If you’ve looked up “types of AI”, you’ve probably seen the number seven. It turns up in online courses, textbooks and news explainers, usually with words like “narrow”, “general” and “self-aware” that sound more like a science fiction film than the phone in your pocket.
This guide explains all seven in plain English, with an everyday example of each. It also answers the question most people care about: which of these actually exist today?
You’ll also learn the other AI words you hear in the news, such as generative AI and AI agents, and how they fit together.
What are 7 types of AI?
The 7 types of AI usually listed are three types sorted by capability (narrow AI, general AI and super AI) and four types sorted by how they work (reactive machines, limited memory, theory of mind and self-aware AI). Only narrow AI, reactive machines and limited memory AI clearly exist today. The others are goals or ideas for the future.
Here’s the full list:
- Narrow AI (also called weak AI)
- General AI (also called AGI or strong AI)
- Super AI (also called artificial superintelligence or ASI)
- Reactive machines
- Limited memory AI
- Theory of mind AI
- Self-aware AI
Think of it as two ways of sorting the same thing, like sorting cars by size or by fuel type. One AI product usually fits somewhere in each list. IBM, for example, classes ChatGPT as narrow AI in the first list and as limited memory AI in the second.
The 3 types of AI by capability
The first list asks a simple question: how much can the AI do, compared with a person?
1. Narrow AI (weak AI)
Narrow AI is built to do one job, or one family of jobs, very well. It can’t step outside that job the way a person can. According to IBM, narrow AI is the only type of AI that exists today.
Everyday examples: Siri and Alexa, face recognition on your phone, email spam filters, map directions and chatbots such as ChatGPT and Google’s Gemini (which has replaced Google Assistant on Android).
“Weak” doesn’t mean feeble. A narrow AI can beat any human at chess or sort a million photos in seconds. It just can’t decide on its own to go and learn beekeeping.
Today’s chatbots stretch this idea. They can write, translate, summarize and help with homework, which is why some people argue they’re more than narrow. Most explainers, including IBM’s, still put them in this group.
2. General AI (AGI)
Plain talk: AGI
AGI stands for artificial general intelligence. It means AI that could learn and do almost any mental task a person can, and use what it learned in one area to handle something new.
IBM and Syracuse University’s School of Information Studies both describe general AI as theoretical. So does it exist? This is where the news gets confusing.
In September 2026, OpenAI released a new model called GPT-6 Astra, and its president, Greg Brockman, told reporters “Welcome to the AGI era,” according to Axios. Many AI researchers quickly disagreed. Part of the problem is that nobody agrees on a test for AGI. Britannica, updated this month, still describes AGI as “controversial and out of reach.”
The fair answer for now: some tech companies say AGI has arrived or is close, many experts dispute it, and there’s no agreed way to measure it. For everyday purposes, the AI you can use is still best treated as a clever tool, not a general mind.
3. Super AI (ASI)
Super AI would be smarter than the best humans at almost everything, from science to strategy to reading people. IBM describes it as theoretical.
This is the AI of the movies, and it’s what people usually mean when they worry about AI “taking over”. It does not exist.
The 4 types of AI by how they work
The second list asks a different question: can the AI remember and learn, and could it ever understand minds, including its own?
4. Reactive machines
A reactive machine has no memory. It looks at what’s in front of it right now and reacts the same way every time.
Everyday example: IBM’s Deep Blue, the chess computer that beat world champion Garry Kasparov in 1997. It judged each board position fresh, without learning from past games. Simple computer opponents in card and board games work in a similar way.
Exists today? Yes, and it has for decades.
5. Limited memory AI
Limited memory AI uses past and recent information to decide what to do. Most of the AI you use today falls here.
Everyday examples: a chatbot that remembers what you said earlier in the conversation, and a self-driving car that keeps track of the speed and position of nearby cars. IBM lists ChatGPT, virtual assistants and self-driving cars in this group.
“Limited” is the key word. The memory is short-term or narrow. Many chatbots forget everything when you start a new chat, unless you turn on a memory setting that saves some details.
Exists today? Yes.
6. Theory of mind AI
Theory of mind AI would understand that people have thoughts, feelings, beliefs and intentions, and would adjust how it behaves to match. IBM calls it theoretical, and Syracuse describes it as a future stage of AI.
Today’s chatbots can notice words like “I’m worried” and reply kindly. That’s pattern-matching from huge amounts of text, not real understanding of how you feel.
Exists today? No.
7. Self-aware AI
Self-aware AI would have a sense of itself: its own awareness, feelings and wants. Syracuse says it exists only in science fiction and theoretical discussions.
Exists today? No.
The 7 types at a glance
| Type | Exists today? | Example |
|---|---|---|
| Narrow AI | Yes | Siri, spam filters, ChatGPT |
| General AI (AGI) | Disputed | None agreed |
| Super AI (ASI) | No | Movie robots |
| Reactive machines | Yes | Deep Blue chess computer |
| Limited memory | Yes | Chatbots, self-driving cars |
| Theory of mind | No | None yet |
| Self-aware | No | Science fiction |
Is this an official list?
No. The “7 types of AI” is a popular teaching framework, not an official standard. No government or standards body defines AI this way. It’s two separate lists joined together, and you’ll see some sources teach only the “4 types” or only the “3 types”.
The lists also overlap. Reactive machines and limited memory AI are both kinds of narrow AI, described from a different angle.
Britannica points to a deeper problem: without a reasonably precise test for when a machine counts as intelligent, there’s no objective way to tell whether an AI project has succeeded. That’s a big reason the AGI debate keeps going in circles.
Other AI words you’ll hear in the news
The seven types are mostly about what AI could become. News stories, ads and app stores use different labels, based on what the AI does. These overlap too, and one product can be several at once.
- Generative AI creates new content such as text, pictures, audio, video or computer code when you ask, according to IBM’s definition. Example: ChatGPT drafting a letter, or an app making a picture from a description.
- Predictive AI uses past patterns to guess what’s likely next. Examples: your bank texting you about an unusual purchase, or a map app predicting traffic.
- Conversational AI talks back in everyday language, by text or voice. Examples: chatbots, Siri, Alexa and the chat windows on company websites.
- Computer vision understands pictures and video. Examples: face recognition, and searching your photos for “beach”.
- Speech and voice AI turns speech into text and back again. Examples: dictation, live captions and voice assistants. The same technology can copy a person’s voice, which is why every family should have a safe word against AI voice-clone scams.
- Recommendation systems suggest what you might like. Examples: “Because you watched…” on Netflix, or “customers also bought” when shopping online.
- AI agents take actions for you, not just give answers.
Plain talk: AI agent
An AI agent is AI that can carry out a task on its own for you, such as filling in a form or moving through websites, rather than only answering questions. IBM describes agents as systems that perform tasks on behalf of a user.
Agents are the newest of these words in everyday news. OpenAI’s GPT-6 Astra, for example, is designed to work directly inside software on a computer, not just suggest what to do next, according to Axios.
So what type is ChatGPT? It’s generative AI and conversational AI in news terms, and narrow, limited memory AI in the seven-type list. All of those labels are true at once.
What this means for you
You don’t need to memorize the seven types. But knowing them helps you read the headlines calmly:
- Every AI you can use is a tool. It’s good at some jobs and poor at others. It isn’t a mind that knows you.
- It can sound human without understanding you. Friendly words aren’t friendship, and a confident tone isn’t proof it’s right.
- Check anything important. AI makes mistakes a person wouldn’t, and it states them just as confidently. Our guide Is ChatGPT Safe and Free? covers the three safety rules that matter most.
- Be careful with agents. An AI that can act for you can also make mistakes for you.
Safety tip
Never give an AI agent or chatbot your passwords, bank or card numbers, or ID numbers such as a Social Security or Medicare number. If an AI tool offers to book or buy something, check every detail yourself before you confirm a payment.
A good way to test your understanding is to ask a chatbot to explain these ideas back to you in your own terms.
Type or say this
“Explain the difference between narrow AI and general AI using an example from cooking. Keep it short and simple.”
Where to go from here
If you’re helping a parent or grandparent make sense of all this, our guide on how to explain AI to an older person has analogies and a 10-minute first demo. For the key facts in one place, read what seniors should know about AI.
When you’re ready to try an assistant yourself, our book AI for Seniors, Plainly Explained (coming soon) walks you through setting one up safely and asking good questions, one step at a time.
