Machine learning doesn’t mean a machine isactually thinking. Machine learning has more to do with statisticsand pattern recognition than it does about being a thinking robot that’s going to takeover the world. Your insights intobots, big data, and artificial intelligence. So machine learning is becoming more and moreprevalent in the apps you use today.
So anything from voice recognition, driverlesscars, or recommendations that you get on Netflix, or Spotify. Machine learning picks up on patterns in data,and then it makes predictions based on those patterns. Let me give you an example. LEGO man, brick, LEGO man, brick, LEGO man…whatdo you think will come next? Brick? Yup! You’ve just looked at a pattern and made aprediction based on that pattern.
A machine learning algorithm would do reallywell with this as well. Except it would do a lot better with somethingthat we could not even realize is a pattern. Lots and lots of data, with lots and lotsof different types of LEGO pieces. It could find a common thread where we couldn’t.
So why is it considered a part of artificialintelligence? Well, it is because Machine Learning learnsvery similar to the way we do. Think about children. A parent reads a book to their child and pointsout a dog in the book. this is us teaching a baby what a dog is. Later mom and child are watching a cartoon,and mom points out a cartoon pup and says to the baby “hey, that’s a dog”. Child and grandparent are walking down thestreet and they see a dog. Grandparent says “that’s a dog”.
Our brains will see a new dog, and even thoughwe’ve never that breed of dog before we’ll still know that it’s a dog. Why? Because we’ve seen enough patterns to makean assumption and a prediction. With machine learning, we teach machines ina very similar way. Instead of teaching the machine learning modelwe call it training.
We train the machine learning model. So we give the machine learning model thousands and thousands of pictures of dogs. The machine learning model picks up on patternsin the photos of dogs. Then when we present it with a new picture,it understands that that too is a dog.
What we don’t tell the machine learning modelis “look out for two eyes, two ears, a nose, a tail, long hair, short hair”. That type of teaching for a machine is whatwe know as an algorithm. Three things I want you to note about MachineLearning. First. Machine learning actually still requires alot of human effort. We need human labelled data for the machineto interpret. So we need to feed it images that humans haveidentified is in fact a dog.
At the second part of machine learning, weneed to see what its outputs are, and we need to validate if it’s right or wrong. Just like if a baby points at a cat and says”hey this is a dog” we need a human there to say “no child”. We need humans. And there’s a whole industry around humanslabelling data for machine learning. Like this company, or this company, or thiscompany. Their whole purpose is to have humans annotatedata and say “this is a stop sign, this is a truck” so that driverless cars can thenlearn to recognize things on the street.
The second thing to note about machine learningis that machines can be quite biased. So remember I talked about humans feedingdata to the machines? Well if humans input false or biased data,the machine is going to spit very biased data out. Amazon had this problem. Amazon taught a machine learning model toreview all the resumes of people that apply to amazon jobs. Problem being, is that humans were very biasedin choosing some resumes over others.
So the machine, what happened? It amplified the bias. And three. Last thing I want you to remember about machine learning is that it’s still a very nascent industry. A lot of companies are taking the lead in machine learning. Specifically, the ones that have lots of data- because you need data for machine learning. So this includes, Google, Amazon….but machine learning right now isn’t a sentient thinking machine. It’s just looking at patterns and making predictions.