AI: Unexplainable, Unpredictable, Uncontrollable (Chapman & Hall/CRC Artificial Intelligence and Robotics Series)

✍ Scribed by Roman V. Yampolskiy

• Publisher: Chapman and Hall/CRC
• Year: 2024
• Tongue: English
• Leaves: 264
• Edition: 1
• Category: Library

✦ Synopsis

Delving into the deeply enigmatic nature of Artificial Intelligence (AI), AI: Unexplainable, Unpredictable, Uncontrollable explores the various reasons why the field is so challenging. Written by one of the founders of the field of AI safety, this book addresses some of the most fascinating questions facing humanity, including the nature of intelligence, consciousness, values and knowledge.

Moving from a broad introduction to the core problems, such as the unpredictability of AI outcomes or the difficulty in explaining AI decisions, this book arrives at more complex questions of ownership and control, conducting an in-depth analysis of potential hazards and unintentional consequences. The book then concludes with philosophical and existential considerations, probing into questions of AI personhood, consciousness, and the distinction between human intelligence and artificial general intelligence (AGI).

Bridging the gap between technical intricacies and philosophical musings, AI: Unexplainable, Unpredictable, Uncontrollable appeals to both AI experts and enthusiasts looking for a comprehensive understanding of the field, whilst also being written for a general audience with minimal technical jargon.

✦ Table of Contents

Cover
Half Title
Series Page
Title Page
Copyright Page
Dedication
Contents
Acknowledgments
Author
1. Introduction
1.1. Introduction
1.2. The AI Control Problem
1.3. Obstacles to Controlling AI
1.4. Defining Safe AI
1.5. On Governability of AI
1.6. Conclusions
1.7. About the Book
References
2. Unpredictability
2.1. Introduction to Unpredictability
2.2. Predictability: What We Can Predict – A Literature Review
2.3. Cognitive Uncontainability
2.4. Conclusions
References
3. Unexplainability and Incomprehensibility
3.1. Introduction
3.2. Literature Review
3.3. Unexplainability
3.4. Incomprehensibility
3.5. Conclusions
Notes
References
4. Unverifiability
4.1. On Observers and Verifiers
4.2. Historical Perspective
4.3. Classification of Verifiers
4.4. Unverifiability
4.5. Unverifiability of Software
4.5.1. Unverifiability of Artificial Intelligence
4.6. Conclusions and Future Work
Notes
References
5. Unownability
5.1. Introduction
5.1.1. Proposals for Establishing Ownership
5.2. Obstacles to Ownership
5.3. Conclusions
References
6. Uncontrollability
6.1. Introduction
6.2. AI Control Problem
6.2.1. Types of Control Problems
6.2.2. Formal Definition
6.3. Previous Work
6.3.1. Controllable
6.3.2. Uncontrollable
6.4. Proving Uncontrollability
6.5. Multidisciplinary Evidence for Uncontrollability of AI
6.5.1. Control Theory
6.5.2. Philosophy
6.5.3. Public Choice Theory
6.5.4. Justice (Unfairness)
6.5.5. Computer Science Theory
6.5.6. Cybersecurity
6.5.7. Software Engineering
6.5.8. Information Technology
6.5.9. Learnability
6.5.10. Economics
6.5.11. Engineering
6.5.12. Astronomy
6.5.13. Physics
6.6. Evidence from AI Safety Research for Uncontrollability of AI
6.6.1. Value Alignment
6.6.2. Brittleness
6.6.3. Unidentifiability
6.6.4. Uncontainability
6.6.5. Uninterruptability
6.6.6. AI Failures
6.6.7. Unpredictability
6.6.8. Unexplainability and Incomprehensibility
6.6.9. Unprovability
6.6.10. Unverifiability
6.6.11. Reward Hacking
6.6.12. Intractability
6.6.13. Goal Uncertainty
6.6.14. Complementarity
6.6.15. Multidimensionality of Problem Space
6.7. Discussion
6.8. Conclusions
Notes
References
7. Pathways to Danger
7.1. Taxonomy of Pathways to Dangerous AI
7.1.1. On Purpose – Pre-Deployment
7.1.2. On Purpose – Post-Deployment
7.1.3. By Mistake – Pre-Deployment
7.1.4. By Mistake – Post-Deployment
7.1.5. Environment – Pre-Deployment
7.1.6. Environment – Post-Deployment
7.1.7. Independently – Pre-Deployment
7.1.8. Independently – Post-Deployment
7.2. Conclusions
References
8. Accidents
8.1. Introduction
8.2. AI Failures
8.2.1. Preventing AI Failures
8.3. AI Safety
8.4. Cybersecurity vs. AI Safety
8.5. Conclusions
Notes
References
9. Personhood
9.1. Introduction to AI Personhood
9.2. Selfish Memes
9.3. Human Indignity
9.4. Legal-System Hacking
9.5. Human Safety
9.6. Conclusions
Notes
References
10. Consciousness
10.1. Introduction to the Problem of Consciousness
10.2. Test for Detecting Qualia
10.3. Computers Can Experience Illusions, and so Are Conscious
10.3.1. Qualia Computing
10.4. Purpose of Consciousness
10.4.1. Qualia Engineering
10.5. Consciousness and Artificial Intelligence
10.6. Conclusions and Conjectures
Note
References
11. Personal Universes
11.1. Introduction to the Multi-Agent Value Alignment Problem
11.2. Individual Simulated Universes
11.3. Benefits and Shortcomings of Personalized Universes
11.4. Conclusions
Note
References
12. Human ≠ AGI
12.1. Introduction
12.2. Prior Work
12.3. Humans Are Not AGI
12.4. Conclusions
Note
References
13. Skepticism
13.1. Introduction to AI Risk Skepticism
13.2. Types of AI Risk Skeptics
13.2.1. Skeptics of Strawman
13.3. Arguments for AI Risk Skepticism
13.3.1. Priorities Objections
13.3.2. Technical Objections
13.3.3. AI Safety-Related Objections
13.3.4. Ethical Objections
13.3.5. Biased Objections
13.3.6. Miscellaneous Objections
13.4. Countermeasures for AI Risk Skepticism
13.5. Conclusions
Notes
References
Index