Artificial intelligence (AI) is the umbrella, not a synonym for machine learning. AI covers systems designed to perform capabilities associated with human intelligence—such as perception, learning, reasoning, planning, communication and physical action. Machine learning (ML) is one way to build such systems, and deep learning is a neural-network-based form of ML. Language, vision, speech and robotics describe capabilities or application domains that can use those methods.
There is no universally fixed checklist called “the components of AI.” NASA and NIST both note that definitions vary by source and purpose, while the European Commission’s taxonomy is intended to map an ecosystem that includes neighboring technologies. The map below is therefore a practical guide, not a single official standard.
AI, ML and deep learning: the nested relationship
Think of the relationship as nested circles:
- AI: the broad field of systems and techniques associated with intelligent capabilities.
- Machine learning: an AI approach in which a model learns patterns from data to make predictions or decisions instead of relying only on rules written by developers. Google Cloud describes ML as a type of AI.
- Deep learning: a subfield of ML that uses artificial neural networks with multiple layers. It is not a name for all AI. Google Cloud places deep learning inside ML.
An AI application can use ML, deep learning, hand-coded rules, or several of these together. Calling a product “AI” does not tell you which method it uses.
What counts as a component?
The labels below mix two different kinds of description. Approaches explain how a system is built (for example, ML or an expert system). Capabilities and domains explain what it handles (for example, language, images or physical movement). They are complementary, not competing boxes. The European Commission’s AI Watch taxonomy explicitly maps AI and adjacent domains rather than claiming one exhaustive list.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →#1 Best Overall
| Label | What it describes | Typical inputs or outputs | Relationship to AI |
|---|---|---|---|
| Artificial intelligence | Broad field and set of goals | Perception, decisions, language, plans or actions | Umbrella term |
| Machine learning | Learning approach | Data used to produce predictions or decisions | Subset of AI |
| Deep learning | Multilayer neural-network approach | Often text, audio, images or other high-dimensional data | Subset of ML |
| Natural language processing | Language domain | Text and language-related tasks | AI domain that may use ML or rules |
| Computer vision and perception | Perception domain | Images, video and other sensor data | AI capability or application area |
| Speech and dialogue | Spoken interaction | Audio, transcriptions and spoken responses | Language-related capability |
| Reasoning, decisions and planning | Problem-solving capabilities | States, constraints, goals and actions | AI capabilities |
| Robotics and physical action | Embodied application domain | Sensor data, controls and movement | AI-adjacent or AI application domain |
| Knowledge-based or expert systems | Encoded-knowledge approach | Rules, facts and an inference procedure | Historical and continuing AI approach |
Core AI capabilities and domains
Learning from data
In ML, training data is used to estimate patterns that can later support classification, prediction, recommendation or control. The learned model is only one part of a deployed system: data collection, evaluation, software rules and human oversight may surround it.
Perception and computer vision
Perception converts signals from cameras, microphones or other sensors into useful representations. Computer vision focuses on visual information such as images and video. The International Telecommunication Union lists vision and perception among AI disciplines and methods; vision systems may identify objects, segment scenes or detect events, but the exact task depends on the application.
Language, speech and dialogue
Natural language processing (NLP) enables computers to process human language, including tasks such as understanding, translation, retrieval and generation. Speech systems add spoken input or output, while dialogue systems manage an interaction over multiple turns. NLP is a domain that can be implemented with ML and deep learning; it is not a separate level beside them. ISO’s overview discusses language processing, deep learning, computer vision and robotics together as parts of the wider AI landscape: ISO, “What is artificial intelligence (AI)?”
Rank #2
Reasoning and problem solving
Reasoning systems use representations, rules, probabilities or learned models to draw conclusions. Problem solving chooses a way to reach a goal, often by searching alternatives or satisfying constraints. These capabilities can be combined with perception and language: an assistant may interpret a request, reason over available information and select an action.
Decision-making and planning
Decision-making selects an action under stated objectives, information and constraints. Planning lays out a sequence of actions, sometimes while conditions change. ITU identifies decisions, planning and problem solving among AI capabilities, and NASA’s explanation includes cognition and planning in its account of human-like tasks: NASA, “What is Artificial Intelligence?”
Robotics and physical action
Robotics connects AI software to sensors, actuators and a physical environment. A robot may combine vision, localization, planning and control; none of those components alone defines the whole robot. NASA includes physical action in its description of AI-related tasks, while the European Commission treats robotics as a neighboring technological domain useful for understanding the AI ecosystem.
Knowledge-based and expert systems
Earlier AI systems commonly represented specialist knowledge as explicit facts and rules, then applied an inference process to reach conclusions. These expert systems do not necessarily learn from data. The Australian Government’s National AI Centre describes expert systems alongside later developments such as neural networks, computer vision and NLP: “Artificial intelligence explained.” Rule-based components remain useful when policies must be explicit, auditable or stable, and they can coexist with learned models.
Why different sources give different component lists
Definitions answer different questions. NASA presents AI in terms of systems solving tasks that may require human-like perception, cognition, planning, learning, communication or physical action. NIST’s glossary collects multiple source-specific definitions, including formulations centered on human-like tasks and techniques that approximate cognitive work: NIST, “artificial intelligence.” A standards glossary, a policy taxonomy and a classroom explanation therefore emphasize different boundaries.
For practical reading, ask what a label is doing:
- Is it a field or goal? AI is the broad category.
- Is it a method? ML, deep learning, symbolic rules and search describe ways to build systems.
- Is it a capability or domain? Language, vision, speech, planning and robotics describe what the system perceives, produces or controls.
- Is it a deployed system? A product combines models, data pipelines, interfaces, infrastructure, safeguards and people.
How the pieces combine in one system
Consider a warehouse robot that moves an item after a spoken request. Speech recognition converts audio to text; NLP interprets the request; a knowledge store identifies the item; vision and other sensors locate it; planning selects a route; control software drives the motors; monitoring and human procedures handle exceptions. ML or deep learning may power several stages, while explicit rules enforce safety constraints. Calling the whole product “AI” is accurate at the system level but does not identify any single algorithm.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Common category mistakes
“AI and ML are the same thing”
They are not. ML is one approach within AI, so an AI system can include ML without every AI system being an ML system.
“Deep learning is a separate alternative to AI”
Deep learning is nested inside ML, which is nested inside AI. It is a particular neural-network approach, not a competing umbrella.
“NLP, vision and robotics are interchangeable types”
They usually identify domains or capabilities, whereas ML and deep learning identify methods. One application can contain all of them.
Best Value
“Every AI system learns from data”
Rule-based and knowledge-based systems can use explicitly encoded knowledge. Modern products often mix learned and hand-authored components.
A practical way to describe an AI system
- Name the capability: for example, classify images, answer questions, forecast demand or plan a route.
- Identify the inputs and outputs: text, speech, images, sensor readings, decisions or physical commands.
- Identify the method: learned model, deep neural network, rules, search, optimization or a combination.
- Place it in the larger system: include data preparation, interfaces, monitoring, safeguards and human review.
- State the boundary: explain what the system does not perceive, decide or control, and which conditions may reduce its reliability.
This description is more informative than labeling a product “AI-powered” without saying what it does.
Quick Recap
Key takeaways
- AI is broader than ML; ML is an approach within AI.
- Deep learning is a multilayer-neural-network subfield of ML.
- Language, vision, speech, reasoning, planning and robotics describe capabilities or domains that can be implemented with several methods.
- “Components of AI” is an explanatory framework, not a universally settled taxonomy.
- Real systems combine methods, capabilities, software and human processes rather than fitting one exclusive category.
