AI Fundamentals

Learning objectives

After this lesson you will be able to:

• Classify artificial intelligence according to the EU AI Act definition
• Explain the difference between machine learning and generative AI
• Recognize typical risks such as bias and hallucinations
• Name the limits and practical challenges of AI

What is artificial intelligence?

Artificial intelligence (AI) refers to computer systems that can perform tasks that normally require human intelligence. The EU AI Act defines AI systems as software that works with machine learning or logic-based approaches and generates results such as predictions, content, decisions or recommendations.

Important: AI has no consciousness and no understanding in the human sense. AI systems recognize patterns in large volumes of data and apply those patterns to new situations.

Machine learning vs. generative AI

Machine learning (ML) is a subfield of AI. ML algorithms learn patterns from data without being explicitly programmed for each individual task. Applications include spam filters, recommendation engines and fraud detection.

Generative AI is a subcategory of ML that produces new content:

• LLMs (large language models) such as GPT, Claude or Llama generate text, code and dialogue
• Image generation such as DALL-E, Midjourney or Stable Diffusion
• Audio and video generation produces speech, music or video

An LLM works like an extremely advanced text-completion mechanism: it predicts, word by word, which word should come next.

Bias & discrimination

AI models learn from training data. If that data contains historical distortions, the model learns them too.

Examples of bias:

• Gender bias: an AI recruiting tool prefers male candidates because the training data contained more male executives
• Ethnic bias: facial recognition performs worse on people with darker skin
• Cultural bias: a language model understands cultural contexts from English-speaking countries better

Hallucinations

An AI system "hallucinates" when it delivers convincing but factually wrong information. This happens because LLMs imitate statistical patterns without any underlying understanding of truth or factuality.

The limits of AI

• No real understanding: AI does not understand what it says or does
• Dependence on data quality: poor training data → poor results
• No general knowledge: AI is limited to its training domain
• Context loss: in longer conversations, LLMs lose the context
• Compute intensity: large models consume enormous amounts of energy
• No causality: AI recognizes correlations, but not causes

→ Go deeper: EU AI Act & risk classes — how the EU AI Act regulates AI systems