7 pillars for understanding AI
Machine learning, language, vision, robotics, speech, expert systems and planning: explore the field beyond chat.
These seven areas are a Noosigma AI learning map, not an official or exhaustive division of computer science. They overlap: a robot can use vision, learning and planning at the same time. Curricula such as ACM/IEEE/AAAI CS2023 organize AI into more knowledge units.
Text reviewed on
Machine learning
Models adjust parameters from data or feedback to recognize patterns, predict outcomes or generate content. Development still requires human choices about data, goals and evaluation. Deep learning is a family of methods based on multilayer neural networks.
Concrete example. A filter can learn to classify messages as spam from labeled examples.
Do not confuse. Learning from data does not mean learning alone without design, supervision or tests. An ordinary chat does not retrain the model.
Natural language processing (NLP)
Studies methods for working with human language in tasks such as classification, search, translation, information extraction and text generation. LLMs are one family of models used for some of these tasks, not the whole field.
Concrete example. A system can identify questions in a message and retrieve related documents before answering.
Do not confuse. Fluent text does not guarantee understanding or factual accuracy.
Computer vision
Analyzes images and video to locate, classify, measure or track elements. It can combine geometry, pattern recognition and machine learning.
Concrete example. A system can locate road signs or flag findings in medical images for professional review.
Do not confuse. Finding a visual pattern is not the same as understanding the whole scene or producing a reliable diagnosis without validation.
Robotics
Combines sensors, perception, planning, control and actuators to act in the physical world. AI can help a robot adapt its actions when the environment changes.
Concrete example. A warehouse robot can recalculate its route when a passage is blocked.
Do not confuse. Not every robot uses machine learning; fixed repetitive movements are also automation.
Speech recognition and synthesis
Speech recognition turns spoken audio into text. Speech synthesis goes the other way, from text to audio. These audio tasks can be combined with NLP but have their own challenges involving signal, noise, accents and voice.
Concrete example. An assistant can transcribe a request, interpret it and then speak its response.
Do not confuse. Recognizing words, identifying the speaker and generating a voice are different tasks.
Expert systems
Represent domain knowledge and use rules or inference mechanisms to support conclusions. They may be small or large and can also combine with methods learned from data.
Concrete example. A support system can relate fault symptoms to documented rules and suggest checks for a technician.
Do not confuse. Explicit rules do not replace expert review or guarantee correct results outside the intended domain.
Planning, scheduling and optimization
Planning selects action sequences to meet goals; scheduling allocates tasks over time and resources; optimization compares alternatives under objectives and constraints. Search, heuristics and evolutionary algorithms are among the possible techniques.
Concrete example. An operation can assign deliveries to vehicles and choose routes while respecting time windows and capacity.
Do not confuse. A computed solution depends on the data and constraints; it is not necessarily the best possible and does not require predicting the future.