Start learning

Start here

Start learning AI with no background

You don't need a degree, a math background, or a job in tech. You need twenty minutes a day and a tolerance for being wrong. Here's what to do first, and what to skip.

Four beliefs that stop people before they start

"I need to be good at math first"

You need math to build certain things. You don't need it to understand what machine learning is, why a model can be confidently wrong, or how to judge whether an AI system is doing its job. When the math does turn up you'll want it, because by then it's answering a question you've already hit. Learning it up front, with no question attached, is how people quit in week two.

"I should learn to code first"

Half true, and a popular way to postpone starting. Code lets you test an idea, which is how you find out whether you understood it. But you can build a real picture of how AI works before you write a line, and the code comes faster once you know what you're coding toward.

"The field moves too fast to bother"

The headlines move fast. The ideas underneath barely move. What a model is and why it fails on inputs it has never seen hasn't changed in years. If you feel permanently behind, you're tracking product launches rather than learning the subject.

"I'll just ask a chatbot"

Ask it. It's good at explanations and it's free. Then watch what happens: you read a clear paragraph, you nod, and a week later you can't reconstruct a word of it.

What's missing is being wrong first — committing to an answer, watching it fail, finding out which belief produced it. You can't ask for that. It has to be arranged.

Two routes through the same idea. Reading an explanation ends with nothing a week later; guessing first, getting it wrong and seeing why leaves the idea in place. Read the explanation Read it Sounds right Gone in a week Feels like learning. Guess first Guess Get it wrong Still there Feels worse. Is learning.
Both routes cost the same ten minutes. Only one of them leaves you with anything.
The one-line version

Don't prepare to learn AI. Start, and pick up each prerequisite the moment it blocks you — which is also the moment you can see the point of it.

Week one

  1. Pick a fixed twenty minutes. The same twenty minutes every day beats three hours on Sunday, and it isn't close. Here's why.
  2. Guess before you read. A wrong guess costs nothing and tells you which belief you were carrying.
  3. Don't take notes yet. Close the page and say the idea out loud instead. If you can't, you've found that out in fifteen seconds rather than at the exam.
  4. Sleep on the confusing bit. Coming back tomorrow works better than grinding at it tonight, and it's far less miserable.
  5. Stop at twenty minutes even when it's going well. Marathons are how people burn out on day nine.

What you can postpone

A tall stack of imagined prerequisites — calculus, linear algebra, a computer science degree, Python fluency, a GPU, a development environment — beside the two things actually needed on day one: twenty minutes a day and a willingness to guess. What you think you need What you need A dev environment A GPU Python fluency A CS degree Linear algebra Calculus Willing to guess 20 minutes a day
The left-hand stack is why most people never start. None of it is needed on day one.
  • Linear algebra and calculus — until a lesson makes you want them.
  • Picking a specialty — vision, language, robotics. The foundations are shared, and you can't choose well before you know what the options involve.
  • Setting up a development environment — a real obstacle, and not a real prerequisite. Deal with it when you have something to run.
  • Research papers — written for people who already have the context. Starting there teaches you mainly that you're stupid, which is false and discouraging.

How to tell it's working

Not by ground covered. By whether you can predict: say what a system will do in a situation you've never been shown, and how it will break. Someone who has only read about it can pick the right answer out of a list, which is a different thing.

Test yourself with the five checks in How to tell whether you actually understand something.

Next