I have worked on computers -- taking them apart, putting them back together, and everything in between -- for over four years now. Machines, including computers, are created from several constituent parts physically connected by hardware (wires) and integrated by software. The functions of most machines we encounter today are merely the sum of their parts. Computers power on because their power supplies send electricity to their motherboards which power and coordinate the processor, RAM, and hard drive. Operating system files are stored on the now-awake hard drive and the operating system carries out a pre-programmed sequence of commands to boot and generate your desktop screen. From there, every software you open and use, every document you write, every website you visit, is brought to you by pre-programmed code and commands.
By contrast, machine learning, in the words of Arthur Samuel in 1959 gives "computers the ability to learn without being explicitly programmed". In other words, machine learning allows computers and other machines to move past the sum of their parts and perform new, complex functions. Yet again another explanation -- machine learning allows machines to learn from data, generating responses and judgments without being directly programmed to do so. In movies, machine learning has been hyperbolized to the Nth degree, like in Spike Jonze's Her (2013), in which an 'operating system' gradually learns more and more about serving its owner, to the point where they fall in love.
Outside of science fiction, however, machine learning is very real and impacts our daily lives. This blog post will outline a couple of instances of machine learning that affect your life today!
With Drover being a green transportation company, I've gotta start by featuring Tesla's machine learning innovations. Widely introduced in 2015 and nearly perfected by fall 2016, Tesla Autopilot allows the Model S and Model X to drive without direct human input. For legal reasons, they recommend that drivers still keep their hands on the wheel, but I'm pretty confident most people with self-driving cars take that fine print loosely.
Tesla Autopilot works through the combined action of hardware and software. Cameras and sensors all over the car observe data about the vehicle's surroundings, including proximity of nearby vehicles and obstacles. Software in the onboard computer takes observed data and makes judgments about speed, braking, and defensive maneuvering. For example, if the Tesla senses the vehicle in front of you slowing down, it will apply the brakes itself and match the lead car's new velocity. Now, everything I have described is more or less directly programmed into the car.
Machine learning comes in as your Tesla's Autopilot ages. As Autopilot logs driving miles, it grows and learns from past experiences. For example, if you take the same commute from home to work every day, that drive on Autopilot will become increasingly smooth and the car will become increasingly decisive about decisions along that route as it learns about the obstacles, turns, and nuances of your most common route. This is machine learning at its finest -- extrapolating beyond observed data to self-correct itself beyond the capacity of its explicit programming.
Netflix also implements machine learning in its recommended movies feature. At the most basic level, Netflix provides a library of movies and TV shows for you to stream for the cost of their monthly subscription. You interact with their interface on your computer, phone, TV, Xbox, etc, and select movies and TV shows to watch. Netflix delivers this content to you through instant streaming. Everything described is explicitly coded by people into Netflix's programming.
Beyond this, however, Netflix generates "Top Picks for You" suggestions... these suggestions are brought to you by machine learning algorithms. Netflix's programming observes your viewing preferences and makes recommendations for similar content that their machine algorithms calculate you will enjoy. There is not a human being generating these lists for us, they are generated by the power of Netflix's machine learning algorithms. Again, this is an example of Netflix taking given data (what you watch) and then inferring and making judgements that go beyond its explicit programming.
My last example is Amazon -- the online retail behemoth and one of my favorite companies in the history of ever. Amazon's website is designed to provide access to the company's practically infinite inventory of everything you could possibly imagine. Your process from typing an item into the search bar, to selecting an item, to checking out from your cart, is all part of Amazon's explicit coding.
Machine learning kicks in however with Amazon's recommendations.
Similar to Netflix's recommendations, Amazon first collects data on every item I buy, every item I look for, and everything I search for in their inventory. Then, machine learning algorithms take that raw data and extrapolate shopping suggestions and similar or complementary items that I may be interested. Again, this goes well beyond any lines of code directly entered by a human being. Amazon's programming is able to learn about me and my shopping patterns and then give me customized feedback and recommendations.
Though this article just scratches the surface on machine learning, I hope you feel you have gained a deeper understanding of what machine learning is without me getting too technical. Additionally, I hope you learned something new or exciting about some of these very common examples of machine learning in our daily lives!