.4.4 Integrating mJDM accounts to scope a version of the capacity in machines
15.5 Major design policy themes around artificial ethical agents
15.5.1 Design policy around endogeneity
15.5.2 Design policy around generality
15.5.3 Design policy around mentalization
15.5.4 Design policy around legibility
15.5.5 Design policy around duty responsiveness
15.6 Conclusion
16. Self-visualization for the human–machine mind–body problem
16.1 Foreword: About this paper
16.1.1 Relevance of cognitive entropy
16.2 Introduction
16.2.1 Computation functionalism and the polarity issue between computer science and logic
16.2.2 The mind-body problem
16.2.3 Precedence relations and m-compactness
16.2.4 Consistency forces us to merge Physicalism and Idealism
16.3 Virtual consciousness and machine consciousness
16.3.1 Decision-making and higher-order representations in machine consciousness
16.3.2 Vanishing of first-order representations, Kant's principle, and rejection of the empirical basis
16.3.3 Meta-consciousness
16.3.4 The algebraic type degeneration assumption
16.3.5 Trialism, local monism, and virtual thinking
16.3.6 Hopf algebras, trialism as oscillations on infinity
16.3.7 Machine consciousness and video games
16.4 Human–machine interactions and the mind-body problem
16.4.1 Consistency at Planck's length and alogism
16.4.2 Descartes' diagrams
16.4.3 Completeness of the alogical doubt state
16.4.4 Descartes' diagrams and quantum physics
16.5 Quantum physics and machine consciousness
16.5.1 Sense-making, quantum measurement and consciousness
16.5.2 Wigner's numerous friends
16.5.3 Consciousness and the m-property
16.5.4 Identities and co-identities or the problem of initial conditions
16.6 Conclusion
17. Knowledge, consciousness, and debate: advancing the science of autonomous human–machine teams
17.1 Introduction
17.2 A review of knowledge across selected disciplines
17.2.1 What is knowledge to systems engineering
17.2.2 What is knowledge to philosophers?
17.2.3 What is knowledge to social scientists?
17.2.3.1 What is knowledge to citizens making decisions to cleanup wastes
17.2.3.2 What is knowledge under autocrats?
17.2.3.3 What is knowledge to consciousness?
17.2.4 What is knowledge to information theorists?
17.2.4.1 Intuition
17.2.4.2 Details
17.2.5 What is knowledge to physicists?
17.3 What is knowledge to us?
17.4 Discussion
17.4.1 Systems engineering
17.4.2 Philosophy
17.4.3 Social science
17.4.3.1 Citizen recommendations to government agencies for the cleanup of their wastes
17.4.3.2 Business
17.4.4 Consciousness
17.4.4.1 Authoritarianism
17.4.4.2 Information theory
17.4.5 Physics
17.5 Conclusions
Index

William Lawless

William Lawless is professor of mathematics and psychology at Paine College, GA. For his PhD topic on group dynamics, he theorized about the causes of tragic mistakes made by large organizations with world-class scientists and engineers. After his PhD in 1992, DOE invited him to join its citizens advisory board (CAB) at DOE’s Savannah River Site (SRS), Aiken, SC. As a founding member, he coauthored numerous recommendations on environmental remediation from radioactive wastes (e.g., the regulated closure in 1997 of the first two high-level radioactive waste tanks in the USA). He is a member of INCOSE, IEEE, AAAI and AAAS. His research today is on autonomous human-machine teams (A-HMT). He is the lead editor of seven published books on artificial intelligence. He was lead organizer of a special issue on “human-machine teams and explainable AI” by AI Magazine (2019). He has authored over 85 articles and book chapters, and over 175 peer-reviewed proceedings. He was the lead organizer of twelve AAAI symposia at Stanford (2020). Since 2018, he has also been serving on the Office of Naval Research's Advisory Boards for the Science of Artificial Intelligence and Command Decision Making.

Affiliations and expertise

Department of Mathematics, Sciences and Technology, and Department of Social Sciences, School of Arts and Sciences, Paine College, Augusta, GA, USA

Ranjeev Mittu

Ranjeev Mittu is the Branch Head for the Information and Decision Sciences Branch within the Information Technology Division at the U.S. Naval Research Laboratory (NRL). He leads a multidisciplinary group of scientists and engineers conducting research and advanced development in visual analytics, human performance assessment, decision support systems, and enterprise systems. Mr. Mittu’s research expertise is in multi-agent systems, human-systems integration, artificial intelligence (AI), machine learning, data mining and pattern recognition; and he has authored and/or coedited eleven books on the topic of AI in collaboration with the national and international scientific communities spanning academia and defense. Mr. Mittu received a Master of Science Degree in Electrical Engineering in 1995 from The Johns Hopkins University in Baltimore, MD.

The views expressed in this Work do not necessarily represent the views of the Department of the Navy, the Department of Defense, or the United States.

Affiliations and expertise

Information and Decision Sciences Branch, US Naval Research Laboratory (NRL), Washington, DC, USA

Donald Sofge

Don Sofge is a computer scientist and roboticist at the Naval Research Laboratory (NRL) with 36 years of experience in artificial intelligence, machine learning, and control systems R&D, the last 23 years at NRL. He leads the Distributed Autonomous Systems Section in the Navy Center for Applied Research in Artificial Intelligence (NCARAI), where he develops nature-inspired computing paradigms to challenging problems in sensing, artificial intelligence, and control of autonomous robotic systems. He has more than 200 refereed publications including 12 edited books in robotics, artificial intelligence, machine learning, planning, sensing, control, and related disciplines.

The views expressed in this Work do not necessarily represent the views of the Department of the Navy, the Department of Defense, or the United States.

Affiliations and expertise

Navy Center for Applied Research in Artificial Intelligence, US Naval Research Laboratory (NRL), Washington, DC, USA

Hesham Fouad

Hesham Fouad is the section head for the Intelligent Decision Support Section within the Information Technology Division at the U.S. Naval Research Laboratory (NRL). Dr. Fouad has over 35 years of experience as a Computer Scientist working in both industry and academia. He began his career at IBM Advanced Technologies where he worked on the first commercially available expert system development and runtime environment. Expert System Environment (ESE) was developed through a collaborative effort between the AI team at Stanford University and IBM. Dr. Fouad worked with the Stanford team to integrate new capabilities into ESE such as Frame Based Reasoning. Dr. Fouad also worked with the Voice Recognition group, led by Kai-Fu Lee, at Carnegie Mellon university to transition their S&T into IBM’s Via Voice product. Finally, Dr. Fouad worked with the MIT media lab’s Project Athena group on a collaborative effort to integrate a Hypermedia capability into the OS/2 operating system.

Dr. Fouad left IBM to pursue a Doctoral Degree in Computer Science at The George Washington University where he conducted his dissertation research on optimal real-time scheduling algorithms for Imprecise Computations using a fair scheduling strategy. This work was the basis of a startup company that Dr. Fouad founded and ran for 13 years. During that time, Dr. Fouad developed and managed the production of a line of commercial software for synthetic spatial audio design and production. He also conducted research on, and managed several ONR funded BAA and SBIR awards on adaptive training in virtual environments during that time.

Dr. Fouad maintained his involvement with academia both as an adjunct professor at the George Washington University, and as chair of a newly created undergraduate degree program in Computer Science with a focus on Game Development at the Art Institute of Washington. Since joining NRL, Dr. Fouad has led numerous efforts with three transitions to-date. He has procured millions in funding from Navy, Marines, Army, and OUSD sources. He has mentored SEAP students and has taken on a variety of roles within The Technical Cooperation Program.

Affiliations and expertise

Information Management and Decision Architectures Branch, Information Technology Division, U.S. Naval Research Laboratory, Washington, DC, USA

Life Sciences

Physical Sciences & Engineering

Social Sciences & Humanities

Health

Interdependent Human-Machine Teams

The Path to Autonomy

Purchase options

World Book Day Celebration!

Institutional subscription on ScienceDirect

William Lawless

Ranjeev Mittu

Donald Sofge

Hesham Fouad

Related books