04. Types of AI

Two Ways to Classify AI

AI is commonly classified by:

Capability — how broad is its intelligence?
Functionality — how does it learn and remember?

graph TD
    AI[Artificial Intelligence]
    AI --> CAP[By Capability]
    AI --> FUNC[By Functionality]

    CAP --> NAI["Narrow AI\nExists today"]
    CAP --> AGI["General AI - AGI\nResearch stage"]
    CAP --> ASI["Super AI - ASI\nHypothetical"]

    FUNC --> RM["Reactive Machines\nNo memory"]
    FUNC --> LM["Limited Memory\nUses recent past"]
    FUNC --> TOM["Theory of Mind\nResearch only"]
    FUNC --> SA["Self-Aware\nDoes not exist"]

    style NAI fill:#14532d,stroke:#059669,color:#e2e8f0
    style AGI fill:#78350f,stroke:#d97706,color:#e2e8f0
    style ASI fill:#450a0a,stroke:#ef4444,color:#e2e8f0
    style RM fill:#14532d,stroke:#059669,color:#e2e8f0
    style LM fill:#14532d,stroke:#059669,color:#e2e8f0
    style TOM fill:#78350f,stroke:#d97706,color:#e2e8f0
    style SA fill:#450a0a,stroke:#ef4444,color:#e2e8f0

By Capability

Narrow AI (Weak AI)

Designed to perform one specific task — and only that task.

ChatGPT → generates text
AlphaGo → plays Go
Spotify → recommends music
Tesla Autopilot → drives cars

This is all that exists today. Narrow AI can outperform humans in its domain but has zero ability outside it.

General AI (AGI — Artificial General Intelligence)

A machine that can perform any intellectual task a human can.

Understands context across domains
Learns new skills without retraining
Reasons, plans, and adapts broadly

Does not exist yet. Active research area. Timelines debated — decades away at minimum.

Super AI (ASI — Artificial Superintelligence)

An AI that surpasses human intelligence in every domain.

Hypothetical
Would self-improve recursively
Subject of significant philosophical and safety debate

By Functionality

Type	Memory	Examples
Reactive Machines	No memory, no learning — pure input/output	Deep Blue (chess), early game AI
Limited Memory	Uses recent past to inform decisions	Self-driving cars, recommendation systems
Theory of Mind	Understands emotions/intentions of others	Research stage only
Self-Aware	Conscious, aware of own existence	Hypothetical, does not exist

Where We Are Today

Narrow AI ✓  (exists, widely deployed)
AGI       ✗  (research, no consensus on timeline)
ASI       ✗  (hypothetical)

Reactive     ✓  (Deep Blue)
Limited Mem  ✓  (ChatGPT, Tesla, Netflix)
Theory of Mind ✗  (research)
Self-Aware   ✗  (science fiction)

Other Common Classifications

Term	Meaning
Supervised Learning AI	Trained on labeled data
Unsupervised Learning AI	Finds patterns without labels
Reinforcement Learning AI	Learns via reward/punishment
Generative AI	Creates new content (text, images, audio)
Discriminative AI	Classifies / makes decisions

Interview Questions

Q: What is the difference between Narrow AI and AGI?

A: Narrow AI excels at one specific task — like translating text or detecting objects in images — but has no ability outside that domain. AGI would handle any intellectual task a human can, across all domains. Only Narrow AI exists today; AGI is still a research goal.

Q: What are the four types of AI by functionality?

A: (1) Reactive Machines — no memory, pure stimulus-response; (2) Limited Memory — uses recent history to improve decisions; (3) Theory of Mind — understands others’ mental states (not yet achieved); (4) Self-Aware — conscious AI (purely hypothetical). All practical AI today falls into the Limited Memory category.