This might be kinda boring.
When we all hear about Machine Learning these days, we first think of AI. Which is an understandable misconception, because the two have become almost impossible to separate in everyday conversations.
But from the basic understanding I got from my MSc, Machine Learning is essentially about teaching computers to learn patterns from data and use those patterns to make predictions or decisions.
It is one of the foundations behind many of the AI systems we use today. Whether it is predicting the next word or token, recommending something we might like, recognising an image, or trying to predict what might happen next, there is a lot more going on underneath, but that is one of the fundamentals.
And this is where I realised something interesting.
There are actually many steps before you get to the exciting part of training a model.
The first one is data.
And data is not always this beautiful, clean spreadsheet that is ready to be fed into a model.
Depending on where it comes from and what you intend to use it for, it can be messy, incomplete, inconsistent or only partially cleaned.
We also learnt about different ways of obtaining data, including primary data: data you collect directly and secondary data, where you use data that has already been collected by someone else or obtained from another source.
Then comes cleaning and preparing the data.
Formatting it.
Removing what you don’t need.
Dealing with missing or inconsistent values.
Making sure the data actually makes sense for the problem you’re trying to solve.
And then, after all of that, you can start getting into the part everyone seems to associate with Machine Learning: training the model.
I gave all these summaries because one of the biggest things my MSc taught me about Machine Learning wasn’t actually a specific algorithm.
It was that sometimes the most exciting things have a lot of boring, repeated steps behind them.
And honestly, that changed the way I see the whole AI space.
It is quite similar to software engineering in that there is a lot of work that happens before you get to the final product. But it is also vastly different because of the importance of the data, the repeated experimentation, and the uncertainty involved in the outcome.
I still have fun training models to do things like football predictions and looking at whether we can use data to predict things in the stock market.
But then you run into one very interesting problem:
Humans.
Numbers can be measured.
Patterns can be found.
Historical behaviour can be analysed.
But humans are far from predictable.
A football team can have the “better” statistics and still lose.
A player can suddenly have an incredible game.
A market can react to something nobody expected.
And that is one of the things I find fascinating about Machine Learning.
We are using machines to find patterns in a world that is created and influenced by humans, while humans themselves continue to introduce uncertainty into those patterns.
So maybe Machine Learning isn’t really about making machines smarter than humans.
Maybe it is about enhancing what humans can do with information.
And as the machines get better, I think the interesting part will continue to be how we, as humans, choose to use them.
