What are the 7 types of AI? It is one of the first questions people ask when they start learning about artificial intelligence, and the answer is simpler than it sounds. The "7 types" come from combining two ways of classifying AI: what it can do (3 types) and how it works (4 types). Understanding them helps you cut through hype and see where today's AI really stands. If you want to go further, Teyro teaches AI concepts like these in 15-minute gamified lessons.
Here is each type explained in plain English, with real examples and an honest look at which ones exist today.
Direct Answer: The 7 Types of AI
The 7 types of AI are grouped into two classifications. By capability: (1) Artificial Narrow Intelligence, (2) Artificial General Intelligence and (3) Artificial Superintelligence. By functionality: (4) reactive machines, (5) limited memory AI, (6) theory of mind AI and (7) self-aware AI. Only narrow AI, reactive machines and limited memory AI exist today.
| # | Type | Classification | Exists Today? | Example |
|---|---|---|---|---|
| 1 | Artificial Narrow Intelligence (ANI) | Capability | Yes | ChatGPT, Siri, Netflix recommendations |
| 2 | Artificial General Intelligence (AGI) | Capability | Not yet (debated) | Human-level AI across all tasks |
| 3 | Artificial Superintelligence (ASI) | Capability | No | AI surpassing humans in every domain |
| 4 | Reactive machines | Functionality | Yes | IBM Deep Blue (chess) |
| 5 | Limited memory AI | Functionality | Yes | Self-driving cars, chatbots |
| 6 | Theory of mind AI | Functionality | No | AI that truly understands emotions and beliefs |
| 7 | Self-aware AI | Functionality | No | Conscious AI with a sense of self |
Types by Capability: How Smart Is the AI?
1. Artificial Narrow Intelligence (ANI)
Also called weak AI, narrow AI is designed or trained for specific tasks. It can be superhuman within its domain but cannot reliably transfer that intelligence elsewhere.
Examples:
- Voice assistants like Siri and Alexa
- Recommendation systems on YouTube, Netflix and Spotify
- Spam filters and fraud detection
- Face recognition on your phone
- AI chatbots like ChatGPT, Claude and Gemini
Is ChatGPT narrow AI? Yes, by most definitions. Modern AI assistants are remarkably broad, handling writing, coding, analysis and conversation, but they still lack the full, flexible, reliable general intelligence humans have. This is why some researchers describe them as a very broad form of narrow AI, and why the boundary with AGI is hotly debated.
2. Artificial General Intelligence (AGI)
AGI would match human-level intelligence across any intellectual task: learning new skills quickly, reasoning in unfamiliar situations and transferring knowledge between domains.
Status: not achieved by any widely accepted standard. Rapid progress has intensified debate, and leading AI labs openly aim to build it, but experts disagree on both how to define AGI and when (or whether) it will arrive.
3. Artificial Superintelligence (ASI)
ASI would surpass the best human minds in every domain, including science, creativity, strategy and social skills.
Status: purely hypothetical. It is discussed mainly in AI safety research and long-term forecasting.
Types by Functionality: How Does the AI Work?
This classification, popularised by researcher Arend Hintze, describes AI by how it uses memory and understanding. It is also known as the 4 types of AI, covered in more technical depth in our guide to the 4 types of AI and machine learning.
4. Reactive Machines
The simplest type. Reactive machines respond to the current input with no memory of the past and no learning from experience.
Example: IBM's Deep Blue, which beat world chess champion Garry Kasparov in 1997. It evaluated the board in front of it but did not remember previous games or learn from them during play.
Other examples include simple rule-based game opponents and basic filters.
5. Limited Memory AI
Limited memory AI learns from historical data and can use recent information to make decisions. Almost all modern AI falls here.
Examples:
- Self-driving cars, which track the speed and position of nearby vehicles over the last few seconds
- Large language models like ChatGPT and Claude, trained on huge datasets and using the context of your conversation
- Recommendation engines that learn from your viewing history
- Image recognition trained on millions of labelled photos
"Limited" means the memory is temporary or fixed: a chatbot's model does not permanently learn from each conversation the way a person does (some products add memory features on top, but the underlying model stays the same until it is retrained).
6. Theory of Mind AI
Theory of mind is the human ability to understand that others have their own beliefs, emotions and intentions. An AI with genuine theory of mind would truly understand people, not just predict likely words about them.
Status: not achieved. Today's AI can often describe emotions convincingly and respond empathetically, but there is no scientific consensus that it genuinely understands mental states. Research continues.
7. Self-Aware AI
The final stage: AI with consciousness, a sense of self, and its own subjective experience and desires.
Status: purely hypothetical. We do not have a scientific way to measure consciousness in machines, and no current system is considered self-aware.
How the Two Classifications Fit Together
CAPABILITY
Narrow (ANI) ──> General (AGI) ──> Super (ASI)
│
│ Today's AI lives here
▼
FUNCTIONALITY
Reactive ──> Limited Memory ──> Theory of Mind ──> Self-Aware
(exists) (exists: today's (not yet) (hypothetical)
AI assistants)
Every AI product you use today is narrow AI with limited memory. The other types are goals, research directions or thought experiments.
Other Ways to Classify AI
You may also see AI grouped by technique or application, which is often more useful for learners:
| Classification | Types |
|---|---|
| By learning method | Supervised, unsupervised, reinforcement and self-supervised learning |
| By technique | Rule-based systems, machine learning, deep learning, generative AI |
| By application | Computer vision, natural language processing, speech recognition, robotics, recommendation systems |
| By output | Predictive AI (forecasts, classifications) vs generative AI (text, images, audio, code) |
If you are learning AI for a career, the technique and application categories are where the practical skills live. See what skills are needed for AI.
Why Knowing the Types of AI Matters
- You can separate hype from reality. When a headline claims AI is "conscious," you know that no self-aware AI exists.
- You understand limitations. Limited memory AI learns patterns from data, so it inherits data gaps and biases and can be confidently wrong.
- You can have informed conversations with colleagues, managers and clients about what AI can realistically do.
- It is the foundation for deeper learning. Next steps include how machine learning works and how to use AI tools effectively. Start with how to start learning AI.
Duolingo-Style Roadmap: Understand AI in 15 Minutes a Day
[Level 1: Novice (0–500 XP)] ──> The 7 types of AI, key vocabulary, AI in daily life
│ Goal: explain AI types to a friend
▼
[Level 2: Builder (500–1500 XP)] ──> How machine learning and LLMs work, their limits
│ Goal: use AI tools with informed judgement
▼
[Level 3: Pro (1500+ XP)] ──> Python, ML basics and building with AI
Goal: your first AI-powered project
The Bottom Line
The 7 types of AI are three capability levels (narrow, general, super) and four functionality levels (reactive, limited memory, theory of mind, self-aware). Everything in use today, from Netflix recommendations to ChatGPT, is narrow AI with limited memory. The rest are future possibilities, some debated and some purely hypothetical. Knowing the difference makes you a sharper, more confident AI learner.
Keep building your AI understanding with daily lessons on Teyro.
Related: The 4 types of AI and machine learning · Is AI hard to learn? · Can I learn AI for free? · Best AI for beginners
