ONLINEAGENT_OPS 2026.Q3 HOME HUBSMODELSBUILDINGCOMPANIESFILMMAKINGSECURITYLIVING WITH ITSKILLSALL HUBS, ONE TERMINAL → ARTICLES CRAFT RECORD BLOG TOOLS SKILLS SEARCH /
HOME›ARTICLES›TYPES OF AI, EXPLAINED: NARROW, GENERAL AND
ARTICLES · EXPLAINERBEGINNER

Types of AI, Explained: Narrow, General and Generative

The types of AI in plain English: narrow, general and super AI by capability, and predictive, generative, conversational and agentic AI by the job they do.

READ11 min
WORDS2,106
SECTIONS7
TYPEEXPLAINER
CHECKED5 OCT 26

There are two standard ways to sort the types of AI, and mixing them up is what makes the question confusing. One sorts AI by how capable it is: narrow, general and super AI. Only narrow AI exists today. The other sorts AI by the job it does: predictive, generative, conversational and agentic AI. Every AI you use today is narrow AI doing one or more of those four jobs.

◈ IN PLAIN TERMS

Think of sorting vehicles. You can sort them by range: a city scooter, a car that can go anywhere on the roads, an imagined machine that could go anywhere at all. Or you can sort them by job: carrying people, carrying goods, racing. A delivery van has a range and a job. AI is the same: every system has a capability level and a job, and the two lists answer different questions.

◈ THE QUICK LIST: 11 TYPES OF AI, ONE LINE EACH
  1. Narrow AI built for one task or a small set; IBM says every other type is theoretical.
  2. General AI (AGI) AI that could handle nearly any thinking task a person can; still theoretical, with no agreed definition.
  3. Superintelligence (ASI) the idea of AI that outperforms the best humans at almost everything; it does not exist.
  4. Predictive AI learns from past data to spot patterns and forecast, like a spam filter or a fraud alert.
  5. Generative AI makes new text, pictures, sound or code from a request.
  6. Conversational AI talks with you in everyday words, like a voice assistant or a help-desk chatbot.
  7. Agentic AI works towards a goal and takes steps on its own, so its mistakes are actions.
  8. Reactive machines no memory and one very specific task, like the Deep Blue chess computer.
  9. Limited memory AI can recall past events; IBM puts ChatGPT, Siri, Alexa and self-driving cars here.
  10. Theory of mind AI would understand the thoughts and emotions of others; theoretical.
  11. Self-aware AI would be aware of its own internal conditions; theoretical.

Two ways to sort AI

The plain point: “how capable is it?” and “what does it do?” are separate questions, each with its own list.

Way of sortingThe question it answersThe types
By capabilityAsks: how wide a range of tasks can it handle, compared with a person?Types: narrow AI (exists), general AI (theoretical), superintelligence (theoretical).
By jobAsks: what does it actually do for you?Types: predictive, generative, conversational and agentic AI, all in use today.

A third list, sorting AI by memory and awareness, also turns up in searches. It is covered further down.

By capability: narrow, general and super AI

The plain point: this ladder runs from what exists today to what is only imagined.

Narrow AI: the only kind that exists

Narrow AI, also called weak AI, is built for one task or a small set of tasks. IBM, which builds and sells AI systems, says narrow AI “is the only type of AI that exists today. Any other form of AI is theoretical. It can be trained to perform a single or narrow task, often far faster and better than a human mind can.”1

Narrow does not mean weak in the everyday sense. The 2023 “Levels of AGI” research paper ranks some narrow systems as superhuman at their one job, naming “AlphaFold”, which predicts protein shapes, and the chess engine “StockFish”.2 Today’s chatbots stretch the label: they handle a huge range of topics, which is why some people argue they are partly general. Where they sit is disputed, as AGI meaning explains.

General AI (AGI): matching people across the board

Artificial general intelligence means AI that could handle nearly any thinking task a person can, including new ones it was not trained for. IBM calls it “today nothing more than a theoretical concept”, describing AI that “can use previous learnings and skills to accomplish new tasks in a different context”.3

There is no single agreed definition. OpenAI’s 2018 charter used an economic one, “Highly autonomous systems that outperform humans at most economically valuable work”, as the Levels of AGI paper records. That paper instead proposes a ladder of levels, and placed the frontier language models of 2023 at Level 1, “Emerging AGI”.4 AGI meaning maps the competing definitions.

Superintelligence: beyond any person

Superintelligence (or ASI, artificial superintelligence) is the idea of AI that outperforms the best humans at almost everything. The philosopher Nick Bostrom defined it in 1997 as “an intellect that is much smarter than the best human brains in practically every field, including scientific creativity, general wisdom and social skills.”5 IBM calls ASI “a hypothetical software-based artificial intelligence (AI) system with an intellectual scope beyond human intelligence.”6 Hypothetical is the key word: it does not exist, and whether or when it could is argued over. Is AI dangerous? sets out who says what.

TAKEAWAY

When someone says “AI”, they almost always mean narrow AI. General AI and superintelligence are ideas about the future, not products you can use.

By job: predictive, generative, conversational and agentic AI

The plain point: this is the list that describes the AI you actually meet, sorted by what it does.

Predictive AI: spots patterns and forecasts

The oldest everyday kind. IBM: “Predictive AI blends statistical analysis with machine learning algorithms to find data patterns and forecast future outcomes.” Its example: “Financial institutions use predictive AI models to forecast market trends”.7 Spam filters and “what to watch next” suggestions work in the same spirit: they learn from past data and make a call. IBM lists Netflix’s viewing recommendations as one example of narrow AI at work.3

Generative AI: makes new content

Generative AI creates things. IBM: it “responds to a user’s prompt or request with generated original content, such as audio, images, software code, text or video.”8 Chatbots writing emails and image makers drawing from a sentence are both generative AI. It is what most news stories mean by “AI” today.

Conversational AI: talks with you

IBM defines it as “technologies, such as chatbots or virtual agents, that users can talk to.”9 Voice assistants and chatbots belong here. Older ones followed scripts; today’s AI chatbots write each reply fresh using a large language model, which makes them conversational and generative at once. How does ChatGPT work? shows what happens inside.

Agentic AI: takes actions for you

The newest kind. IBM: “Agentic AI is an artificial intelligence system that can accomplish a specific goal with limited supervision.”10 An AI agent does not just answer; it uses tools such as a browser or your calendar and decides its own next step. Anthropic calls agents “systems where LLMs dynamically direct their own processes and tool usage”.11 Because it acts, its mistakes are actions, which is why agents need more care.

JobWhat it doesEveryday example
PredictiveDoes: learns from past data and makes a call or a forecast.Like: a spam filter, a fraud alert, a “you might like” list.
GenerativeDoes: makes new text, pictures, sound or code from a request.Like: a chatbot drafting an email, an app drawing a picture.
ConversationalDoes: talks with you in everyday words, typed or spoken.Like: a voice assistant, a help-desk chatbot.
AgenticDoes: works towards a goal and takes steps on its own.Like: an assistant that books the train rather than telling you how.

How the two lists fit together

The plain point: any real system gets one label from each list, and often several from the second.

Take a modern chatbot. On capability, it is narrow AI by IBM’s standard, since IBM counts nothing else as existing. On job, it is conversational and generative, and when it is allowed to use tools and act, agentic too. A bank’s fraud checker is narrow and predictive. A self-driving car is narrow, and mostly predictive: it reads the road and forecasts what will happen next. None of them is general AI.

That is also why “Is ChatGPT AGI?” has no simple yes or no. Under IBM’s list, no; under looser definitions, people argue it is partly there. The honest answer is to ask which definition is in use.

The other list: reactive, limited memory, theory of mind, self-aware

The plain point: many explainers also sort AI by how it handles memory and minds. Only the first two types exist.

IBM gives this list as “functionality-based” types. Reactive machines have “no memory and are designed to perform a very specific task”; its example is Deep Blue, the chess computer that “beat chess grandmaster Garry Kasparov in the late 1990s”. Limited memory AI can “recall past events and outcomes and monitor specific objects or situations over time”, and IBM puts ChatGPT, Siri, Alexa and self-driving cars here. Theory of mind AI would “understand the thoughts and emotions of other entities”, and self-aware AI would have awareness of “its own internal conditions”. Both of those are theoretical.3 When someone asks for “the four types of AI”, this is usually the list they mean.

Where machine learning, deep learning and LLMs fit

The plain point: these words describe how AI is built, not which type it is.

They nest inside each other. IBM describes the order: “Directly underneath AI, we have machine learning”, then “Deep learning is a subset of machine learning that uses multilayered neural networks”, and generative AI “refers to deep learning models that can create complex original content”.12 So a chatbot is narrow AI (capability), generative and conversational (job), built with machine learning and deep learning (method). Artificial intelligence meaning draws the family tree.

This page covers the standard categories. For the AI products you meet day to day, drawn as characters, see the eight kinds of AI you already use; for image, video, audio and text tools for creators, see understanding AI model types.

TAKEAWAY

To pin down any AI, ask three things: how capable is it (almost certainly narrow), what job does it do (predict, generate, converse or act), and how was it built. Most confusion comes from answering one question with a word from another list.

Questions people ask

Which type of AI is ChatGPT?

By capability it is narrow AI under IBM’s definition, and by job it is generative and conversational, becoming agentic when it uses tools. IBM also lists ChatGPT as an example of “limited memory” AI.13

Does AGI exist yet?

Not by the usual definitions: IBM calls it “nothing more than a theoretical concept”. Some researchers argue today’s chatbots are an early, partial step; AGI meaning lays out the dispute.3

What are the 4 types of AI?

Usually reactive machines, limited memory, theory of mind and self-aware AI, the list IBM calls functionality-based. Only the first two exist; the other two are theoretical.

Is generative AI the same as AI?

No. Generative AI is one job within AI: making new content. Spam filters and recommendation feeds are AI too, but they predict rather than create.

What is the difference between AI and machine learning?

AI is the whole field; machine learning is the main method inside it, where software learns from examples instead of following hand-written rules. See machine learning meaning.

QUICK CHECK · 3 QUESTIONS

Pick an answer to see if you got it. Nothing is recorded.

1. According to the page, what are the two standard ways to sort the types of AI?

  • By whether it is free or paid, and by who made it
  • By how capable it is, and by the job it does
  • By whether it works with text or with pictures
Show the answer

Answer: By how capable it is, and by the job it does. The page says one list sorts AI by capability (narrow, general, super) and the other by job (predictive, generative, conversational, agentic).

2. Which type of AI does the page say is the only kind that exists today?

  • Narrow AI
  • General AI (AGI)
  • Superintelligence
Show the answer

Answer: Narrow AI. The page quotes IBM saying narrow AI is the only type of AI that exists today, and any other form is theoretical.

3. Is generative AI the same thing as AI, according to the page?

  • Yes, the two words mean the same thing
  • Yes, because AI is mainly about making content
  • No, it is one job within AI; spam filters predict rather than create
Show the answer

Answer: No, it is one job within AI; spam filters predict rather than create. The page says generative AI is one job, making new content, while spam filters and recommendation feeds are AI that predict.

Related on this site: Generative AI vs Predictive AI · Narrow AI vs AGI vs Superintelligence

SOURCES
  1. 1IBM, Types of Artificial Intelligence, undated page, read at source 5 Oct 2026Artificial Narrow AI “is the only type of AI that exists today. Any other form of AI is theoretical. It can be trained to perform a single or narrow task, often far faster and better than a human mind can.” ↑
  2. 2Morris et al., Levels of AGI for Operationalizing Progress on the Path to AGI, arXiv:2311.02462 v5, 24 Sep 2025, read at source 5 Oct 2026 in the HTML full text, Table 1, Superhuman Narrow AI: “AlphaFold … AlphaZero … StockFish”.↑
  3. 3IBM, Types of Artificial Intelligence, read at source 5 Oct 2026General AI is “today nothing more than a theoretical concept. AGI can use previous learnings and skills to accomplish new tasks in a different context without the need for human beings to train the underlying models.” ↑a“Netflix’s viewing recommendations are powered by models that process data sets collected from viewing history”. ↑breactive machines have “no memory and are designed to perform a very specific task”; Deep Blue “beat chess grandmaster Garry Kasparov in the late 1990s”; limited memory AI can “recall past events and outcomes and monitor specific objects or situations over time”; theory of mind AI would “understand the thoughts and emotions of other entities”; self-aware AI would have awareness of “its own internal conditions and traits along with human emotions”. ↑cGeneral AI is “today nothing more than a theoretical concept.” ↑d
  4. 4Morris et al., Levels of AGI, v5, read at source 5 Oct 2026OpenAI’s charter definition, “Highly autonomous systems that outperform humans at most economically valuable work”. · read at source 17 Sep 2026 (as cited on this site’s AGI meaning page): “current frontier language models would therefore be considered a Level 1 General AI (“Emerging AGI”)”. ↑
  5. 5Nick Bostrom, How Long Before Superintelligence?, 1997, read at source 5 Oct 2026“an intellect that is much smarter than the best human brains in practically every field, including scientific creativity, general wisdom and social skills.” ↑
  6. 6IBM, What Is Artificial Superintelligence?, 14 Dec 2023, read at source 5 Oct 2026“Artificial superintelligence (ASI) is a hypothetical software-based artificial intelligence (AI) system with an intellectual scope beyond human intelligence.” ↑
  7. 7IBM, Generative AI vs. predictive AI: What’s the difference?, 9 Aug 2024, read at source 5 Oct 2026“Predictive AI blends statistical analysis with machine learning algorithms to find data patterns and forecast future outcomes.” · “Financial institutions use predictive AI models to forecast market trends, stock prices and other economic factors”. ↑
  8. 8IBM, Generative AI vs. predictive AI, 9 Aug 2024, read at source 5 Oct 2026“Generative AI (gen AI) is artificial intelligence that responds to a user’s prompt or request with generated original content, such as audio, images, software code, text or video.” ↑
  9. 9IBM, What is Conversational AI?, 28 Sep 2021, read at source 5 Oct 2026“Conversational artificial intelligence (AI) refers to technologies, such as chatbots or virtual agents, that users can talk to.” ↑
  10. 10IBM, What is Agentic AI?, undated page, read at source 5 Oct 2026“Agentic AI is an artificial intelligence system that can accomplish a specific goal with limited supervision.” ↑
  11. 11Anthropic, Building effective agents, 19 Dec 2024, read at source 1 Oct 2026 (as cited on this site’s AI agent page)“Agents, on the other hand, are systems where LLMs dynamically direct their own processes and tool usage”. ↑
  12. 12IBM, What Is Artificial Intelligence (AI)?, updated 15 Jun 2026, read at source 1 Oct 2026 (as cited on this site’s artificial intelligence page): “Directly underneath AI, we have machine learning” … “Deep learning is a subset of machine learning that uses multilayered neural networks” … “Generative AI … refers to deep learning models that can create complex original content”.↑
  13. 13IBM, Types of Artificial Intelligence, read at source 5 Oct 2026, which lists ChatGPT, Siri, Alexa and self-driving cars under limited memory AI.↑

Morris et al., Levels of AGI for Operationalizing Progress on the Path to AGI, arXiv:2311.02462 v5, 24 Sep 2025, read at source 5 Oct 2026 (and 17 Sep 2026).

The comparisons on this page (sorting vehicles by range and by job), the two-axis table and the examples of how real systems fit both lists (the chatbot, the fraud checker, the self-driving car) are this site’s own ways of explaining, not claims from these sources.

Free printable AI cheat sheets, and AI news every week.

Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.

SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK
aipromptgeneer://$ about this siteAI PROMPT GENEERAI explained in plain English. Every claim links to its source, with the date it was checked.NEWSLETTERWhat changed in AI each week, in plain English.SUBSCRIBE FREE ↗AI PROMPT GENEER · MACHINE ASSISTANCE DISCLOSEDEVERY ARTICLE CARRIES ITS OWN CHECKED DATE