There is a whole generation of learners right now whose entire idea of artificial intelligence is a chat box, and this course is a useful corrective, because it teaches the field as it existed and largely still exists underneath the hype. Ansaf Salleb-Aouissi runs Columbia's survey on edX, and it covers the classical canon, intelligent agents, uninformed and informed search, adversarial search for game playing, constraint satisfaction, the fundamentals of machine learning, and logic and knowledge representation, before opening out into planning, robotics, and a first look at language. This is the material that the famous Russell and Norvig textbook is built from, and learning it gives you something most prompt-era courses cannot, which is the ability to see that intelligence in machines has a long structured history and a set of techniques that did not stop being true when transformers arrived. The programming projects are the making of it.
Implementing a search agent or a game-playing algorithm yourself is what turns these from lecture concepts into things you actually understand, and it is the difference between this and a passive overview. The honest framing is about expectations. This is classical AI first, so if you came hoping for weeks on deep neural networks and large language models you will be disappointed, those get a nod rather than a deep dive, and you should take a dedicated modern course alongside it for that. It also asks for real effort, comfortable Python and a willingness to engage with the maths, and it runs at the length and intensity of an actual university semester, so the eight to ten hours a week is not marketing, it is the real cost.
Audited for free and taken seriously, it is one of the best ways to build the foundation that makes everything newer easier to place, and I would happily recommend pairing it with a deep learning course to get a picture of the field that is genuinely complete rather than fashionable.