What is machine learning?
Machine learning is a branch of AI in which a system learns from data without being explicitly programmed, and thereby recognises patterns, makes predictions, or automates decisions.
DEFINITION
Machine learning is the method by which almost all modern AI systems are trained. Instead of exact rules, a computer receives many examples and learns patterns from them. If an ML model receives thousands of emails marked as to which are spam, it learns to recognise new spam emails on its own. The model learns from mistakes and improves with more data. There are three main approaches: supervised learning with labelled examples as the training basis; unsupervised learning, where the model finds structures in data itself; and reinforcement learning, where the model learns optimal behaviour through reward and punishment. Machine learning is the technical foundation for language models, image recognition systems, recommendation algorithms, and many other everyday AI applications.
CONNECTIONS
Leadership
Machine learning learns through feedback loops — just like successful organisations. A feedback culture that openly documents and evaluates failures is therefore also the basic prerequisite for good ML datasets.
Agility
ML models can improve velocity forecasts by evaluating historical sprint data. That makes capacity planning less intuitive and more data-driven, leading to more realistic commitments.
Project Management
Machine learning suits risk identification in complex projects: patterns from historical project data (delays, budget overruns) can point early to similar patterns in the current project.
KEY POINTS
- ML systems learn from data without explicitly programmed rules.
- Three main approaches: supervised, unsupervised, and reinforcement learning.
- The quality and quantity of training data determine model quality.
- ML is the technical foundation for LLMs, image recognition, and recommendation systems.
- ML models can reflect gaps or errors in training data.
EXAMPLE
An insurer trains an ML model on historical claims data to assess new applications automatically. The model recognises patterns that indicate fraud and flags suspicious cases for manual review. The team of twelve case handlers can focus on complex cases and processes twice as many applications overall.
MISCONCEPTIONS
Does an ML system really learn like a human?
No. ML optimises mathematical functions based on data. That has little to do with human understanding of learning or intuition. The term “learning” is a useful analogy, not a precise description.
Do I need huge amounts of data for machine learning?
Not always. For certain tasks, a few hundred examples are enough. Complex models like LLMs need very large datasets, but many specialised ML applications work with far less.