AI system: inference, not just rules

An AI system is software that learns patterns from data to produce outputs like predictions, content, or decisions. Unlike a traditional program that follows hard-coded if-then rules, an AI system adapts its behaviour based on the input it receives. For example, a spam filter that uses a fixed list of blocked words is rule-based, whereas one that learns from millions of emails to flag new spam patterns is an AI system. This distinction is crucial: if an EPSO scenario describes a program that always applies the same logic regardless of input, it is not an AI system.

To spot the difference quickly, ask: does the software change its output based on patterns it has learned, or does it follow a fixed set of instructions? If the answer is the latter, it is not AI. Another trick is to look for keywords like 'training data', 'model', or 'inference' in the question—these signal an AI system. Conversely, terms like 'hard-coded', 'rule-based', or 'deterministic' point to a non-AI program. Remember that the AI Act definition focuses on inference, not just automation.

A compact way to test yourself: imagine a program that recommends movies. If it uses a list of genres you selected, it is rule-based. If it analyses your past ratings and similar users' choices to predict what you might like, it is an AI system. Keep this contrast in mind: rules versus learned patterns.