Artificial Intelligence
AI
In short
Artificial intelligence is the umbrella term for software that infers from examples instead of following rules someone wrote down. An ordinary program does exactly what it was told. An AI system finds patterns in data and applies them to cases nobody anticipated. Both of its properties follow from that: it is useful on problems nobody could specify, and it can be wrong without anything being broken.
One difference, and everything follows from it
Ordinary software is deterministic: same input, same output, every time. An AI system is not. It returns the most probable answer, not the demonstrably correct one.
That is not a defect a later version fixes. It is how the thing is built. Once you accept it, the useful question changes from «is the AI reliable» to «what happens when it is wrong, and who notices».
Three terms people use interchangeably, and shouldn't
Artificial intelligence covers anything that handles tasks a person would otherwise do. Machine learning is the subset that learns this from data rather than from rules. Generative AI is a subset of that again – the part that produces new content: text, images, code.
When someone says «AI» today they usually mean the smallest of the three. The confusion has consequences: fraud detection in your accounting is AI, produces nothing, and needs no chat interface whatsoever.
Where it stalls in a smaller company – rarely at the model
The models are good enough for almost any task a small or mid-sized company has. What is usually missing is one of three things: data in one place instead of four systems, a bounded problem instead of «something with AI», and a person who approves before anything leaves the building.
The third is the one most often skipped. An AI that sends email on its own saves ten minutes and costs a client when it misfires. We automate the draft, never the send.
What is measured, and what is sales copy
One figure with a source: the EU AI Act (Regulation (EU) 2024/1689) requires in Art. 50(4) that AI-generated content be disclosed when it is made available to the public. That is binding law, not guidance – and it reaches Swiss companies that operate into the EU.
What is not measured is nearly every productivity figure in a vendor deck. Anyone promising «40 % more efficiency» has not measured it in your company. We label figures from our own practice as exactly that, and we run the numbers on your case before the work starts.
How we handle it
ALPENIQ AI starts with one bounded case that can be costed, not with a platform. The next case follows once the first one holds.
How we work: AI consulting, process automation and data analytics.
Related terms
A term in your quote that nobody explained?
In a strategy call we translate the offer in front of you – even when it did not come from us.