What is artificial intelligence?
Artificial intelligence is changing how companies work, make decisions and communicate with customers. This page explains what AI is, how it works and where it is used today.
What does artificial intelligence mean?
Artificial intelligence (AI) is a branch of computer science that develops systems which solve tasks that normally require human intelligence. In German the term is Künstliche Intelligenz (KI) – both abbreviations mean the same thing.
The term originated in 1956 at the Dartmouth Conference in the USA. There, researchers first formulated the idea of teaching machines to think.
Today, AI systems recognise patterns in data, make predictions and generate new content.
How it works
How does AI work?
AI works by having algorithms analyse large amounts of data, recognise patterns and derive rules from them that they apply to new situations.
Algorithm
A defined sequence of computing steps that solves a problem – comparable to a recipe.
Training
The model learns from example data and optimises millions of internal parameters over many passes.
Inference
The trained model applies its knowledge to new, unknown data and delivers results.
Categories
What types of AI are there?
AI can be divided into three levels: narrow AI, general AI and superintelligence. All of today's applications – from ChatGPT to image recognition – are narrow AI.
Narrow AI solves one specific task. A chess computer plays chess but cannot write emails.
General AI could handle any cognitive task at a human level – it does not yet exist. Superintelligence would surpass human intelligence in all areas and remains a theoretical concept.
Classification
What levels of AI are there?
Besides the division into narrow and general, research distinguishes four functional levels of AI systems.
Reactive machines
React only to current inputs without memory – such as IBM's chess computer Deep Blue.
Limited memory
Use past data for decisions – self-driving cars and ChatGPT work this way.
Theory of mind
Could understand human emotions and intentions – does not yet exist.
Self-awareness
Would have its own consciousness – purely hypothetical.
Machine learning
What is machine learning?
Machine learning is a subfield of AI in which algorithms learn from data and improve without being explicitly programmed. It forms the basis for most modern AI applications.
Instead of specifying fixed rules, you show the algorithm examples. It recognises patterns on its own and derives rules from them.
The difference from classic programming: a developer does not write if-then logic but defines a learning goal.
Training
Which learning methods does AI use?
AI systems use three basic learning methods, which are applied depending on the task and the data available.
Supervised learning
The model trains on data for which the correct answers are known – such as images with the label “cat” or “dog”.
Unsupervised learning
The model finds structures in data on its own without given labels – such as customer groups with similar buying behaviour.
Reinforcement learning
The model learns through reward and punishment – a robot practises grasping and receives positive feedback on success.
Deep learning
What is deep learning?
Deep learning is a subfield of machine learning that uses artificial neural networks with many layers to recognise complex patterns. It is behind ChatGPT, image generators and speech recognition.
The transformer – an architecture developed in 2017 – processes text in parallel rather than sequentially. It forms the basis for GPT, Claude and Gemini.
Architecture
What are neural networks?
Neural networks are modelled on the human brain. Artificial neurons receive signals, compute them and pass the result on to the next layer.
A neural network consists of an input layer, one or more hidden layers and an output layer. The more layers, the more complex the relationships the network can learn.
Language models
What are large language models?
Large language models (LLMs) are AI models trained on huge amounts of text to understand and generate human language. Well-known examples are GPT (OpenAI), Claude (Anthropic) and Gemini (Google).
Token
The smallest unit an LLM processes – a word, a syllable or a punctuation mark.
Context window
The maximum amount of text an LLM can process at once – currently up to two million tokens.
RLHF
Human raters sort answers by quality – that is how the model learns to prefer helpful results.
Prompts
What is prompt engineering?
Prompt engineering describes the practice of formulating inputs for AI models so that they deliver precise and useful results. The quality of the prompt directly influences the quality of the answer.
A good prompt defines role, context, format and constraints for the model. Instead of “Write a text”, “You are a copywriter for B2B software. Write a product description in 100 words.” works much better.
Creating content
What is generative AI?
Generative AI creates new content – text, images, code, music or videos. It learns from training data which patterns are typical and produces similar but original results.
ChatGPT writes text, DALL-E generates images, GitHub Copilot completes code. Unlike analytical AI, which evaluates existing data, generative AI creates something new.
The breakthrough came at the end of 2022, when ChatGPT reached 100 million users within two months.
Agents
How do AI agents work?
AI agents are systems that plan and execute tasks on their own and use external tools when needed. Unlike chatbots, they do not only react to inputs but act proactively.
An AI agent can work through multi-step workflows – for example researching data, analysing it and producing a report. Companies use AI agents for customer support, data processing and process automation.
Everyday life
Where do we encounter AI in everyday life?
Most people use AI every day without consciously noticing it – in voice assistants, navigation apps, email filters and product recommendations.
Voice assistants
Siri, Alexa and Google Assistant understand spoken commands and carry out actions.
Streaming
Netflix and Spotify analyse usage behaviour and recommend matching content.
Medicine
AI analyses X-ray images and supports diagnoses with pattern recognition.
Navigation
Google Maps calculates routes based on real-time traffic data and historical patterns.
Business
How can AI be used in companies?
AI tools automate routine tasks and free up capacity for strategic work. Which AI use cases offer the greatest leverage depends on industry and processes.
How AI in mid-sized companies is used in concrete terms is shown in the practical overview with fields of application and entry points.
Customer support
Chatbots answer standard enquiries around the clock and escalate complex cases to staff.
Data analysis
AI recognises patterns in sales figures, customer behaviour and market data faster than manual analysis.
Content creation
Language models draft texts, summaries and translations in seconds.
Risks
What risks does artificial intelligence have?
AI systems occasionally generate false information that sounds plausible. This phenomenon is called hallucination – it arises because language models calculate probabilities and do not check facts.
Bias is another risk. If training data under-represents certain groups, the model reflects this distortion in its results.
An application screening system could systematically disadvantage certain candidates. Countermeasures are diverse training data, regular audits and human control.
Besides hallucinations and bias, cyberattacks, energy consumption and regulatory obligations are among the central AI risks for companies.
Limits
What can AI not do?
AI has no consciousness, no understanding and no intentions – it calculates probabilities based on statistical patterns.
No world knowledge
AI only knows what is in its training data – current events are missing without web access.
No logical reasoning
AI simulates conclusions but does not understand causality in the human sense.
Energy consumption
Training and operating large models consume considerable amounts of electricity and computing resources.
Law
How does the EU regulate artificial intelligence?
The EU AI Act is the first comprehensive AI law in the world. It divides AI applications into four risk categories – from minimal to unacceptable.
The higher the risk, the stricter the requirements for transparency, safety and documentation.
In parallel, the existing data protection rules of the GDPR apply. AI models need large amounts of data for training; the GDPR demands data minimisation. Companies must actively take both sets of rules into account.
FAQ
What are the most frequently asked questions about AI?
Artificial intelligence is the umbrella term for all systems that imitate human cognitive abilities. Machine learning is a subfield of AI in which algorithms learn from data without being explicitly programmed.
Every machine learning system is AI, but not every AI is based on machine learning. Rule-based expert systems are also AI but work with fixed if-then rules instead of learned patterns.
ChatGPT (basic version), Google Gemini, Microsoft Copilot and Claude offer free access with a limited feature set. For image generation, tools such as DALL-E (limited credits) or Stable Diffusion (open source) are available.
The free versions are suitable for first experiences with AI. For professional use in companies, the paid variants with higher limits, better models and data protection guarantees pay off.
AI mainly changes professions with a high share of routine tasks: clerical work, accounting, simple programming and translation. Creative, social and strategic professions are supported but not replaced.
New job profiles emerge in parallel – such as prompt engineer, AI trainer or AI consultant. Those who understand and use AI as a tool benefit most from the development.
AI learns by analysing large amounts of data and recognising patterns. In supervised learning, the model receives examples with correct answers. In unsupervised learning, it finds structures on its own.
In reinforcement learning, the model optimises its behaviour through reward and punishment. LLMs such as ChatGPT combine pre-training on large amounts of text with human feedback (RLHF) to generate helpful answers.
A hallucination arises when an AI model generates false or invented information that sounds plausible. That happens because language models calculate probabilities and do not check facts.
Countermeasures are retrieval-augmented generation (RAG), grounding with current data and human review. No current AI system is completely free of hallucinations.
Transparency
Sources used
- Own research
- Stanford HAI: “Artificial Intelligence Index Report” (2025)
- McKinsey: “The State of AI” (2025)
- European Parliament: “Artificial Intelligence Act – AI Act” (2024)
- Reuters: “ChatGPT sets record for fastest-growing user base” (2023)