In 1937, when computers were first introduced to the world, no one knew the full potential of their abilities. The more we understood the way computers functioned the smarter it would get. We utilized computers to store data, have a way to communicate just like how you can read this right now, etc. Only half of the world has access to this thing called the internet.
Fast forward to 2020, the algorithms in these computers that we have are now sometimes unexplainable. Why is that? This is because we have made the computer/AI algorithm smarter than we can imagine. When programming an algorithm to give you recommendations on what to watch next on YouTube, that is AI learning and paying attention to what you are interested in based on what you are watching.
The founders of this field Minsky and McCarthy describe AI as any task performed by a program or a machine, if a human were to do the same, we would say the human had applied intelligence to finish the task. Now, this is a very simple definition of what AI is. The goal is to create a model that answers our questions correctly most of the time.
AI systems are typically capable of having some behaviors we have as humans, such as planning, learning, problem-solving, etc. There are many implementations of AI and can be found almost anywhere you go, the device you are using right now to read this has an AI in it; the ability for google to understand you is called natural language processing, and when spam emails are separated that is an unsupervised model, and many more examples.
Machine learning is how AI can be trained to do a certain task. Machine learning is where large amounts of data are put into the computer system to perform the task that you want it to. You train the computer before it is used. Just as a professional athlete like Usain Bolt is trained to run the specific event he does. To become better at what they do they are trained for a duration of time, it is not something you can master in a night. The same goes for a computer, it has to be trained with a humongous amount of data.
There are many subcategories in machine learning. These algorithms are used for different tasks and answering questions that we are not able to solve or not notice. Three popular types of learning are Unsupervised Learning, Supervised Learning, and Reinforcement Learning.
Unsupervised Learning
When separating coloured cubes into different groups you have to make groups based on the colour because it is an attribute that sets them apart and can be defined as a group. This is an example of what an AI can do if large amounts of data are given and no parameters were set. The model will measure similarities to group data together and group them together into different categories.
This can be a problem at times because it will sort them one way and not actually haven’t quite worked. It can do a job best when you set parameters and have the number of groups wanted, this is called K-means. K–Means is an algorithm that segments data into clusters to study similarities. This includes information on customer behavior, which can be used for targeted marketing. The system looks at similarities between observations (for example, customers) and establishes a centroid, which is the center of a cluster
Supervised Learning
This is an algorithm used when there is enough data to put in and work in functions. Supervised learning can work well with enough data that have labels. It is mainly used to identify objects, recognize gestures or a voice or clarify spam emails.
The machine trains on a data set to recognize something. For instance, you want the machine to differentiate a huge amount of animal pictures to recognize a cat and a dog. The information the machine has is how the inferences and how well the job will get done. The more information the better. If you want the machine to improve and imitate actions we can do, the algorithm never gets better.
In Supervised Learning there can be cases where there is over-fitting, which means that the function you are training is perfect and when something unknown and new is put into the program it crashes, putting it in a random spot. This is an algorithm that is used to make predictions and identify things and when you overdo it, the system becomes worse, making it brittle.