“AI” is used for everything from a spam filter to a chatbot to a self-driving car, which makes it hard to know what people mean. In practice, AI is a family of fields that overlap. This guide explains the main branches of AI, what each does, where you meet them every day and how they combine in real products.
Key takeaways
- AI is the broad goal; machine learning is the main method; deep learning powers most modern advances; generative AI is the newest layer.
- Other branches include natural language processing, computer vision, speech, robotics and expert systems.
- Most real products combine several branches.
- Today’s AI is “narrow”: very good at specific tasks, not general human-like intelligence.
The main branches
| Branch | What it does | Everyday example |
|---|---|---|
| Machine learning | Learns patterns from data to predict or classify | Spam filters, product recommendations, credit scoring |
| Deep learning | Uses large neural networks for images, speech and language | Face unlock, voice assistants, translation |
| Natural language processing (NLP) | Understands and produces human language | Chatbots, search, email classification, summarisation |
| Computer vision | Interprets images and video | Quality inspection, medical imaging, number-plate reading |
| Speech recognition and synthesis | Turns speech to text and back | Voice typing, call-centre transcription |
| Robotics | Perceives and acts in the physical world | Warehouse robots, cobots; see cobots and the future of work |
| Expert systems and rules | Encode expert knowledge as rules | Tax software, diagnostic checklists |
| Generative AI | Creates new text, images, code and audio | ChatGPT, image generators, coding assistants |
Machine learning in one minute
Instead of writing rules by hand, you give an algorithm examples and let it find patterns. There are three broad styles: supervised learning (learn from labelled examples, such as emails marked spam), unsupervised learning (find groups or structure in unlabelled data) and reinforcement learning (learn by trial and error with rewards). Predicting next month’s sales is machine learning; so is grouping customers by behaviour.
Generative AI and large language models
Generative AI models learn the patterns of text, images or audio so well that they can produce new examples. Large language models (LLMs) are the best-known type. For how these terms differ, read LLMs vs generative AI and generative AI vs predictive AI. For safe use at work, see dos and don’ts of generative AI.
Types of AI by capability
- Narrow AI does one kind of task well. Every AI product in use today is narrow, even when it looks general.
- General AI (AGI) would match human ability across many tasks. Researchers disagree on when, or whether, it will exist.
- Superintelligence is a hypothetical level beyond human ability.
How branches combine in real products
| Product | Branches involved |
|---|---|
| Self-driving car | Computer vision, deep learning, planning, robotics |
| Voice assistant | Speech recognition, NLP, generative AI, text to speech |
| Invoice automation | Computer vision (OCR), NLP, machine learning, RPA; see intelligent document automation |
| Customer support bot | NLP, generative AI, retrieval of company knowledge; see chatbot vs conversational AI |
| Predictive maintenance | Machine learning on sensor data, rules, alerts |
Where AI meets automation
AI is increasingly the “brain” that reads, decides and plans, while automation tools carry out the steps. See RPA vs intelligent automation and cognitive RPA for how that works in business processes.
Frequently asked questions
What is the difference between AI, machine learning and deep learning?
AI is the broad field. Machine learning is a method within AI where systems learn from data. Deep learning is a type of machine learning using large neural networks.
Is ChatGPT machine learning?
Yes. It is a large language model, which is a deep learning model, trained with machine learning methods.
Which branch of AI is most used in business?
Machine learning for prediction and classification is the most widespread, with natural language processing and generative AI growing quickly.
