A series of instructions to solve a certain problem (see also Knowing-it-all).
The individual commands must be unique and executed step by step. An algorithm usually requires an input and returns an output. Examples of algorithms are computer programs and electronic circuits, but also building instructions or cooking recipes. Certain algorithms are attributed to artificial intelligence.
Machines that emulate intelligent behaviour on the basis of algorithms.
An exact definition is difficult because the concept of intelligence itself is not clearly defined. As a result, AI includes a whole spectrum of methods, disciplines and applications: computer programs that can play chess or, alternatively, chatbots that chat with social network users. Certain subareas of robotics are just as much part of AI as expert systems which are to help make optimal decisions in a limited range. To an increasing degree, ethical, social or legal considerations are also part of AI. Machine learning is regarded as a key technology within the AI.
AI algorithms that learn from data and examples to solve problems.
They acquire “knowledge” by means of examples (training data) or through independently recognising patterns in data. This enables them to subsequently assess unknown data of a similar nature. When identifying faces, for example, an algorithm can learn that pupillary distance, face shape and nose size are crucial factors for recognition. For the “face” concept, the algorithm extracts a characteristic pattern from each image. The more data the algorithm has at its disposal, the more precise the recognition becomes. Considerable progress has been made in recent years with the help of artificial neural networks.
Artificial neural networks
Mathematical models inspired by the way the brain works.
Signals – the input data of the algorithm – are fed into interconnected units, referred to as “mathematical nerve cells”. The artificial nerve cells process the information and generate further signals using simple mathematical equations, which they pass on to subordinate “cells”. Finally, an output layer generates a result. There can be several layers of these nerve cells, which are linked to each other in different ways, between the input layer and the output layer.
Learning strengthens, weakens or changes the connections between individual cells. Advancements in computer technology and the availability of large amounts of data have enabled deep learning in such artificial networks.
Machine learning in neural networks with many layers, so-called “deep” networks.
Here, too, algorithms analyse large data sets and can then evaluate unknown data of a similar nature. However, the network models are much more complex due to the many layers. As a result, the algorithm has many degrees of freedom in which to network and, in order to solve a task, can independently learn to extract optimal and possibly very complex traits. When identifying faces, it can thus discover finer criteria than pupillary distance or nose size that are helpful for recognition. Programmers help the software to “learn” by giving feedback on whether a result is right or wrong, but they do not correct the process.
Graphics: Bernd Struckmeyer, ONYXprj/Shuttersctock.com