1Four Approaches to AI
Russell and Norvig's Artificial Intelligence: A Modern Approach defines AI along two dimensions: thought processes vs behaviour, and human-like vs rational. This produces four distinct schools of thought, each leading to different research methodologies.
Thinking Humanly — Cognitive Modelling
AI that thinks like a real human. Requires understanding how humans actually think. Draws on cognitive science and experimental psychology to build precise models of the human mind.
Acting Humanly — Turing Test Approach
AI that acts like a human regardless of internal mechanism. The Turing Test is the canonical benchmark. Requires capabilities: NLP, knowledge representation, automated reasoning, machine learning, computer vision, robotics.
Thinking Rationally — Laws of Thought
AI that reasons correctly using formal logic. Uses symbolic logic to capture the laws of rational thought. Challenge: much human reasoning is probabilistic, informal, and context-dependent — hard to reduce to symbols.
Acting Rationally — Rational Agent
AI that acts to achieve the best outcome. The rational agent approach is the dominant paradigm in modern AI. It does not require human-like reasoning — only that the agent acts to maximize its performance measure.
2Three Levels: ANI, AGI and ASI
AI capability is often classified by scope. Every AI system deployed today — every large language model, every self-driving car, every recommendation engine — falls into the first category.
ANI — Artificial Narrow Intelligence
Performs a single task or a narrow domain of tasks. Google Translate cannot recognize images; AlphaGo cannot write code. ANI is the only form of AI humanity has achieved. It is not conscious and not driven by emotions.
AGI — Artificial General Intelligence
Human-level intelligence — the ability to reason and perform any intellectual task a human can. Also called "Strong AI." AGI has not been achieved. It would require AI to understand abstraction, strategy, imagination, and consciousness — things we barely understand in humans.
ASI — Artificial Super Intelligence
Oxford philosopher Nick Bostrom: "any intellect that greatly exceeds the cognitive performance of humans in virtually all domains of interest." Entirely hypothetical. ASI is the subject of long-term AI safety research.
3AI Methodologies
Within the ANI category, there are four major methodologies for building intelligent systems. Each makes different assumptions about what "intelligence" means and how to achieve it computationally.
Search Agents
The agent perceives its environment via sensors, processes through a knowledge base, and acts via actuators. It treats the goal state as a target and tries to find a sequence of actions to reach it. Covered in the next post: AI Agents & Uninformed Search.
Symbolic AI (Rule-Based)
All domain rules are encoded by hand into a program. Fast at inference, slow to build. Dominant before the 1970s. Expert systems are a key example. Now largely replaced by ML for most tasks.
Machine Learning
Algorithms find correlations in data automatically without explicit programming. The machine learns all the rules from data. The dominant paradigm today. Includes supervised, unsupervised, and reinforcement learning.
Expert Systems
Computer systems that emulate decisions of human experts. Maintain a knowledge base of expert facts plus an inference engine. Still used in medicine and law where rules are relatively stable and interpretability is mandatory.
4Why This Taxonomy Matters
Choosing the wrong framework wastes time. If you need a system that explains its reasoning (medical diagnosis, legal decisions), a neural network's black-box nature makes it a poor fit — consider rule-based or expert system approaches. If you need to handle raw sensor data at scale, symbolic AI will not scale. Knowing the taxonomy helps you start with the right tool.