Machine Learning is a subset of AI, and is an approach to try to achieve… Oh really, another fancy word and long descriptions?
Let’s make it simple!
I will start with Artificial Intelligence, and then move to Machine Learning. Why so? Wait for it! The Wikipedia definition for AI: “Artificial intelligence (AI) is the intelligence of machines or software, as opposed to the intelligence of humans or animals.”.
By definition, it’s all about software, and machines, and a machine requires some kind of code that we can execute. When we write code we are telling a computer what to do.
When we code a program/application, we say we are programming. Programming is simply giving input values/data to a function/method and expecting output values/data.
What can we do in programming with and without Machine Learning?
- Without ML.
- How to get from location A to Location B. For example: find a route to my house from my current location. Easy to code, “simple if/else” statements.
- With ML.
- How to recognize something or someone? For example: Is this person angry? Find a cat in the picture. Not that easy to code. Requires a lot of input and output examples.
Programming without Machine Learning
Often our programs are procedural. That is we have IF this THEN do that, we use this type of conditions, we tell the computer what to do. As programmers, we write the function that takes an input, and all the computer does is actually just calculate output using the function, and the input.
Example
As an input let’s provide a list of numbers [1, 2, 3], we give it to a function and we get an output of [2, 4, 6].
So clearly this function multiplies our list items by 2.
It is easy to tell a computer how to find a route to my house from my current location, isn’t it? We need GPS coordinates for my current location and GPS coordinates for my house location. Using a well-known method named triangulation – https://en.wikipedia.org/wiki/Triangulation. Of course, in the cities, we have many roads, crossings, etc., and we want to find the shortest path – without walking through the buildings – in this case, there is a well-known algorithm for that called “Dijkstra’s algorithm” – https://en.wikipedia.org/wiki/Dijkstra%27s_algorithm, and many others https://en.wikipedia.org/wiki/Shortest_path_problem.
Programming with Machine Learning
Machine Learning changes the above idea about programming. Using ML we say, give me the input and give me what you want the output to be. Give me all the input data that you have, and tell me what is the output that you want. As developers, we give this input and output, and with Machine Learning, we say, machine, you figure out what this function is. We don’t know how to tell a machine how to do that. You, machine figure it out yourself. We are just training the machine to create that function. To do it well a machine needs lots and lots of data.
With ML the function is generated or created by the machine. Instead of us writing the function, as we have before, with Machine Learning, the machine writes the function. This is Machine Learning. The word function can be replaced with a model. Here comes the mighty ML model – it’s a function that receives an input and returns an output. Another word might be an algorithm, an ML algorithm, or an ML bot. The problem with Machine Learning Models is that these functions are really hard to understand because the computer comes up with these functions and knows how they work.
Example
As an input let’s provide a hundreds images that may or may not have cats in it, and as an output let’s provide fifty cat images.
With Machine Learning, the machine writes the function that finds a cat.
But to tell my computer what a cat is. Is almost impossible, right? Because there are so many factors. A cat has legs and a tail, but a dog has legs and a tail as well. A cat has fur, but then the computer is going to ask me, what is fur? But with machine learning, we can actually teach a computer to know what a cat is. Or at least pretend like it knows.
At the end, I will finish the sentence from the very beginning of this article. Machine Learning is a subset of AI, and is an approach to try to achieve Artificial Intelligence through systems that can find patterns in a set of data. The Stanford University “definition” – AI is the science of getting computers to act without being explicitly programmed.
NOTE: In this article, I’m just barely scratching the surface. This topic needs more reading and research on your own. As a good starting point, readers may check a few useful resources in my other article – How did my journey with AI & ML start?.
Leave a Reply