Pass path: Learn the 4 AI perspectives (Russell & Norvig matrix), Turing Test setup, and ethical pillars of responsible AI. Top path: Contrast rational agent approach with human-centric emulation, trace historical AI winters, and evaluate algorithmic bias mitigation frameworks.
Part 1: Core Concepts (2 Marks)
★ Artificial Intelligence (AI): The study and design of intelligent agents capable of perceiving their environment through sensors and taking autonomous actions through actuators that maximize their expected probability of achieving goals. ★ The Turing Test: An operational test of intelligence proposed by Alan Turing (1950) where a human interrogator questions a hidden human and machine via text; if the interrogator cannot reliably distinguish the machine from the human after 5 minutes, the machine passes. ★ Acting Rationally vs Thinking Rationally: Thinking rationally focuses on formal logic and correct inferential thought processes (“laws of thought”); acting rationally focuses on rational agents that act so as to achieve the best expected outcome given available percepts, even under uncertainty.
Part 2: Structural Frameworks (10 Marks)
Four Perspectives of AI and Foundational Disciplines
- Russell & Norvig’s 2x2 AI Taxonomy:
- Acting Humanly: The Turing Test approach (requires NLP, knowledge representation, automated reasoning, machine learning).
- Thinking Humanly: Cognitive modeling approach (validates theories of human cognition using psychological experiments or brain imaging).
- Thinking Rationally: The “Laws of Thought” approach (formal logic and syllogisms codified in Aristotelian systems).
- Acting Rationally: The Rational Agent approach (standard modern engineering framework; agents maximize expected utility).
- Interdisciplinary Foundations:
- Philosophy: Formal logic, empiricism, mind-body dualism, epistemology.
- Mathematics: Probability theory, graph theory, computability (Gödel’s Incompleteness).
- Economics: Decision theory, game theory, utility maximization.
- Neuroscience: Study of biological neurons and cortical architectures.
- Responsible AI & Ethics:
- Algorithmic Bias: Training on historical datasets reproduces societal discrimination.
- Explainability / Transparency: Critical in high-stakes domains (healthcare, criminal justice).
- Accountability: Determining legal liability for autonomous vehicle crashes or medical misdiagnoses.
Part 3: Distinction Tables
| Criteria | Human-Centric AI (Emulation) | Rationality-Centric AI (Optimization) |
|---|---|---|
| Core Benchmark | Human behavior, cognitive quirks, and empirical tests | Maximum expected utility and mathematical optimality |
| Verification Method | Psychological experiments and Turing Test | Mathematical proofs and performance metric evaluation |
| Tolerance for Errors | Replicates human cognitive biases and limits | Eliminates irrational human errors to maximize goal achievement |
| Dominant Paradigm | Cognitive science, affective computing | Modern machine learning, robotics, automated planning |
Part 4: Worked Problems & Exam Traps
- Easy: What four capabilities must a computer possess to pass the Total Turing Test?
- Answer: Natural Language Processing, Knowledge Representation, Automated Reasoning, and Computer Vision / Robotics (for physical interaction).
- Exam-level: Explain why Russell & Norvig argue that the “Acting Rationally” perspective is superior to the “Acting Humanly” perspective as an engineering foundation for AI.
- Answer: 1. It is mathematically more general than the laws of thought approach (correct inference is only one mechanism for achieving rationality). 2. It is more amenable to scientific development because the standard of rationality is clearly defined mathematically (maximizing utility), whereas human behavior is idiosyncratic, culturally variable, and computationally sub-optimal.
- Twist variant: Does a thermostat satisfy the technical definition of an intelligent agent?
- Answer: Yes, in its simplest form. It possesses sensors (thermometer), actuators (heater switch), and a simple reflex condition-action rule to maintain target room temperature.
Students lose marks here because they define AI solely as “machines that think like humans”, omitting the rational agent optimization paradigm that forms the bedrock of modern computer science.