Machine Learning

Show a model 10,000 photos labelled 'cat' or 'dog'. It tweaks millions of internal dials until it can guess correctly on photos it has never seen. That tweaking process is 'training'.

Supervised: learn from labelled examples. Unsupervised: spot patterns in unlabelled data. Reinforcement: learn from trial and error with rewards (how AlphaGo learned Go).

Bad data, biased data, or too little data. A resume-screening model trained only on past hires will replicate past hiring bias. The model isn't 'wrong' — the data taught it that.