Over the last few years, I am sure that you have heard the words “machine learning” thrown around whenever AI, or any topic related to AI was brought up. Although on a deep level this can get extremely complicated, I am here to break down the key elements of the topic so that you won’t be left in the dark when this and other words like it are brought up.
WHAT IS IT
Contrary to what you may currently believe, machine learning is not artificial intelligence, but rather a subset of it.
Machine learning allows a computer system to learn from tasks without being explicitly programmed to do so.
Once trained, the model moves on by itself without real outside assistance.
This is very different from how traditional coding worked, with the program repeating the same process each time, and if better performance was needed, the code would have to be revisited.
HOW IT ACTUALLY WORKS
The process of how machine learning works at a glance is surprisingly straightforward. The model is given multiple datasets containing images, or data points to train on, and through these images, it picks up on various patterns and then makes informed decisions later based on these patterns.
The simplest example of this could be something like an AI designed to decipher whether there is a dog in an image or not.
To start, the model would be given thousands, if not tens of thousands, of images. Through these, it would pick up patterns about what a dog should look like. Note that these patterns won’t necessarily always be the same as what we agree on. If all the pictures with dogs were in fields, the AI would assign pictures with a green background to be dogs.
This is something that we can clearly see is not true without thinking twice, but for the program, it is incredibly difficult, if not impossible, to properly find these flaws, as it does not have a concept of what a dog really is.
TYPES OF MACHINE LEARNING
There are three main types of machine learning,
- Unsupervised learning
- Supervised learning
- Reinforcement learning
The example I gave above about the dog was an example of Supervised learning. Giving a set of labelled data and allowing the model to identify patterns so that it can make predictions on new data.
If the datasets were not labelled right and wrong, with the AI meant to group them using patterns it observed between similar images with no prior information, this would be considered unsupervised learning.
Reinforcement learning is quite different from the rest and would not fit very well in the example above. This form of machine learning gives the AI direct feedback on each decision that it makes with a score or other ranking system. This is the type of machine learning that trained AI to play chess and Go at an extremely high level, beating the top players in the world.
This is the most exciting form of machine learning that you will see in the news since it is by far the most flamboyant.
The other forms of machine learning are still very important, dealing with issues like spam detection and recommendation algorithms (in conjunction with reinforcement), but rarely make headlines.
REAL WORLD APPLICATIONS
There are countless real-world applications of machine learning, and this number is growing every day.
Some of the major uses include, but are not limited to,
- AI chatbots such as Claude, ChatGPT, Gemini
- AI overviews on sites like Google
- Social media feed curation
- Autocorrect and predictive text
Machine learning is no longer a concept reserved for computer scientists and politicians. To fully understand the way that the world is moving, you must first understand key underlying concepts like these, even at a surface level, so that you can be well informed about the technology shaping the world.
