Overview
AI systems perform tasks such as classification or language processing. Machine learning fits patterns in examples to predict new cases; a spam filter estimates whether a message resembles labeled spam. Predictions are not certainty or human understanding. Use depends on data quality, task, and consequences.
Core ideas
Training examples shape model parameters. Separate test data helps check generalization rather than memorization. Accuracy can hide uneven errors, so inspect precision, recall, calibration, and subgroup performance. Documentation, privacy, oversight, and ways to challenge consequential decisions support responsible use.
Try it
With a public flower-measurement dataset, compare a decision tree against a majority-class baseline. Split training and test examples first. Record accuracy and inspect a confusion matrix for confused classes. Change tree depth and compare. This learning exercise should never judge people or guide high-stakes decisions.
Study pathways
Study statistics, programming, and visualization, then supervised learning and evaluation. Further reading: MIT OpenCourseWare offers learning materials at https://ocw.mit.edu/.