When you log into your streaming service, have you ever wondered how the service seems to know exactly what you want to watch? Or why it suggests shows and movies you wouldn’t have picked on your own but are still perfect for you? It’s all thanks to machine learning.
Machine learning is training computer models to predict outcomes. This happens through a combination of math, science, and data. Because you’re not telling the machine what the right answer is, you have to teach it how to infer what the right answer is, which takes time, experimentation, and a lot of data.
Machine Learning Definition
Machine learning is the intersection of computer science and data science. Using data, algorithms, and statistics, engineers “teach” a computer how to learn. In this case, the machine is learning how to identify patterns in vast amounts of data.
This training is similar to the way humans learn. When you’re trying to master something new, you make mistakes. But over time, you learn from those mistakes and improve. Machine learning is similar in that, over time, the machine learns and improves its accuracy.
>>MORE: We Asked ChatGPT to Write Resumes and Cover Letters. Here’s What It Got Right (and Wrong)
But unlike a human, the machine isn’t thinking. It can’t generate new concepts and ideas. What it’s really doing is making classifications and predictions based on the information it already has. Here’s a simple example of how this operates in real life.
Data Science
Try your hand as a data scientist in this free course from British Airways. You'll scrape data to uncover customer insights and make predictions.
Avg. Time: 3-4 hours
Skills you’ll build: Python, web scraping, machine learning, data visulization
Your streaming service has probably made suggestions that look something like “People who watched X have also watched…” or, “Suggested for you…” The service makes these recommendations based on the information gathered from two data points.
The first is what others have watched. For example, if you watch a ton of sci-fi, the service makes your suggestions based on what other sci-fi aficionados watch. The second data point is what you do with that suggestion. Whether or not you watch the show, the machine integrates that information.
Over time, it learns you’re more of a Star Wars than Star Trek fan, tailors the suggestions accordingly, and offers content that a Star Wars fan is more likely to choose. In a sense, you’re acting as a machine learning engineer, teaching and training your streamer’s model to help it improve its accuracy for you and other viewers.
Is Machine Learning AI?
Artificial intelligence (AI) and machine learning are similar but not the same.
Think of artificial intelligence as the umbrella term for all things related to teaching machines to be more like humans. The idea is that one day humans will create machines that act, think, and reason just like we do.
Machine learning is under the AI umbrella in that machine learning teaches an algorithm how to analyze data and find patterns without help from humans. However, in machine learning, there’s no attempt to teach the model how to draw original conclusions beyond identifying patterns.
Artificial Intelligence
See what it's like to work in artificial intelligence in this free course from Cognizant. You'll conduct exploratory data analysis, build a model, and interpret the results.
Avg. Time: 3-4 hours
Skills you’ll build: Data analysis, machine learning, model interpretation, data modeling, quality assurance
How Does Machine Learning Work?
Machine learning starts with training the model using data sets. The machine examines and compares various data sets to determine what the correct output is.
Let’s go back to the streaming service. How does it decide what the “right” suggestion is for your next binge-fest?
It starts with training. To keep things simple, we’ll consider only movies. The “movie” category is one data set: movies. It’s pretty big, though, so the movies are in genres: horror, rom-coms, thrillers, and so on. Those are additional data sets. Then, you’ve got the viewing habits of everyone who uses the movie category. Who watches only rom-coms, who watches only documentaries, and who bounces around genres? This is yet another data set.
An engineer feeds all this data to the model and lets it analyze the information to identify patterns. For example, the machine may find that most people who watch horror movies never watch rom-coms but do watch thrillers. It could find that people who watch sci-fi movies almost always choose a comedy the next time they stream.
From these patterns, the machine learns and starts predicting that if you watch horror movies, you’re more likely to watch the horror or thriller movie it suggests than the rom-com (unless it includes horror elements!).
Types of Machine Learning
So, that’s how machine learning works. However, just like there are different teaching methods, there are different training methods. The three main types of machine learning — supervised, unsupervised, and reinforced — all have the same end goal, to train the machine to make correct predictions.
Supervised Machine Learning
In supervised learning, the machine’s algorithms are trained using labeled data sets. Labeled data is data that is labeled or tagged by a human, which makes it easier to find patterns. One common example is your email. Messages you mark as spam help the program learn what to do with your incoming messages. Over time, the machine should learn that when the prince of a foreign country emails you about a loan, it’s probably spam and automatically sorts it into that folder.
In the case of our streaming service, movies may be labeled to help the machine understand what elements might appear in which movie. For example, a movie with an alien invasion is probably a sci-fi movie, not a documentary.
Unsupervised Machine Learning
Unsupervised learning means the algorithm analyzes unlabeled data sets, which is exactly what it sounds like. The data doesn’t have any tags or labels, which makes it harder for the model to learn, but also makes it possible to discover patterns and trends the engineers may not have realized existed. For example, the machine may find that people who stream at 2:00 a.m. are more likely to watch documentaries than thrillers.
>>MORE: What Is Data Engineering?
Reinforced Machine Learning
Reinforced machine learning is closer to how humans learn. It starts with labeled data to train the machine and give it a solid understanding of the patterns it should look for. An example might be giving it data on what makes something a sci-fi movie or a thriller.
However, instead of feeding the machine more data to improve the outputs, the machine learns from its success and failures — much like a human. If you continually watched the movies and shows the model suggested, that would be a success and the machine would build on what it’s learned to make similar suggestions. But if you reject all of the suggestions or even some of the suggestions, the machine learns from this and adjusts its suggestions accordingly.
Introduction to Data Science
Experience what it's like to be a data scientist in this free course from Commonwealth Bank. Build practical skills like anonymizing data, documenting processes, and examining APIs.
Avg. Time: 3-4 hours
Skills you’ll build: Data privacy, APIs, database design, unstructured data, data set schema
How Do Companies Use Machine Learning?
But where does all of this lead? Fortunately, it leads to lots of inventions across a variety of fields. Examples include:
- Image analysis: Companies train machines to better analyze images (like being able to tell the difference between a human and a duck).
- Fraud detection: The model learns what your normal spending patterns are and quickly identifies when your credit card has been compromised.
- Chatbots: Sometimes that online conversation with customer service isn’t with a human!
- Medical diagnostics: Machines are being trained to assist doctors with finding medical issues the human doctor might miss.
- Predict the future: Machines are also being trained to predict a probable outcome based on past data, like when a customer is likely to cancel their subscription.
Digital Intelligence
Explore different data jobs in this free course from PwC. You'll learn how to use Python to build a responsible AI, try some quant financial modeling, create predictive models, and more.
Avg. Time: 8-9 hours
Skills you’ll build: Classification modeling, cash flow prediction, large data sets, portfolio valuation.
Machine Learning: The Bottom Line
Though machine learning is relatively new, it’s found its way into much of our lives. Every time you watch a suggested movie, click on a recommended post, or use your credit card, you’re helping train a machine.
Gain a competitive edge in your job search. Enroll in one of Forage’s top data job simulations and sharpen your machine learning skills.
Image credit: Canva
