There are three main categories of AI: narrow AI, artificial general intelligence (AGI) and artificial super intelligence (ASI). Only narrow AI currently exists. All the AI tools you use as a professional, from ChatGPT to image recognition, fall under narrow AI.
There are different types within narrow AI, each of which learn and work in its own way. In this overview, we explain what types of AI there are, how they differ from each other and which types are relevant to your work.
What types of AI currently exist?
AI is classified based on how broadly the system can think and act. The three main categories are narrow AI (limited to specific tasks), general AI (human level) and super AI (above human level).
Narrow AI is the only category that exists now. General AI and super AI are theoretical concepts that researchers are working on, but have not yet been realised.
For your day-to-day work, narrow AI is the only category that matters. All the examples and applications in this article fall under it.
What is narrow AI and what do you use it for?
Narrow AI (also called ‘weak AI’) is designed to perform one specific task or a limited set of tasks. The system is very good at that one task, but cannot do anything beyond that.
Examples you probably already use every day: voice assistants like Siri and Google Assistant, recommendation systems from Netflix and Spotify, and spam filters in your e-mail. All these systems are narrow AI.
Also, tools such as ChatGPT, Claude and Gemini fall under narrow AI. They seem broadly applicable because they can handle text, code and analysis, but under the bonnet they are statistical systems that apply patterns from their training data.
How do different types of AI learn?
Not all AI learns in the same way. The learning method determines what an AI system is suitable for.
In supervised learning, the model trains on labelled data: each input is linked to the desired output. This works well for tasks where you have clear examples, such as recognising spam emails or classifying documents.
In unsupervised learning, the model receives data without labels and has to discover patterns on its own. This is used for grouping customers based on behaviour or detecting anomalies in datasets.
Self-supervised learning is in between. The model learns from the structure in the data itself, without people having to manually add labels. Large language models like ChatGPT and Claude are trained this way: they learn by predicting which word follows in billions of texts.
What is the difference between reactive AI and limited memory AI?
Besides classification by learning method, you can also classify AI by memory and adaptability. This is another way of looking at the same systems.
Reactive AI responds to input without remembering anything or learning from previous interactions. A chess computer that recalculates every move without remembering previous games is reactive AI. This type is reliable but inflexible.
Limited memory AI stores and uses information to make better decisions. This is the type of AI you most commonly encounter in professional applications: fraud detection systems that learn from historical patterns, inventory planning that incorporates seasonal data, or chatbots that remember context within a conversation.
Understand how AI works technically helps you judge when you can trust the output and when not.
What is generative AI and how is it different from other types of AI?
Generative AI is a specific application of narrow AI that can create new content: text, images, code, audio and video. The difference with other types of AI is that generative AI does not just recognise patterns or make predictions, but produces original output.
Whereas predictive AI estimates which customers are likely to cancel, generative AI can write an entire email campaign to retain those customers. Both are useful, but for different purposes.
For professionals, generative AI has become the most visible type of AI. Want to dive deeper into this topic? Then read our article on what generative AI is and how it differs from other forms of AI.
What are AGI and ASI?
Artificial General Intelligence (AGI) is a hypothetical type of AI that can perform any intellectual task that a human can. It could independently learn, reason and adapt to completely new situations without specific training.
AGI does not exist. Researchers are working on it, but mimicking human understanding, emotion and sense of context remains a huge challenge. There is no scientific consensus on when or even if AGI will be realised.
Artificial Super Intelligence (ASI) goes one step further: AI that surpasses human intelligence in all areas. This is purely theoretical and not something you need to consider as a professional right now. You can read more about the ethical and social questions arising from this in our article on Whether AI is dangerous.
What types of AI are relevant to your work?
For everyday professional use, three applications of narrow AI are most relevant.
Supervised learning AI helps with tasks with distinct patterns, such as document classification and sentiment analysis. Limited memory AI is suitable for systems that learn from new input, such as chatbots and fraud detection. Generative AI you use for content creation and analysis: writing texts, summarising data and brainstorming.
In practice, the name of the AI type is less important than the question: does this tool solve a concrete problem in my work? Start with one clear application and build from there. Want to know What you can do concretely with ChatGPT? That gives you a good starting point.
Want to learn how to use AI tools effectively for your own work? In the ChatGPT course from LearnLLM build a working method that you can apply immediately, including checkpoints to verify output.
