When you learned to ride a bike, you were terrible at the start.
Then your Mum told you had to balance a bit to the left you got better but you weren’t good.
A few more goes and you could ride without your training wheels.
After riding a bike 1,000 times it’s harder for you to fall over than not.
Machine learning is getting a computer to do things without explicitly telling it to do so. With things being finding patterns in data.
With lots of data, machine learning algorithms are like you riding a bike for the 1,001st time. Really good.
Without it, they’re like you riding a bike with training wheels for the first time, not so good.
How do they find the patterns? What’s data?
Data can be any kind of information. Pictures, text, numbers. But computers work best with numbers.
Machine learning algorithms find patterns by turning everything into numbers.
A picture is a combination of different pixels each with different colours. A pixel of value 255, 255, 255 is white, 0, 0, 0 is black.
To tell if a dog is an image, a machine learning algorithm processes the pixel numbers of images with dogs and images without dogs and remembers the differences. If the next image it sees numbers are closer to the ones with dogs, it will say there’s a dog in there.
The important thing to remember is the lack of context, you can’t ask the algorithm where the dog is. It will only tell you there is a dog.
Text can be turned into numbers too. If the word dog is 1 and the word pet is 0, the word car might be 75. Why? Because dog and pet are used in more similar context than dog and car.
This is how your email blocks spam automatically. If a new email gets converted to numbers and it looks like other spam email, converted to numbers, the new email will be classified as spam.
There are many more machine learning techniques I’ve skipped over but this is a good start.
If someone has never heard of it before, most people are visual learners.
Paint a picture for them.