Artificial intelligence (AI) is a mixture of mathematics/statistics and information science/data processing. The term is often used to describe computer systems that emulate the cognitive abilities of human intelligence such as perception, communication, learning, planning, acting, or creation. A frequently mentioned sub-area of artificial intelligence is also machine learning in which statistical techniques enabling machines to improve their ability to perform tasks as they gain experience.
Basically, AI consists of algorithms. This is a kind of step-by-step guide that helps to achieve an objective or to solve a task. This enables a computer system to take on human-like tasks and independently find solutions if processes unexpectedly change. However, the algorithms are always designed for certain narrowly-defined tasks – and they need to be trained in advance using lots of data. For instance, an AI system which has been trained in dog pictures is able to detect them in illustrations at a later date once it has seen enough variations of test pictures.
After comprehensive training, algorithms are adept at recognizing patterns and laws. However, they do not understand deeper associations. Not only that; if they are not accordingly programmed and sufficiently trained, they are unable to identify any completely unknown object. And it is no certainty that the algorithm will correctly recognize every object. Artificial intelligence can still get it wrong, even when looking at objects that are easy for humans to recognize and, for example, confuse a blueberry muffin with the head of a chihuahua.
How does AI learn?
One sub-area of machine learning is deep learning. The basis of deep learning is the deep neural network.
This type of AI imitates/emulates the human brain where neurons are linked with each other on many layers and, as a result, can process and save information. If they are trained, they can develop a kind of "experience."
Algorithms for machine learning have existed for a very long time. But the possibility of applying complex mathematical calculations to enormous amounts of data – again and again and faster and faster – is relatively new. And it also owes its success to new and, above all, more powerful computer technology.
Today, algorithms are excellent at pattern recognition. Excellent computer performance means they are able to learn quickly, process lots of information, and instantly recognize patterns in enormous amounts of data. In these clearly defined areas, they already work many times faster than the human brain.
The next stage of this development will be reached over the next few years with the introduction of quantum computing. As a result, calculations will be made even faster so that completely new areas of application will be opened up to AI (keyword: quantum AI).
AI revolutionizing almost all aspects of life
As a key technology, it is currently used in various areas of life. Many people use it every single day in assistance systems but don’t realize it. For instance, AI is used in navigation systems or internet searches. Just as it is in purchase recommendations or in the app that tells us which bus or train to take.
It is also used by banks, insurance companies, as well as in the field of medicine, agriculture, and many other fields with the objective of helping people.
Deutsche Telekom is also using artificial intelligence in many areas. From customer service and the planning of fiber-optic rollout through to an app indicating the best possible train or bus connection and services for corporate customers.
Weak and strong AI: What’s the difference?
The AI that we currently encounter is “weak AI.” It is programmed for certain tasks and performs them (rule-based). It is often able to improve itself. “Strong AI” assumes that the technology systems are able to develop the same level of intelligence as humans. And possibly even exceed them. As things stand, this type of AI is still purely theoretical but is the same technology that we frequently see in sci-fi films.
However, it is important that we take ethical aspects into account for the further development of AI and that we do it now. Whether deliberate or not, incorrectly developed or trained AI can discriminate against certain groups in society, be attacked through security vulnerabilities, or make completely the wrong decisions.
Expert opinions on the topic of artificial intelligence and questions such as “Can algorithms develop feelings?” or “Do we need to fear superintelligence?” are addressed in the specials on the Digital Responsibility initiative.