How Does Machine Learning Work
The most significant element of Artificial Intelligence (AI) is “Learning”.
Before a machine can have AI capability, it needs to learn — just as human beings can only become more intelligent through constantly learning new things.
Therefore, what machine learning is doing is basically trying to mimic human intelligence through learning.
The next question to ask is:
“Where does the machine learn from?”
While human beings look for our own resources (books, internet, experience, etc) to learn, a machine needs human beings to feed it with data for it to learn. This data is called the training dataset that is provided to the computer for it to start training itself.
ChatGPT, an AI Large Language Model (LLM), stands for Chat Generative Pre-trained Transformer. The word “Pre-trained” tells us that all AI models need to be trained before it can perform what it is intended to do.
The next question to ask is:
“How does the machine learn?”
We human beings write the algorithms that tell the machine how to learn from the dataset given to it. An algorithm is a set of step-by-step instructions on how to perform a specific task to solve a problem — in this instance the machine learning algorithm of how to learn from the dataset.
These machine learning algorithms are able to adapt and improve themselves during the course of their learning process.
There are three main categories of machine learning algorithms:
Supervised learning — In supervised learning, the machine learns “under supervision” using labeled datasets. Labeled datasets refers to data that are already known and categorized — for example datasets labeled as house, tree, car, dog, etc. Lots of sample data are fed to the machine for it to learn from.
Unsupervised learning — In unsupervised learning, the machine learns by itself with “no supervision”. It learns through pattern recognition using unlabeled datasets. Unlabeled datasets refer to data that are unknown and the machine tries to identify them through their patterns and structures.
Reinforcement learning — In reinforcement learning, there are also no labeled datasets and the machine mimics how human beings learn — through getting positive or negative feedback. It goes through a trial and error learning process. Errors help in learning through a “reward or penalty” system.
In all the above machine learning algorithms, the objective is to learn from the input (training) datasets so as to be able to “predict” the results when presented with a new set of data.
A newer subset of machine learning called “deep learning” goes further in its capabilities by using more sophisticated artificial neural network algorithms. Artificial neural networks are modeled after the biological neural networks of human brains.
Our human brain consists of billions of neurons and the biological activity of neurons is carried into the concept behind “deep learning” AI models. Neurons are like information messengers — they transmit information to one another.
The equivalent of the human brain’s neurons in deep learning algorithms are called “nodes”. A node is a data point within an AI neural network that can have multiple complex input and output connections to and from it.
Modeling after the human brain’s neurons, nodes are interconnected to make up a huge and sophisticated neural network architecture.
Artificial intelligence has been applied in speech recognition, self-driving cars, face recognition, computer vision, language translation, natural language processing, algorithmic trading, fraud detection, robotics, etc.
While there are dangers of AI, we must always remember that all technologies are subject to misuse and AI is no exception. Be vigilant and at the same time let our moral compass guide us in all our endeavors.
Cheers everyone!
The joy of writing is such that I can share and disseminate what I know. I am also learning what I don’t know. And I realized that there is so much that I still don’t know and I want to know.