AI Ethics & Concepts

Cheatsheet Content

1. Ethics in Artificial Intelligence Moral principles, guidelines, and frameworks for fair, accountable, transparent, and beneficial AI development and use. Key Ethical Principles: Fairness & Non-discrimination: Avoid biases (race, gender, etc.) by using unbiased data and algorithms. Transparency & Explainability (XAI): AI decisions must be understandable to humans; avoid "black boxes." Accountability & Responsibility: Clear responsibility for AI outcomes; who is accountable if harm occurs? Privacy: Design with privacy-by-design, data protection, user consent (e.g., GDPR). Safety & Reliability: Robust, secure, and reliable systems; prevent malicious use. Human Autonomy & Control: AI augments human decision-making; "human-in-the-loop" is crucial. 2. Importance of Ethical AI Prevents Harm & Discrimination: Safeguards against amplifying societal biases. Builds Public Trust: Essential for societal acceptance and adoption of AI. Ensures Legal & Regulatory Compliance: Adherence to evolving AI regulations (e.g., EU AI Act). Promotes Long-term Sustainability: Aligns AI with human values for positive evolution. Mitigates Business Risk: Protects against financial losses, lawsuits, and brand damage. Guides Responsible Innovation: Navigates moral dilemmas (e.g., autonomous weapons). 3. Impact of AI on Jobs Job Displacement (Automation): AI automates routine tasks (data entry, customer service), risking unemployment in some sectors. Job Transformation (Augmentation): AI augments human capabilities (e.g., doctors using AI for diagnostics), shifting job requirements to critical thinking, creativity, and AI management. Job Creation: New roles emerge (AI Trainers, Ethicists, Data Scientists, ML Engineers). Overall Effect: Shift in the nature of work, requiring reskilling and education to manage transition. 4. AI as a Service (AIaaS) Mechanism Cloud-based service model offering pre-built AI tools, APIs, and infrastructure. Key Mechanisms & Components: Cloud-Based Infrastructure: Provider hosts hardware (GPUs/TPUs), software, algorithms. APIs & Microservices: Core AI functions as APIs for: Computer Vision (e.g., image recognition) Natural Language Processing (e.g., sentiment analysis) Speech Services (speech-to-text) Pre-trained ML Models (prediction, recommendation) Machine Learning Platforms: Platforms for custom model building, training, deployment. Bots & Robotics: Chatbot frameworks, Robotic Process Automation (RPA) tools. Pay-as-you-go Model: Usage-based pricing, lowering entry barriers. How it Works: Companies subscribe to an AI API, send data, and receive results without managing underlying AI complexity. 5. Recent Trends in AI Generative AI & LLMs: Dominance of models like GPT-4, Gemini, DALL-E for text, code, image generation. Multimodal AI: Systems processing multiple input types (text, image, audio, video) simultaneously. AI Democratization: Low-code/no-code AI platforms making AI accessible to non-experts. Explainable AI (XAI): Increased focus on interpretability and transparency of complex AI models. AI for Science and Discovery: Accelerating breakthroughs in drug discovery, materials science, climate modeling. Edge AI: Running AI algorithms directly on local devices (smartphones, IoT) for lower latency and privacy. AI Governance and Regulation: Rapid development of frameworks (e.g., EU AI Act) for ethical and safe AI deployment. AI Agentic Workflows: Autonomous AI "agents" that break down goals, use tools, and execute multi-step tasks. 6. Expert System (ES) AI branch emulating human expert decision-making in a specific domain. Uses knowledge and inference procedures. Core Idea: Captures expert knowledge in a "knowledge base" and uses an "inference engine" to apply logical rules for problem-solving. Key Characteristics: Domain-Specific: Limited to a very specific field (e.g., medical diagnosis). Symbolic Reasoning: Uses IF-THEN rules to manipulate symbols. Heuristic: Incorporates expert rules-of-thumb. 7. Components of an Expert System 1. Knowledge Base: Stores domain-specific facts and rules (heuristics) from human experts. 2. Inference Engine: Applies logical rules from the knowledge base to deduce conclusions. Forward Chaining: Data-driven; starts with facts to reach a goal. Backward Chaining: Goal-driven; starts with a hypothesis and works backward. 3. User Interface: Allows non-expert users to interact, input data, and receive advice. 4. Explanation Facility: Justifies the system's reasoning, showing how and why a conclusion was reached. 5. Knowledge Acquisition Facility: Helps domain experts and knowledge engineers input and update knowledge. 8. Advantages, Disadvantages, and Applications of Expert Systems Advantages: Permanence & Consistency: Knowledge preserved indefinitely, applied consistently. Multiple Expertise: Can integrate knowledge from several experts. Availability & Cost-Effectiveness: 24/7 availability, reduces reliance on scarce human experts. Risk Reduction: Usable in hazardous environments. Training Tool: Can train new personnel. Disadvantages: Narrow Domain: Limited, lacks common sense. High Development Cost & Time: Knowledge acquisition is difficult and expensive. Inability to Learn: Cannot learn automatically; requires manual updates. Lack of Creativity: Cannot adapt to novel situations outside predefined rules. Applications: Medical Diagnosis (e.g., MYCIN) Configuration (e.g., XCON for computer systems) Financial Decision Making (loan approval) Process Monitoring & Control Customer Service (troubleshooting) 9. Traditional System vs. Expert System Feature Traditional System Expert System Processing Based on algorithms and precise data processing. Based on symbolic reasoning and heuristic rules. Knowledge Representation Data and programs are integrated; not easily separable. Knowledge Base (facts/rules) separated from Inference Engine. Problem-Solving Uses deterministic, step-by-step procedures. Uses inference, often with uncertainty/incomplete info. Domain General-purpose data processing. Very narrow, specific domain of expertise. Explanation Cannot explain its reasoning process. Has an Explanation Facility. Adaptability Changes require reprogramming the entire system. Updated by modifying the knowledge base. Error Handling Fails or crashes with unexpected inputs. Can handle incomplete data, provide probabilistic answers.