The main types of AI are grouped in three ways: by capability (from narrow AI to AI that would exceed human intelligence), by functionality (how a system processes information and whether it remembers) and by the technology it is built on, such as machine learning or generative AI. Every AI system you can use today, ChatGPT included, is narrow AI.
What you’ll take from this
By the end you will be able to place any AI tool on the map and know what to expect from it.
- Tell apart the types of AI that exist today from the ones that are still hypothetical.
- Understand which type of AI sits behind a chatbot, a spam filter or a self-driving car.
- See which AI technologies companies actually use, with 2025 data.
- Choose the right type of AI for a specific task at work.
What are the different types of AI? The main AI categories
Understanding the different types of AI starts with three classifications that overlap. The first looks at capability, the second at how the system works and the third at the technology it is built with. A single AI system can sit in all three, as ChatGPT does.
| Classification | AI types | Example | Exists today? |
|---|---|---|---|
| By capability | Narrow AI, general AI, super AI | A translation app | Only the narrow level |
| By functionality | Reactive machines, limited memory, theory of mind, self-aware | A virtual assistant uses limited memory | Reactive and limited memory |
| By technology | Machine learning, deep learning, generative AI, language processing, computer vision, rule-based systems | A spam filter learns from examples | All of them |
You do not need to code to choose and use these tools at work, but you do need to know which type of AI you are dealing with. That is where our certificate program in artificial intelligence starts.
Types of AI by capability: narrow AI, general AI and super AI
By capability, there are three types of AI. Only the first level exists today. It is the most common classification because it measures how close machines come to human intelligence: how many different problems they can solve and how independently.
Narrow AI or weak AI (ANI)
Narrow AI, also called weak AI or artificial narrow intelligence, is built to perform a specific task: translate text, recommend a show, recognize a face or draft an email. It can do that task faster and often better than a person, but it cannot step outside it. It is the only kind that exists today, even though some assistants cover so many tasks that they feel general.
Everyday examples include recommendation systems on streaming platforms, facial recognition systems on your phone, fraud detection at your bank and the assistants you use at work. Each one is excellent at its own job and useless at anything else.
General AI or strong AI (AGI)
Artificial general intelligence, also called strong AI, would be a system with human-level intelligence that can learn and perform any intellectual task a human can, applying knowledge across a wide range of problems it has never seen. The major AI labs describe it as their goal, but no current system is generally recognized as AGI. For now it remains a research target, not a product.
Super AI or artificial superintelligence (ASI)
Super AI, or superintelligent AI, would be a hypothetical AI that surpasses human intelligence in every area, from science to creativity to social skills. It drives much of the debate about AI safety and the future of AI. If you want to explore that debate, read our article on artificial superintelligence and whether it is a threat.
Types of AI based on functionality: the 4 types
By functionality, AI is divided into four types. This classification looks at how a system processes information and whether it uses past experience to make decisions.
- Reactive machines (reactive AI). They respond to what is in front of them with no memory of the past. The classic example is Deep Blue, the computer that beat Garry Kasparov at chess in 1997.
- Limited memory AI: it learns from historical data and recent inputs. Conversational AI (chatbots and virtual assistants) and self-driving cars all work this way.
- Theory of mind AI: it would understand people’s emotions, intentions and beliefs. Today it is an area of AI research.
- Self-aware AI: it would be conscious of itself. It is a theoretical AI with no real system behind it and, for now, it belongs to science fiction.
Almost everything you use at work, from the filter that catches spam to the assistant that summarizes your meetings, falls into this second group.
Types of AI models and technologies behind today’s tools
By technology, AI is grouped by the technique a system is built on, and many different types can work together in one product. This is the most useful classification in practice, because it tells you what each system can do and what data it runs on.
Machine learning
Machine learning is the field of AI in which a system uses learning models to find patterns from data instead of following rules written by hand. It comes in three main forms. Supervised learning learns from labeled examples, unsupervised learning finds groups in unlabeled data and reinforcement learning learns by trial and error with rewards. A learning model that flags fraudulent transactions or forecasts demand for a product works this way.
Deep learning and neural networks
Deep learning is a subset of this field that uses artificial neural networks with many layers, loosely inspired by the human brain. It powers speech recognition, machine translation and today’s chatbots. For a closer look at how it works, read our guide to deep learning and artificial intelligence.
Generative AI vs predictive AI
Generative AI creates new content: text, images, video, audio or code. Copilot, Claude and Gemini generate text; Midjourney and Adobe Firefly generate images; ElevenLabs generates voice. Behind the text assistants are large language models (LLMs), AI models trained on huge amounts of text to predict the next word.
Predictive AI does something different: it uses machine learning models trained on past data to forecast a number or a category, such as next month’s sales, the chance a customer will cancel or whether a loan is risky. Many companies get more value from it than from gen AI, because it plugs straight into decisions they already make.
Natural language processing and computer vision
Natural language processing lets machines understand and produce human language: sorting emails, analyzing customer reviews or answering questions in a support chat. Computer vision does the same with images and video, from reading license plates to inspecting parts on a production line or spotting objects on the road.
Two common techniques are image segmentation, which labels every pixel of an image (road, car, pedestrian), and anomaly detection, which flags a part or a payment that does not look like the rest.
Expert systems and AI agents
Expert systems are the oldest form of AI: rule-based AI that makes decisions with “if this, then that” rules written by specialists. An AI agent is the newest: it combines a language model with tools to complete multi-step tasks with some autonomy. Our guide to AI agents covers how they work and the different kinds.
How are the different AI systems trained?
Most AI systems in use today learn from data rather than from rules written by hand. The data an AI uses decides what it can and cannot do, so it helps to know the three ways it is used.
- Labeled data. In supervised learning, people tag each example (this email is spam, this one is not) and the AI model learns to repeat that decision on new cases.
- Self-supervised learning. LLMs learn by hiding part of a text and predicting the missing words, so they can train on huge amounts of text nobody has labeled. This is how the models behind ChatGPT or Copilot are pre-trained before they are adjusted with human feedback.
- Training, validation and test data sets. Teams split their data into three parts: one to train the model, one to tune it and one it has never seen, to check that it works on new cases instead of memorizing the examples.
When an AI tool gives a confident wrong answer, the cause is often here: the data it learned from did not cover your case. That is why decision-making with AI still needs a person who checks the result.
Examples of each type of AI in everyday life and at work
The easiest way to understand the different types of artificial intelligence is to see where you already use them. This table matches each type to an everyday example and a work example:
| Type of AI | Everyday example | Work example |
|---|---|---|
| Reactive machine | A chess program that picks the best next move | A rules engine that routes support tickets by keyword |
| Limited memory | A navigation app that reroutes you around traffic | A meeting assistant that summarizes the call you just had |
| Machine learning | A spam filter in your inbox | A model that scores which leads are most likely to buy |
| Computer vision | Face recognition that opens your phone | Quality checks on a production line |
| Language processing | A voice assistant that sets a timer | Sorting customer emails by topic and urgency |
| Generative AI | Creating an image for a birthday card | Drafting a proposal or a first version of a report |
Notice what these AI applications have in common: none of these examples needs advanced AI beyond a single-purpose system, and none of them thinks like a human. That is the reality of modern AI: very capable systems, each built for a defined job, that you combine to get work done.
What the different types of AI mean for your job
For most professionals, the types of AI that matter day to day are generative AI and the AI agents built on it. They are the ones that change how you write, analyze and plan, and they are the ones you can start using today without technical skills.
Forecasting tools and computer vision usually arrive through software your company buys: a CRM that scores leads, a finance tool that flags unusual payments, a warehouse system that counts stock from camera images. You rarely build these yourself, but you need to understand what they do well enough to trust or question their output.
The skill that carries across every type is judgment: knowing what the AI system was trained to do, where it is likely to fail and when a human needs to check the result. That matters even more with the newest AI capabilities, where agents act on your behalf across several tools.
Teams that treat AI technologies as a set of different tools, each with a clear job, get far more out of them than teams that expect one assistant to do everything.
Which types of AI do companies use most?
Companies mostly use AI to analyze and generate language. According to Eurostat, 20.0% of EU enterprises with 10 or more employees used AI technologies in 2025, up from 13.5% in 2024. These were the most common:
| AI technology | Share of EU enterprises using it (2025) |
|---|---|
| Analyzing written language (text mining) | 11.8% |
| Generating pictures, video or sound | 9.5% |
| Generating written or spoken language | 8.8% |
| Speech recognition | 7.2% |
Individuals are moving faster than companies. According to the same source, 32.7% of people aged 16 to 74 in the EU used generative AI tools in 2025. Many people already use types of AI at home that their employer has not adopted yet.
How does EU regulation classify AI?
The EU AI Act does not classify AI by technology but by risk, and sets different obligations for each level. If your company sells to or operates in the EU, this is the classification that decides your legal duties. According to the European Commission, there are four levels:
| Risk level | What it means | Example |
|---|---|---|
| Unacceptable | Banned | Social scoring |
| High | Strict requirements before use | CV-sorting software for recruitment |
| Transparency | People must be told they are dealing with AI or AI-generated content | Chatbots and deepfakes |
| Minimal | No specific obligations | Spam filters and video games |
The regulation also has separate rules for general-purpose AI models, the kind behind ChatGPT or Copilot, covering transparency and copyright. Knowing this classification matters as much as knowing the technology if your team uses AI at work.
What type of AI is ChatGPT?
ChatGPT is narrow AI by capability, limited memory AI by functionality and generative AI by technology: a large language model built on neural networks. The same applies to Copilot, Claude and Gemini. They are highly capable, but they are not artificial general intelligence.
How to choose the right type of artificial intelligence for each use case
Choosing the right AI starts with the problem you need to solve, not with the tool everyone is talking about. This table covers the most common use case for each type in an office:
| If you need to… | Type of AI | Example tools |
|---|---|---|
| Write, summarize or translate text | Text generation | ChatGPT, Copilot, Claude |
| Create images or video for a campaign | Image and video generation | Midjourney, Adobe Firefly, Runway |
| Automate repetitive tasks across apps | Automation with AI agents | Power Automate, Copilot Studio |
| Forecast sales or spot anomalies in data | Predictive AI | Data analysis models |
A practical rule: if the output is new content, you want generative AI; if the output is a number or a decision, you want a forecasting model. Most teams end up using both, often in the same workflow.
Frequently asked questions about AI types
These are the most common questions about how AI is classified and which AI types exist today.
What are the 4 types of AI?
When people talk about four types of AI, they usually mean the classification by functionality: reactive machines, limited memory AI, theory of mind AI and self-aware AI. Only the first two exist today; the other two are research goals or hypotheses.
What are the 7 types of AI?
The seven types of AI come from combining the two most quoted classifications: the three levels of capability (narrow, general and super) and the four types by functionality (reactive, limited memory, theory of mind and self-aware). They are two ways of looking at the same technology, so every system fits one category in each list.
Which AI type is still hypothetical?
General and super intelligence, theory of mind AI and self-aware AI are all still hypothetical. Every AI system in use today is narrow AI with either reactive or limited memory functionality, however advanced it looks.
What is the difference between AI, machine learning and deep learning?
They are nested circles. Artificial intelligence is the whole field, machine learning is the part where systems learn from data and deep learning is a branch of it based on neural networks with many layers. The third sits inside the second, but not the other way around.
Further reading
These are the official sources we used for this guide.
- AI regulatory framework, European Commission
- Use of artificial intelligence in enterprises, Eurostat
- Artificial intelligence, Wikipedia
If you want to move from knowing the types of AI to using them in your work, the Certificate Program in Artificial Intelligence and Innovation shows you how to apply AI assistants, agents and automation to real tasks, with responsible use built in. It was developed in collaboration with Microsoft and ends with the Microsoft + Founderz Certification.
