My neighbour spent a Saturday last month signing up to four AI courses. By Sunday night she had watched twelve minutes of one, bookmarked the other three, and felt worse about the whole thing than when she started. Nothing about that story is unusual. Free AI courses are everywhere. The hard part is knowing which ones were written for someone with no background at all, and then surviving past the first week.
So this is a shortlist rather than a directory. Four free courses that genuinely assume you know nothing, an honest note on what each one asks of you, and a few tactics for actually reaching the end.
Free is not the same as beginner-friendly
Plenty of courses label themselves introductory and then open lesson two with a matrix. That isn’t malice. Course authors tend to forget what zero feels like.
The giveaway is usually in the prerequisites line. “Basic Python” and “familiarity with linear algebra” mean the course was written for people switching over from software or maths, not for someone starting cold. There’s nothing wrong with those courses. They’re just a bad first step.
If you want to test the water before committing, browsing free AI courses you can open without signing up lets you read lesson one and judge the difficulty yourself in about five minutes. That beats reading reviews.
Four free courses that assume you know nothing
Elements of AI (University of Helsinki)
Six chapters, no coding, roughly 30 hours end to end. It was built by the University of Helsinki and the tech firm Reaktor with an explicit goal of getting one percent of the Finnish population to understand what AI actually is. That public-service framing shows. The writing is plain, the examples are everyday, and you finish able to hold a sensible conversation about neural networks without ever having built one. Free to complete; the certificate costs money if you want it.
Microsoft’s AI for Beginners
A 24-lesson curriculum on GitHub, spread across 12 suggested weeks, covering everything from symbolic AI to computer vision, with quizzes and labs attached to each lesson. Entirely free and self-paced, and it’s one of the few beginner programmes that doesn’t pretend deep learning is the only kind of AI that exists. The free Microsoft AI courses worth downloading get a fuller breakdown elsewhere if you want the details.
Kaggle’s Intro to Machine Learning
Seven lessons, about three hours. You write code in the browser with most of it already filled in, so you spend your time adjusting and running rather than wrestling with a development environment on day one. A good first taste of what training a model involves.
Harvard’s CS50 Introduction to AI with Python
Free to audit on edX. Seven lectures plus projects on search algorithms, logic and probability. It’s the most demanding item here and you shouldn’t touch it first, but it’s the natural next step once the vocabulary stops feeling foreign.
Google’s Machine Learning Crash Course gets recommended alongside these and it deserves the praise, at around 15 hours and 30 lessons with interactive visualisations. It just sits better as course number two, once you’ve seen a loss function and stopped being frightened by it.
What to leave alone for now
Almost every “learn AI” list includes fast.ai, Andrew Ng’s deep learning specialisation and a scattering of research papers. All fine resources. All wrong for week one. The guide to what to learn first and what to skip covers the reasoning in depth, but the short version is this: a course you abandon in frustration teaches you nothing, and starting too high is the single most common reason people abandon one.
How to spot a good beginner course from a time sink
- It shows output before theory. Good courses let you run something in lesson one and explain why it worked in lesson three.
- It uses concrete examples. Predicting house prices, filtering spam, spotting pneumonia on a chest X-ray. Vague definitions of “intelligence” are a warning sign.
- It gives you something to check your work against. Quizzes, notebooks, worked solutions. Passive video watching feels productive and isn’t.
- It’s honest about scope. A three-hour course promising that you’ll “understand the basics” is more trustworthy than one promising mastery.
- It doesn’t need a paid tool. Anything nudging you toward a subscription in module two isn’t really free.
The real obstacle is finishing
Completion rates for free online courses land somewhere between 5 and 15 percent, depending on who’s counting. Paying nothing is part of the problem. There’s no sunk cost dragging you back to lesson four on a wet Tuesday.
A few things beat willpower. Pick one course, not four, and close the other tabs. Book 25 minutes rather than “a few hours this weekend,” because the smaller commitment is the one you keep. Tell a friend which lesson you’re on so there’s someone to be mildly accountable to. And if the urge to switch courses hits at lesson three, ride it out. That urge usually passes within a fortnight and is mostly boredom with the early material. There’s more on making it stick in this piece on how to actually finish a free AI course.
A first month that gets you somewhere
Week one: Elements of AI, chapters one to three. About 90 minutes, no code, and you come out able to explain supervised learning and why a model can be overfitted. Week two: chapters four to six. Week three: Kaggle’s seven lessons, which is a single evening if you sit down properly. Week four: start Microsoft’s curriculum or Google’s Crash Course and see which one holds your attention better.
That’s roughly ten hours across a month. It won’t make you employable, and any course claiming otherwise is selling something. What it will do is give you the vocabulary to read about AI without skimming, the ability to judge a tool or a headline on its merits, and a clear sense of whether you want to go deeper. From there, a structured 12-week learning roadmap makes the next stretch far less vague than “learn AI.”
The other decision nobody raises early enough is direction. Building things means Python, APIs and a lot of debugging. Understanding them means maths, papers and patience. Neither is the more legitimate path, and neither needs deciding this month. Open Elements of AI chapter one tonight and give it half an hour. If it sticks, you’ll know by Thursday.