You can start AI with high-school math. You need intuition for four ideas, not proofs. Here they are, honestly scoped.
1. Vectors — a list of numbers with direction
A data point is a vector: a house = [bedrooms, area, age] = [3, 1200, 10]. Words become vectors (embeddings). "Similar things point in similar directions" — that is the whole intuition behind search and recommendations.
2. Matrices — a grid of numbers
Your entire dataset is a matrix (rows = examples, columns = features). Neural networks are chains of matrix multiplications. You do not compute them by hand — NumPy/PyTorch do — but knowing "data flows through as matrix math" demystifies everything.
import numpy as np inputs = np.array([3, 1200, 10]) # one house (a vector) weights = np.array([50000, 200, -1000]) # learned importance of each feature price = inputs @ weights # @ = matrix/dot product -> a prediction!
3. Derivatives — the slope, for learning
A derivative answers "if I nudge this input, which way and how fast does the output change?" That is exactly what gradient descent needs to know which way is downhill. You will never differentiate by hand — autograd does it — but the concept IS learning.
4. Probability — because the world is uncertain
Models output probabilities: "87% spam". You need: what a probability means (0–1), and that a classifier picks the highest-probability class. Bayes and distributions come later, only if you go deep.
Truth: you can build and ship useful ML with this intuition alone. Add rigor after you have trained a few models and the math has something to attach to. Learning math first, in the abstract, is why most people quit.