Building AI: Going a Step Deeper

Building AI feels noticeably more serious than Elements of AI.

Elements of AI is something you can almost move through conversationally. It introduces the ideas, gets you thinking and gives you a broad foundation. Building AI asks a bit more of you.

I found myself slowing down, rereading sections and spending longer with the exercises. I probably enjoyed it slightly more because of that extra depth.

Learning through intuition, not just definitions

One of the exercises I still remember clearly was simulated annealing.

The course used an interactive example where you could change the temperature and see how that affected the search behaviour. Set it too low and the system stayed close to where it already was. Set it too high and it could wander all over the place. Somewhere in between was the useful balance.

That made the idea much easier to understand than simply reading a definition.

It also gave me a useful mental model for later. Once you've seen how changing a parameter can make a system more constrained or more exploratory, you start recognising similar ideas elsewhere.

Another memorable section covered generative adversarial networks, or GANs, and how two competing networks can be used to generate images.

I certainly didn't finish the lesson knowing how to build a GAN myself, but that wasn't really the point. It meant that when those ideas came up again later, they were no longer completely new.

You don't need strong Python

One reason I'd spent time learning Python beforehand was because Building AI offers different levels for some of its exercises.

The most advanced answers can involve Python.

My Python wasn't good enough to comfortably work at that level, so I completed the course using the middle level instead. I don't feel I missed what I wanted from the course because of that.

If you already know Python well, the higher level gives you another way to engage with the material. If you don't, I wouldn't let that put you off.

The real value for me was understanding the ideas.

Another excellent Helsinki course

The course design was again one of its strongest features.

The University of Helsinki's online learning environment feels like it has been designed around education first. The explanations are clear, the interactive sections are genuinely useful, and there wasn't the constant commercial pressure I'd experienced on some other learning platforms.

You could complete the course without paying. The certificate itself was paid, so I didn't buy one. I simply kept the completion screen instead.

That approach actually makes me more likely to use Helsinki again. If I eventually pay for a more advanced course, they're one of the first places I'd look.

Who I'd recommend it to

Building AI is a natural follow-on from Elements of AI.

Someone who only wants a very broad understanding of AI could probably stop after the first course. But if you've enjoyed Elements of AI enough to complete it, there's a good chance you'll want to keep going anyway.

I think I spent roughly 40 hours over four or five weeks on Building AI. It may have had a recommended study time closer to 50 hours, but either way it felt like time well spent.

What it won't do is show you how to build an AI application next week.

Like Elements of AI, it has that university approach to learning. It's more interested in giving you useful concepts and ways of thinking than walking you through one specific product or framework.

For me, that was exactly what I needed at the time.

After learning some Python, completing Elements of AI and then working through Building AI, the next step finally felt obvious.

I didn't need another introductory course.

I needed to build something.