what sign up
nixm nixm

← learn

Types of AI: the textbook lists, and the one that actually helps

October 14, 2026

There are two textbook ways to sort AI. By capability: narrow AI (one task, everything that exists today), general AI (human-level across tasks, not yet built) and superintelligence (beyond human, hypothetical). By function: reactive machines, limited-memory systems, theory-of-mind AI and self-aware AI; only the first two exist. The split that helps day to day is by what the system does: rule-based, predictive, generative, and agentic. The chatbot you use is narrow, limited-memory, generative AI, increasingly with agent features bolted on.

By capability: the three levels

  • Narrow AI (weak AI). Built for one kind of task: translate, detect fraud, drive on a highway, write text. Every AI that exists is narrow, including the chatbots, however broad they feel. A model that writes legal briefs and poetry is still "narrow" in this sense because it cannot learn a new skill the way a person does; it is wide, not general.
  • General AI (AGI). Human-level ability across most intellectual tasks, including learning new ones from little data. Does not exist. The AGI page covers why nobody agrees how close it is.
  • Superintelligence. Beyond the best humans at essentially everything. Hypothetical; the subject of the long-term safety debate.

By function: the four types

This list comes from a 2016 essay by the researcher Arend Hintze and is quoted everywhere, so it is worth knowing which entries are real.

  • Reactive machines. Respond to the present input with no memory of the past. The classic example is the chess computer that beat Kasparov in 1997: it evaluated the board in front of it and nothing else. Real, and still common in simple systems.
  • Limited memory. Uses recent past data to decide: a self-driving car tracking the cars around it, a chatbot reading the conversation so far. Real; nearly all modern AI, including every language model, is this.
  • Theory of mind. Would understand that other beings have beliefs, intentions and feelings, and act on that understanding. Not built. Chatbots can talk about emotions fluently without this; talking about something and modelling it are different.
  • Self-aware. Would have consciousness and a sense of itself. Not built, and there is no agreed way to tell if it ever were.

By what it does: the split that helps

  • Rule-based. A person wrote the logic. Tax software, the thermostat, early chatbots that matched keywords. Reliable, rigid, and not what people mean by AI any more, though products still call it that.
  • Predictive (classic machine learning). Learned from data to classify or forecast: spam or not, fraud or not, what you will buy, when the machine will fail. Runs most of the economy quietly. The machine learning page explains how.
  • Generative. Learned from data to produce new text, images, video, audio or code. This is the 2022 shift and what "AI" means in conversation now. The generative AI page is the long version.
  • Agentic. Generative models given tools and a loop, so they can carry out multi-step tasks rather than answer one message. The 2025–26 shift. See the AI agents page.

Other ways you will see AI sorted

By medium: language models, vision models, speech models, multimodal models that do several. By method: symbolic AI (rules and logic, the old school) versus machine learning (statistics from data, the new school), with hybrids. By where it runs: cloud AI on a company's servers versus on-device AI on your phone or laptop, which matters for privacy because nothing leaves the device. None of these contradict the lists above; they are just different axes.

So which type are you talking to?

When you type into a chatbot: narrow (capability), limited-memory (function), generative (what it does), usually cloud-hosted (where it runs), and increasingly agentic when it searches or runs code for you. That combination explains both its strengths (broad, fluent, fast) and its limits (no real memory beyond the window, no understanding of you as a person, and, in most products, a server that keeps what you typed). The last point is the one you can choose: nixm is the same type of AI with a different rule about data, keeping no transcript and training on nothing you say.

frequently asked

What are the 3 types of AI?

By capability: narrow AI, which does one kind of task and is everything that exists today; general AI, human-level across tasks, not yet built; and superintelligence, beyond human, hypothetical.

What are the 4 types of AI?

By function: reactive machines with no memory, limited-memory systems that use recent data, theory-of-mind AI that would model other minds, and self-aware AI. Only the first two exist.

What type of AI is ChatGPT?

Narrow by capability, limited-memory by function, generative by what it does, and increasingly agentic when it uses tools like search or code. It is not general AI.

What is the difference between narrow AI and general AI?

Narrow AI performs tasks it was built or trained for, however many. General AI would learn new tasks and domains the way a person does, from little data. Narrow exists; general does not.

Does self-aware AI exist?

No. No current system has consciousness or a sense of self, and there is no agreed test that could confirm it if one did.

What is the most common type of AI today?

Predictive machine learning runs most systems quietly, from spam filters to fraud checks. Generative AI, the chatbots and image tools, is the most visible and the fastest growing.

try it now — no account, no transcript kept. even the signal dissolves in seven days.

seed a signal →

Enter your email and receive a link to sign in. No passwords, just a secure signal to connect.

why connect

  • 20 free credits to start
  • your signals, remembered
  • buy more credits anytime
  • image & video generation features
or
+