You can use AI tools without any math. To understand how models work — or to build them — four areas matter most. The free book Mathematics for Machine Learning by Deisenroth, Faisal and Ong covers all of them.

1. Linear algebra: the language of data

Data inside a model is stored as vectors (lists of numbers) and matrices (tables of numbers). A neural network layer is essentially a matrix multiplication followed by a simple function. Key ideas: vectors, matrices, dot products, and the notion of “distance” between vectors — the basis of embeddings and similarity search.

2. Calculus: how models learn

Training means adjusting parameters to reduce error. The derivative tells you which direction reduces the error fastest; gradient descent follows it step by step. Backpropagation is the chain rule applied efficiently through many layers. You need the intuition of a gradient far more than the ability to compute integrals.

3. Probability: dealing with uncertainty

Language models output probabilities for the next token. Classification models output probabilities for each class. Concepts to know: probability distributions, conditional probability and the idea of sampling.

4. Statistics: knowing if a result is real

Evaluating models is statistics: averages, variance, confidence intervals, and the difference between training and test data. Most “my model improved” claims fail here.

How to learn it efficiently

  • Learn on demand. When a concept appears in a course, study just that.
  • Use visual resources. Geometric intuition for vectors and gradients sticks better than formulas.
  • Code it. Implementing gradient descent in a few lines of Python teaches more than a chapter of theory.

Next: see gradient descent explained with one example.

How much math for which goal

  • Using AI tools at work: essentially none beyond basic numeracy and statistics literacy (understanding percentages, averages and uncertainty).
  • Building applications with AI APIs: basic statistics for evaluation; little else.
  • Applied machine learning engineer: working knowledge of linear algebra, calculus concepts (gradients), probability and statistics.
  • Research scientist: deep linear algebra, multivariable calculus, probability theory, optimization and often information theory.

Choose the depth that matches your goal, and deepen as needed when concepts appear in practice.

A worked micro-example

Consider a model that predicts house prices with a single weight: price = w × size. If a 100 m² house costs 300,000 and the model predicts 250,000 (w = 2,500), the error is 50,000. The loss might be the squared error: 50,000². Calculus tells us how the loss changes as w changes — the gradient — and gradient descent nudges w upward to reduce the error. Linear algebra extends this to thousands of features at once, and probability helps express how confident the prediction is. One tiny example uses all four areas of math.

Resources

Free resources include Khan Academy for foundations, 3Blue1Brown’s visual series on linear algebra and calculus, and university course materials published online. Books such as Mathematics for Machine Learning by Deisenroth, Faisal and Ong are available free online from the authors.

Common sticking points

Learners often struggle with notation more than ideas. Translate formulas into words, write small code examples that compute them, and draw diagrams of vectors and functions. Seeing a matrix multiplication or a gradient in code often makes the math click.

Put the ideas into practice with our gradient descent example and the free learning path to understand AI.

What to watch next

  • Interactive math courses designed for machine learning learners.
  • AI tutors that explain math step by step.
  • Research on the mathematical foundations of deep learning.
  • New textbooks bridging math and modern AI.

Key takeaways

  • Linear algebra, calculus, probability and statistics are the core math of AI.
  • The depth you need depends on your goal, from none to advanced.
  • Learning math alongside practical projects is more effective than studying it in isolation.
  • Free resources make it possible to learn the necessary math at no cost.

Frequently asked questions

Can I learn AI without math?

You can use and build with AI tools without much math. Understanding how models work deeply requires more.

Which math should I learn first?

Start with basic statistics and linear algebra concepts like vectors and matrices, then gradients from calculus.

Is there a free math book for machine learning?

Yes. Mathematics for Machine Learning by Deisenroth, Faisal and Ong is freely available online.