Machine Learning is the discipline of training a computer to find patterns in data. This data could be in the form of sensor readings, images, historical stock market results etc. It can look at any data and try to approximate. It does this by mapping inputs to outputs and make a prediction. For instance you could collect all the data on houses in a particular area (number of bedrooms, type of house, garage, garden, driveway etc and price sold) and predict what price a new property on the market might realistically fetch.
Plant Data
Another example might be the best growing environment for a plant: temperature, humidity, light (colour, brightness, length), soil moisture content, sounds, fertiliser etc and compare to growing rate. As you can imagine this requires a lot of data and Machine learning needs lots of data, the more the better but not just quantity but quality. Feeding in poor data gives you poor predictions. This can include unintentional bias.
Neural Networks
So, how does a machine learn from all this data? One of the most common ways is through something called Neural Networks. A neural network is an algorithm written in code (python is commonly used) that can take huge amounts of data inputs and connect the dots to an output. For instance it can look at tens of thousands of pictures of dogs and cats and when it next sees a picture a dog or a cat it can (or should) be able to identify it with a good degree of certainty.
Cats and dogs
This is why it is called an approximation machine. It does its best with the data it has, yet it is still never 100% certain. We, even a 2 year old human, can usually say with 100% certain whether an animal is a cat of a dog. But the computer doesn’t actually know what a cat or a dog is. It has no concept of what a cat or dog is or does. It only has seen pictures of cats and dogs, nothing more. You might be thinking that today, with ChatGPT or Bard it can now talk knowledgeably about cats and dogs.
ChatGPT
Yes it can appear to but even these Large Language Models (LLM) are really just statistical predictors based on reading about cats and dogs on the internet. They (LLMs not cats and dogs) are not real even if you feel as if you are talking to a human who does seem to understand and know more than you do about any given topic. Machine Learning has made substantial improvements over the last couple of years but there is so much more to Machine Learning than ChatGPT or generative art.
Further reading
https://mitsloan.mit.edu/ideas-made-to-matter/machine-learning-explained
