Types of AI — Your Friendly Neighborhood Taxonomy (but with more memes and fewer biology degrees)

"If AI were a high school, this is where we explain the cliques." — Probably your future robot friend

---

You're already not a complete newbie: we've already sketched the AI origin story in History of AI and compared machine smarts to human smarts in AI vs Human Intelligence. Now let's sort the party into who belongs in which group. This isn't just academic hair-splitting — these categories help answer practical questions like: what can current systems actually do, what should we realistically fear, and which sci-fi apocalypse belongs in the 'never' folder.

Big picture: Two popular axes for splitting AI

There are two common ways people classify AI. Think of them as two different maps of the same city:

1. By capability — how powerful and general the intelligence is (Narrow, General, Super).
2. By functionality — how the system perceives and reasons (Reactive, Limited Memory, Theory of Mind, Self-aware).

Both are useful. Capability tells you the scale of the brainpower. Functionality tells you how that brainpower operates.

---

Part A — By capability: how general is the smarts?

1) Narrow AI (also called Weak AI)

• Definition: Systems built to do one specific task extremely well.
• Examples: Voice assistants (Siri), recommendation engines, AlphaGo, current large language models when used for specific tasks.
• Why it matters: This is 99.9% of what exists today. Narrow AI can beat humans at a single thing, but it has no idea about literally everything else.

Analogy: Narrow AI is the sushi chef who can make the perfect nigiri but thinks a microwave is sorcery.

2) General AI (AGI — Artificial General Intelligence)

• Definition: A system that can perform any intellectual task a human can, across domains, and learn new tasks without being retrained from scratch.
• Status: Theoretical / research goal. Not achieved yet.

Analogy: AGI is the chef who not only cooks every cuisine but also manages the restaurant, negotiates supply contracts, and writes the Yelp reviews — effortlessly.

3) Superintelligence (ASI — Artificial Superintelligence)

• Definition: An intelligence that surpasses the best human minds in practically every field — creativity, wisdom, social skills, science.
• Status: Hypothetical, but the source of most dramatic ethical and safety debates.

Analogy: Superintelligence is if the chef becomes a culinary deity who can invent food that makes you cry tears of joy and solve world hunger before breakfast.

---

Part B — By functionality: how does the AI think?

This taxonomy is practical for engineers and cognitive scientists. It was popularized in AI education and gives a sense of progression.

1) Reactive Machines

• Definition: No memory; respond to inputs with fixed rules or learned policies.
• Example: IBM's Deep Blue (chess engine). It looks at the board and decides the best move; it doesn't learn from past games beyond search heuristics.
• Limitation: Can't use past experiences to inform new decisions.

2) Limited Memory

• Definition: Can use recent data to make decisions — the most useful category for real-world AI today.
• Examples: Self-driving cars (use sensor history to predict other vehicles), modern neural networks that use past examples during training and short-term context at runtime (like language models with a context window).

3) Theory of Mind (not yet achieved)

• Definition: Systems that understand that other agents have beliefs, desires, and intentions which affect their behavior.
• Why it's big: Social intelligence, negotiation, empathy — these require Theory of Mind.
• Status: Research stage. We can model some aspects, but no AI has full human-like theory-of-mind capabilities.

4) Self-aware (definitely sci-fi today)

• Definition: Systems with consciousness, self-reflection, subjective experience.
• Status: Hypothetical. Raises philosophical and ethical questions far beyond engineering.

---

Quick table: mapping capability to functionality

Capability	Typical functionality examples	Real today?
Narrow AI (ANI)	Reactive machines, limited memory	Yes
General AI (AGI)	Limited memory to Theory of Mind	No (goal)
Superintelligence (ASI)	Beyond Theory of Mind and Self-aware	No (speculative)

---

Real-world examples and the many times people get confused

• ChatGPT and friends are Narrow AI — even if they seem chatty and occasionally philosophical. They are powerful limited-memory systems trained on lots of data; they don't understand the way humans do (see AI vs Human Intelligence for that nuance).

• Self-driving research often uses limited memory plus predictive models — that's why they can drive in complex situations but still make mistakes human drivers wouldn't.

• Superintelligence stories are fun, but remember: history and current tech show slow, incremental progress. Sudden jumps to AGI or ASI would require breakthroughs we don't yet have evidence for.

---

Why people keep misunderstanding this

• Hollywood compresses decades into two hours of plot — so people think AGI is just a firmware update away.
• Marketing loves the word "AI" — everything from your thermostat to your sandwich press gets the AI badge.
• Clever systems can imitate human-like behavior well enough to trigger anthropomorphic assumptions.

Question to ask: When someone says "AI did X," ask what X is and whether that required general reasoning or a very specific optimization.

---

Practical takeaways (so you can sound smart at parties)

• Most AI today = Narrow AI: excellent at single tasks, not magically intelligent.
• Function matters: Reactive vs Limited Memory isn't just jargon — it determines what problems an AI can solve.
• AGI and ASI are research & ethics topics, not products: worth studying, but separate from today's deployed systems.

"Treat current AI like a very talented intern, not a demigod." — Also probably your future robot friend

---

Closing: What to learn next

• If you want hands-on skills, focus on limited-memory systems: supervised learning, reinforcement learning, and sequence models.
• If you're curious about long-term impacts, read up on AGI safety, ethics, and the philosophy of mind (we touched on intelligence differences in AI vs Human Intelligence).

Key final thought: Classifications are maps, not the territory. They help you navigate promises, products, and precautions — but always look under the hood of any "AI" claim.

Go forth with curiosity and skepticism. Bring snacks.