Whether in industry or in the private sphere – artificial intelligence is on everyone’s lips. But what does artificial intelligence mean? In this article, we answer the most important questions on this topic
Artificial intelligence is one of the really big future topics of our society
- What is artificial intelligence?
- How intelligent is artificial intelligence really?
- How does artificial intelligence work?
- How do you program artificial intelligence?
- What do we need artificial intelligence for?
- Where is artificial intelligence used?
- Smart machines: How is artificial intelligence changing our lives?
- How is artificial intelligence changing our society?
What is artificial intelligence?
When developing systems with artificial intelligence (abbreviated: KI, English: Artificial Intelligence), researchers and developers try to emulate human perception and human action using machines. An official definition of the term Artificial However, there is no such thing as intelligence, which is partly due to the abstract nature of the term intelligence and the rapid change in the subject area. Therefore, AI is mainly defined by its properties and the associated sub-areas, such as speech recognition, image processing or machine learning.
Properties that are characteristic of AI are autonomy and adaptivity. AI systems have the ability to perform tasks in complex environments without constant human guidance. They are also able to improve their performance independently by learning from their experiences.
In general, the systems are divided into weak and strong AI:
⦁ Weak AI refers to machines that can replace a single human cognitive ability. Systems with weak AI perform a specific task and behave intelligently.
⦁ A strong AI would be a machine that has the same abilities as a human or even surpasses human abilities. The system with strong AI could thus fulfil any intellectual task. It would be intelligent (compared to the weak AI that just behaves intelligently) and would be conscious.
The AI solutions that exist to date all fall into the ‘weak AI’ category. Strong AI systems only exist in the field of science fiction.
The term artificial intelligence includes many sub-areas, which in turn include numerous AI methods and AI applications, for example:
⦁ knowledge acquisition and representation
⦁ Pattern recognition (images, text, speech)
⦁ Prediction (Big Data / Predictive Analytics)
Machine learning (deep learning/neural networks)
Machine learning or the special form of deep learning, in particular, are often regarded as the most important method of artificial intelligence, because learning from experience and the ability to generate new actions from it is evidence of intelligence.
How intelligent is artificial intelligence really?
The answer to this question is closely related to the definition of the word ‘intelligent’. Because the question is whether intelligence is the same as intelligent behaviour or whether intelligence requires a brain and a consciousness.
Provided that intelligence equals intelligent behaviour, the English mathematician and logician Alan Turing developed a test to determine whether a machine is intelligent. In the so-called Turing test, a human investigator interacts with two chat partners – one of them is human, one is a computer. If the investigator is not able to distinguish artificially from human intelligence by exchanging written messages, the computer has passed the test and must therefore have reached the level of human intelligence.
But even if a machine behaves intelligently and thus passes the Turing test, it still does not have a human brain. If this is assumed for real intelligence, then only strong AI systems would be really intelligent.
For the areas in which AI is currently being used, the question of ‘real intelligence’ is not that important. What is far more important is how well a system can perform the task for which it is designed. In this context, the following often applies: artificial intelligence is only as smart as the data with which it is trained
How does artificial intelligence work?
The topic of artificial intelligence includes many AI methods and AI applications, each of which has different functionalities. In general, the intelligent behaviour of the technologies is simulated using computer science and mathematics/statistics. The computers are trained for specific tasks, often by processing large amounts of data and recognizing patterns in it.
This requires certain skills that can be divided into four areas: perception, understanding, action and learning.
- Perceive: This skill creates the data that an AI-powered system operates on. 1. Sensors of all kinds, including cameras and microphones, are used for this purpose.
- Understanding: This part of the AI is the processing component from which the control panel emanates. The data obtained is processed using statistical analysis, speech/image recognition, algorithms that define special rules, or machine learning.
- Action: This is the output component of the AI that issues commands to connected devices, triggers follow-up processes, or outputs other things such as images, translations, and the like.
- Learning: This area is special about current AI technologies. You can learn from mistakes and feedback from the areas of ‘understanding’ and ‘acting’ during the training phase and also during operation.
Algorithms and neural networks play an important role in both ‘understanding’ and ‘learning’.
What are algorithms?
Algorithms are detailed and systematic instructions that define step by step how a mathematical problem can be solved. Algorithms are implemented, ie translated into a programming language, so that the computer/machine can generate the desired solution from the given information.
There are, for example, sorting or search algorithms (these include Google’s algorithms), but also algorithms that can make complex decisions based on all relevant factors. To do this, an algorithm takes all the factors (which have to be defined beforehand by a human) and links them in all possible variants. He then goes through every possible combination and its consequences and compares them with the programmed specifications. From this, the algorithm calculates the answer that is most likely to be correct.
The class of learning algorithms, which are summarized under the generic term machine learning, is particularly important for AI. You can learn from a large number of example cases and derive general rules. After the learning phase, you can apply these insights to real cases. Among the machine learning algorithms, there are also deep learning algorithms that are used to analyze and process particularly large amounts of data – in connection with neural networks.
How do neural networks work?
Artificial neural networks (ANN) are the attempt to artificially reproduce aspects of the human brain known from research. It is mainly about simulating and using the interaction of nerve cells (neurons) and their connections (synapses).
The goals of artificial neural networks are, on the one hand, a better understanding of the human brain and the processes taking place there and, on the other hand, advances in machine learning by linking huge amounts of data (big data) and deep learning techniques.
A single neuron is a simple information processor that can examine exactly one aspect of information. To process very complex information, many neurons are therefore connected to each other. These connections are called synapses, they form a complex network between the neurons. A neural network (regardless of whether it is biological or artificial) consists of a large number of neurons that can receive signals and transmit them to other neurons via connections.
There are three types of neuron layers in an ANN:
- Input Layer: The first layer of neurons for raw processing of the information. These neurons feed pre-processed information into the network.
- Hidden Layer(s): One to a finite number of neuron layers with linked neurons. During the training, you will learn how to read patterns from the raw information using examples. From these, complex typical characteristics are worked out in order to solve the task.
- Output Layer: The last neuron layer – the so-called output layer. It can represent any possible outcome.
The big advantage of artificial neural networks compared to ‘classical’ data processing is that they can consider several aspects at the same time. While a ‘simple’ algorithm analyzes each property of a data set one after the other, in the layers of the ANN all properties can be analyzed simultaneously.
ANN is used, for example, for image processing, speech recognition, early warning systems or process optimization. It is precisely this process optimization or error detection (via image processing or audio analysis) that is already being used in companies as part of Industry 4.0.
How do you program artificial intelligence?
How a system is programmed with artificial intelligence, depends a lot on what you want the system to do. As with the programming of any computer program, the first priority is choosing the right programming language. The most commonly used programming languages for AI are:
- Python has become the de facto standard for machine learning and big data analysis. It is very versatile and suitable for numerous fields of application, especially in the area of artificial intelligence and the Internet of Things.
- Prolog is considered the most important logical programming language and is therefore particularly suitable for simulators, generators, as well as systems for diagnosis and prognosis – exactly what should often be created in AI projects. For example, the language processing components of IBM’s AI ‘Watson’ are written in Prolog. In addition to Prolog, Watson and the associated DeepQA software engine are also based on Java and C++, among other things.
- find out why IBM is opening its Watson IoT headquarters in Munich.
- As an object-oriented programming language, Java is very versatile and can therefore also be used well for AI programming. Java is particularly suitable for natural language processing, search algorithms and neural networks. Java is slower than C++ but easier to program.
- C++ is particularly relevant for time-critical AI projects, as it is the fastest computer language. It is also particularly extensive and strong in statistics. This makes it ideal for machine learning and neural networks.
- LISP is particularly strong in inductive logic and machine learning, making it well-suited for AI as well.
- R is mainly used in the context of big data for data analysis. The programming language is mainly suitable for statistical methods and allows the free calculation of statistics and graphics.
In addition to an AI-friendly programming language, program libraries and frameworks such as TensorFlow, Torch, Keras or Caffe are required for AI implementation. The most important subprograms, routines and algorithms for AI development are already implemented here, which can then be transferred to your own AI solution.
What do we need artificial intelligence for?
Algorithmic decision-making systems are intended to support or replace people in making decisions. In the area of big data, in particular, this is necessary in order to make decisions possible at all, because here people would not be able to make any well-founded decisions based on a large amount of data alone.
Processes can be accelerated with AI support, especially with the increasing computing capacity of computers. This makes it possible to carry out tasks that humans or machines without AI could not previously handle due to time constraints.
In addition, AI-based systems, especially machine learning, make it possible to predict what will happen in the future.
In general, artificial intelligence is one of many tools to solve problems. AI makes it possible to do things that were previously not possible. It allows overcoming long-standing borders. Therefore, AI systems make a strong contribution to competitiveness.
Where is artificial intelligence used?
The areas of application for artificial Intelligence are extremely diverse. They range from spam filters and product recommendations in e-commerce or streaming services to language assistants such as Alexa or Siri and chess computers to self-driving cars. With the digitization of industry, ie Industry 4.0, many companies are also adopting AI-based systems (although not all Industry 4.0 applications contain AI in the same way).
According to the Federal Association of the Digital Economy, AI solutions are generally found in the area of data collection and analysis in the industrial environment. Examples for this are
- optical inspections through AI applications in the field of quality assurance, which make error or process analyzes possible,
- Process automation in manufacturing and assembly, which includes self-regulating adjustment of control parameters,
- Robots with artificial intelligence, which can perform even better⦁ ‘bin picking’,
- forward-looking data analysis in order to maintain machines and systems as required ( ⦁ predictive maintenance),
- voice control of machines,
- Chatbots for ⦁ virtual customer service other
- ERP systems with AI strategies that optimize internal production structures and processes.
Smart machines: How is artificial intelligence changing our lives?
Spam filters, personalized advertising on the Internet, more relevant search results in online searches, chatbots for customer service inquiries and navigation devices that include traffic jams and construction sites in the route calculation – these are all systems that work with artificial intelligence and are changing our daily lives.
In many cases, it ensures greater efficiency: while in the past you could spend hours waiting in the manufacturer’s queue because of every small question or problem with a product, today chatbots can provide quick and uncomplicated help in some cases. Cars park automatically, stays in lane and keep their distance, and recognize street signs.
Our smartphones help us write messages, correct our spelling mistakes and even predict which words we will use. And you recognize our face and then unlock it automatically.
How does Face ID / facial recognition work?
What is certain is that artificial intelligence is and will be finding its way into many areas of life. Until autonomous driving spreads it will take a while, but smart home applications (eg for saving energy) and digital voice assistants are becoming more and more popular. While only 390 million people used AI-supported assistants in 2015, in 2019 there were almost 1.38 billion. According to a forecast by Tractica, the numbers will continue to rise; By 2021, voice assistants are expected to have 1.83 billion users.
Not only are the users of language assistants such as Apple’s Siri, Amazon’s Alexa or Google Assistant increasing, more and more people are also using AI systems in general: 73 per cent of those surveyed in a Bitkom study stated that they had already used a simple application based on builds AI.
And that’s no coincidence: The majority of Germans consider artificial intelligence to be useful – 88 per cent of the Germans questioned in a PwC survey stated that artificial intelligence will help to master the challenges of the future. They consider AI to be particularly helpful in the areas of cyber security (49 per cent), clean energy/climate change (45 per cent) and protection against diseases (43 per cent).
So if AI researchers manage to develop systems that can master these very challenges, they will be a thing of the past. The technologies can help reduce traffic congestion, speed up administrative tasks, and improve medical diagnosis and treatment.
How is artificial intelligence changing our society?
Artificial intelligence has an impact on almost all areas of our society. It changes how we communicate, meet new people, consume news and do our work – in positive and negative ways.
For example, many people consume News less via platforms aimed at the general public such as television, but in social networks, where they receive personalized content with AI. From 2013 to 2019, the use of social media as a news source in Germany has increased from 18 per cent to 34 per cent; This is according to the Digital News Report 2019 of the Reuters Institute for the Study of Journalism.
According to a Kantar study, the main reasons for this are easy access to a wide range of news sources and the ability to comment on news and share it with others. While on the one hand-personalized messages are very pleasant, they also encourage the emergence of so-called ‘filter bubbles’ and the separation of different social classes. This can lead to societal problems.
Autonomous means of transport and intelligent traffic control will not only support climate protection in the future but will also create more free time for people, which in turn can be used for work or hobbies. At the same time, people’s areas of responsibility are shifting to areas in which creativity and empathy are required.
How do self-driving cars work?
Likewise, artificial intelligence will transform policing and the justice system. Face recognition will become just as common as fingerprints. And analysis algorithms will also become more relevant: Courts in the USA, for example, already use software that calculates the risk of reoffending by criminals. The calculations are based on information about the person and their radius, but also on the Analysis of all similar crimes committed.
The emergence of many new technologies is changing our society not only directly, but also indirectly. For example, according to the KI-Bundesverband, the compulsory subject ‘digital education’ or ‘computer science’ urgently needs to be introduced in German schools. Because it is irresponsible to release young people into the increasingly digitized world without active support.