artificial intelligence

Artificial intelligence — AI for short — is an umbrella term for computer programs that solve tasks that would previously have required human judgment. This includes understanding text, recognizing speech, describing images, or learning from experience. AI is not a single method, but an entire family of methods. What they have in common: the program is not programmed step by step by hand — it develops its abilities on its own by analyzing large amounts of examples. The result is a model, meaning a kind of stored knowledge that the program can apply to new tasks.

Why AI is changing so much right now

AI is not a new concept — its basic ideas date back to the 1950s. For a long time, these approaches failed because computers were too slow and datasets too small. This has changed fundamentally over the past ten years. Graphics cards, originally built for video games, turned out to be ideal for AI computations. At the same time, the internet made billions of texts, images, and videos accessible — enough learning material to train truly powerful models.

The economic significance is enormous. Companies developing leading AI models are now valued on the stock market higher than the world’s largest oil companies. At the same time, AI is transforming entire professional fields: tasks that used to take hours — such as summarizing a contract or checking an X-ray for abnormalities — are now completed by AI systems in seconds.

How an AI model acquires its abilities

The central mechanism is called training. The program is fed many examples — for instance, millions of labeled photos of cats and dogs. At first, the program tries to wildly guess which image shows which animal. If it’s wrong, it is corrected internally. This process is repeated billions of times until its error rate is small enough. What remains afterward is a network of millions or billions of numerical values that store the learned knowledge in compressed form.

Roughly, two phases are distinguished: training and inference. During training, the model learns — this takes weeks and consumes a lot of electricity. During inference, the finished model is used for real requests. Both phases are technically very different. A common misconception is to believe that an AI model actually thinks or understands. It recognizes patterns in data and produces an answer based on them — without consciousness or intention behind it.

There are also different forms of AI. Narrow AI — also called weak AI — is specialized for one task, such as playing chess or detecting spam. So-called general-purpose models, like large language models, can instead handle many different tasks without having been trained individually for each one. True general AI, which matches human thinking in every respect, does not yet exist — it is a research goal, not a finished product.

AI in everyday life, in products, and in the news

AI is encountered constantly today, usually without being noticed. When Spotify makes a suggestion, when Google Maps calculates a detour, or when a smartphone unlocks by facial recognition — a trained model lies behind all of it. Far more visible are applications like ChatGPT or Gemini, programs that let you communicate in natural language via text or speech.

In medicine, AI helps detect tumors in images earlier than a human doctor could. In the financial industry, AI systems check in real time whether a credit card payment looks suspicious. And in schools, AI is increasingly being discussed — both as a learning tool and as a source of problems, for instance when homework is written by machine.

In business news, the term usually appears in connection with investment, regulation, or the competition between the US and China. In 2024, the EU became the first major economic power to introduce binding rules for the use of AI with the so-called AI Act, a law regulating AI systems. This shows that AI is no longer merely a technical topic, but a social and political issue.