Central Limit Theorem

The Central Limit Theorem is the reason averages feel “predictable” even when individual data points are messy. It explains why, with enough data, the mean of a sample starts behaving like something simple and familiar.

What it says (without the scary wording)
Take many random samples of the same size from a population, compute the mean of each sample, and look at the distribution of those means. The Central Limit Theorem says that as the sample size grows, that distribution becomes approximately normal (Gaussian), even if the original data are skewed or weirdly shaped.

The key technical pieces
The theorem applies under fairly broad conditions (independent observations, finite variance, and not-too-extreme dependence). It also tells you how spread out those sample means are:

• The sample mean centers around the true population mean μ.
• Its standard deviation is the standard error: σ/√n, where σ is the population standard deviation and n is sample size.
• Bigger n → smaller uncertainty in the mean.

Concrete examples

• House prices: Individual prices are skewed, but the average price from many random neighborhoods tends to be close to normal.
• Exam scores: Even if one class has a lopsided score distribution, the mean score across many randomly selected groups of students becomes roughly Gaussian.
• Manufacturing: Average thickness from samples of parts stabilizes and supports tight quality control limits.

Why it matters in AI/ML
Many ML practices lean on “means are approximately normal”: confidence intervals for metrics (accuracy, RMSE), A/B testing, estimating uncertainty with repeated sampling, and approximations behind methods like least squares. Even when models are complex, the CLT often justifies treating aggregated estimates as near-Gaussian, enabling practical inference.