\u201eArtificial Intelligence\u201c and \u201eMachine Learning\u201c are often mistaken for the same thing in everyday life. In job postings, in the media, and on product pages, they usually appear side by side, as if they were interchangeable. They are not. If you\u2019re planning to pursue further education or want to understand what everyone is talking about right now, making a clear distinction between them will take you much further. This article breaks down the three terms that are constantly confused: AI, machine learning, and deep learning.<\/p>\n\n\n\n
Artificial intelligence is the umbrella term for all methods that enable computers to perform tasks that would otherwise require human intelligence. Machine learning is a subset of this: the method by which a system learns from data rather than following fixed rules. In short: All machine learning is AI, but not all AI is machine learning.<\/p>\n\n\n\n
Artificial intelligence<\/a> describes a goal, not a specific technique. What is meant is that a machine performs tasks that would otherwise require human thought: understanding language, interpreting images, planning, and making decisions.<\/p>\n\n\n\n It\u2019s important to note that there are different ways to achieve this. An AI can be based on rigid if-then rules that a human has written by hand. A simple chess program or a rule-based system that sorts applications according to fixed criteria is already AI, without any learning involved. Such systems are easy to understand, but they will never perform better than their rules allow.<\/p>\n\n\n\n One point clears up a lot of misunderstandings: All systems currently in use<\/a> are what are known as \u201enarrow AI.\u201c They are specialized for a narrowly defined task. A translation system translates; a recommendation system makes recommendations. There is no AI with general common sense that can freely switch between tasks. When the media refers to \u201eAI,\u201c they almost always mean a specialized system.<\/p>\n\n\n\n Machine learning is the branch of AI that has seen the most progress in recent years. Instead of being given rules, a model learns from examples. You show it historical data, it recognizes patterns in it, and applies them to new cases.<\/p>\n\n\n\n The difference from rule-based AI is fundamental. A rule-based system can only do what someone has programmed it to do. A learning system discovers connections that no one had thought of, as long as there is enough data available. However, its result is never a fixed truth, but rather a probability.<\/p>\n\n\n\n An example makes the difference clear. A rule-based spam filter blocks emails containing specific, predefined words. A machine-learning spam filter receives tens of thousands of flagged emails and deduces for itself what constitutes spam. It adapts when senders change their tactics. The rule-based filter does not.<\/p>\n\n\n\n An important practical consideration: Traditional machine learning requires people to prepare the data and identify the right features. This preparation\u2014known as feature engineering\u2014is often the most time-consuming part, and that is precisely where data experts come in. How this Step-by-Step Data-Driven Learning Works<\/a>, as shown in the detailed overview.<\/p>\n\n\n\n A third term crops up in almost every text on the subject: deep learning. It is not an alternative to machine learning, but rather a special case of it.<\/p>\n\n\n\n Deep learning works with artificial neural networks<\/a>, which consist of many layers. That\u2019s why they\u2019re called \u201edeep.\u201c The practical advantage: These networks can process unstructured raw data\u2014such as images, speech, or audio\u2014directly and identify the important features on their own, without a human having to label them beforehand.<\/p>\n\n\n\n This is what makes deep learning so effective at tasks that are too challenging for traditional machine learning: Facial Recognition, Language Translation, and Speech Recognition<\/a> These are typical everyday examples. The cost is high. Deep learning requires very large amounts of data\u2014often millions of examples\u2014and a great deal of computing power. And there\u2019s a catch that matters especially in critical areas: The decisions made by deep neural networks are virtually impossible for humans to understand. This is referred to as the \u201eblack box\u201c\u2014a problem that Support for research institutes such as Fraunhofer IKS<\/a> work.<\/p>\n\n\n\n The easiest way to remember these three terms is to think of them as circles nested inside one another. Deep learning is a subset of machine learning, and machine learning is a subset of artificial intelligence.<\/p>\n\n\n\nMachine Learning: The Learning Method<\/h2>\n\n\n\n
Deep Learning: The Deep Layer<\/h2>\n\n\n\n
A Comparison of AI, Machine Learning, and Deep Learning<\/h2>\n\n\n\n