Machine learning is the method that lets software learn from examples instead of being programmed with fixed rules – the system finds the patterns in the data itself and applies them to new situations.
Classic software does exactly what it is told – no more, no less. Machine learning turns that around: instead of describing the solution, you show the system thousands of examples of the task and let it find the pattern itself. That is why modern AI can write, translate and recognise images without anyone having formulated the rules for doing so.
The price of that ability is predictability. A system that has learned from examples inherits the biases of those examples and can fail in ways nobody foresaw. That is why machine learning goes hand in hand with human oversight – not as a technical detail, but as a division of labour.
In practice
Language models such as ChatGPT and Claude are machine learning at scale: they have learned the patterns of language from enormous volumes of text. That explains both their strength – they write fluently about almost anything – and their weakness: they can sound confident when they are wrong.
How to explain it to leadership
»Machine learning means the software has learned from the past. Our job is to judge when the past is a good guide – and when it is not.«
The concept is the foundation of generative AI – and of the whole development I describe in What is AI?.