by Albert Major

Introduction to Artificial Intelligence

INTRODUCTION TO AI ETHICS

• Article examines both present and future ethical challenges in AI development
• Written by leading experts Nick Bostrom and Eliezer Yudkowsky
• Addresses current practical issues and potential future scenarios
• Takes a balanced approach considering both benefits and risks

CURRENT AI ETHICAL CHALLENGES

1. Unintended Discrimination:
   • Example of mortgage approval AI showing racial bias despite being "race-blind"
   • AI can learn discriminatory patterns from historical data
   • May use proxy variables (like zip codes) that correlate with protected characteristics
2. Key Requirements for AI Systems:
   • Transparency: Decision-making process must be understandable
   • Predictability: Users should be able to anticipate AI behavior
   • Accountability: Clear responsibility for AI actions
   • Robustness: Resistance to manipulation and exploitation
   • Auditability: Ability to investigate and verify AI decisions
   • Error correction: Ability to identify and fix mistakes

TYPES OF ARTIFICIAL INTELLIGENCE

1. Domain-Specific AI:
   • Limited to specific tasks (chess, image recognition, etc.)
   • Can exceed human performance in narrow domains
   • Examples:
     • Deep Blue (chess)
     • Image recognition systems
     • Language processing tools
   • Limitations: Cannot transfer learning to other domains
2. Artificial General Intelligence (AGI):
   • Capable of human-like learning and adaptation
   • Features:
     • General problem-solving ability
     • Transfer learning between domains
     • Adaptability to new situations
   • Currently theoretical but actively researched
   • Represents a significant leap from current AI

MORAL STATUS AND RIGHTS

1. Current AI Systems:
   • No moral status
   • Treated as tools/property
   • No inherent rights
2. Future AI Moral Consideration Based On:
   • Sentience: Ability to experience feelings/sensations
   • Sapience: Self-awareness and reasoning capability
   • Consciousness: Subjective experiences
3. Non-Discrimination Principles:
   • Substrate Non-Discrimination: Physical composition shouldn't matter
   • Ontogeny Non-Discrimination: Origin/creation method shouldn't matter
   • Moral status should be based on capabilities, not composition

UNIQUE AI CHARACTERISTICS

1. Time Perception:
   • Subjective vs. objective time experience
   • Potential for vastly accelerated thinking
   • Ethical implications:
     • Justice and punishment
     • Fair treatment
     • Resource allocation