Updated 2026-11-18 · Research-based guide · Includes a practical decision worksheet
Lists of “AI types” often mix research fields, training methods, and descriptions of capability. That makes the subject harder to learn. Use several complementary maps instead: what the system does, how it learns, and what kind of claim is being made about its ability.
Map the task first
Language processing concerns work with language; computer vision concerns images or visual information; planning concerns selecting actions toward a goal. Robotics combines sensing and action in physical systems. These fields can overlap in one project. A warehouse robot, for example, may use visual perception and planning rather than belong to a single exclusive category.
Separate methods from application fields
In supervised learning, training examples include target labels or values. Unsupervised learning seeks patterns without that same target structure. Reinforcement learning involves learning behavior from interactions and rewards. These are broad learning setups, not a complete list of AI techniques. Rule-based search or reasoning also belongs in an AI map and should not be erased by a focus on modern learned models.
Place deep learning within the map
Deep learning uses neural networks with multiple processing layers. It is one family of machine-learning approaches, not a synonym for all AI. Generative systems produce outputs such as text or images, while predictive tasks may estimate a category or number. A system can be both deep-learning-based and generative; the labels answer different questions.
Use one example across the maps
Imagine a fictional service desk that sorts incoming requests and drafts a response. Request sorting is a classification task; response drafting is generation. The workflow may also contain ordinary rules that route urgent cases to a person. Calling the entire process “an AI agent” tells you less than explaining which steps use which methods and who can authorize an action.
Treat capability labels carefully
Terms such as narrow AI or general intelligence describe different kinds of capability claims and are not a practical procurement specification by themselves. Ask what a system can do under defined conditions, what evidence supports that claim, and where it fails. A strong result on one benchmark does not demonstrate reliability across every task described by an ambitious label.
Turn terminology into study questions
For each unfamiliar term, record the task, required data, output, evaluation, and limits. Russell and Norvig’s textbook provides a broad map beyond machine learning; Google’s crash course covers selected learning concepts. Use the AI comparison guide to distinguish neighboring fields and project planning to put one part of the map into practice.
Questions readers ask
Are machine learning and AI identical?
Machine learning is an important part of AI, but AI also includes approaches such as search, planning, and symbolic reasoning.
Is generative AI a learning method?
It describes a class of output-generating capabilities. To understand a particular system, also ask how it was trained and how its outputs are evaluated.
Put this guide into practice
Map a recommendation feature into data collection, prediction, ranking and evaluation. Distinguish the business task from the technique used. Several techniques may serve one task; one technique may appear in several applications.
| Check | Evidence to record |
|---|---|
| Task definition | Observation, source or test result; responsible person; follow-up date. |
| Technique choice | Observation, source or test result; responsible person; follow-up date. |
| Evaluation measure | Observation, source or test result; responsible person; follow-up date. |
This is a proposed exercise using illustrative circumstances, not a report of a test or visit performed by our publication.
Sources and further reading
Source pages checked November 16, 2026. The planning examples and checklists are editorial suggestions, not reported product tests or guaranteed outcomes.
Published by Sadddam Hussain. Research-led guidance for US readers. Read our editorial standards.
