From virtual chess players to self-driving cars, machine learning has become a very important part of society, and it's growing quickly. Today, demand for human programmers is growing so fast that they've figured out how to get their computers to teach themselves with machine learning.
Before continuing, it's important to note that machine learning is an ever-advancing field and there are many ways to achieve it. For the sake of simplicity, I'll be covering only the most successful and popular types of machine learning. Through understanding what machine learning is and how it can be used, we'll talk about the basics of artificial intelligence.
According to the definition of machine learning, a computer that “learns” is able to access data to deepen their understanding of a topic. One of the most accurate and popular ways to have a machine learn is called deep learning, or deep neural networks. These systems consist of three parts: inputs, hidden layers, and outputs. Between every two parts, there are connections with different weights, and these weights decipher which inputs should be seen as more important and which should be ignored.
In the beginning, these weights are randomly generated and the computer has no idea how to achieve any task, creating random answers to our questions. This is where the training comes in. Like our brains, computers need to be taught to understand what they are doing. A popular way to train a computer is a process called Back Propagation. Our computer will run our program, set our inputs, and give us its output result. Now we can give the program the correct answers to check its work. The computer’s error is calculated by subtracting its output results from the given answers, then the computer slightly adjusts the weights of the connections, generating results that become more and more correct. This process occurs thousands of times over seconds, and a computer can learn to complete a task with little to no error very rapidly.
Now, that we understand how machine learning works, what can it achieve? While computers can be trained to solve many problems, a computer can rarely complete multiple jobs with a single program. Instead, programmers design their machine learning models based on the problems they are looking to solve. Machine learning can achieve a variety of objectives. For example, Google created a machine learning program that could compose music. Other machine learning programs can recognize objects in images; this is how self-driving cars work. Some play video games, write their own programs, conduct research alongside humans, and much, much more. One of the most important advancements in machine learning was when a project created by Elon Musk's company, OpenAI, played the game DOTA 2 against a champion and won. This game is said to be one of the most complex strategy games ever created!
So yes, machine learning can educate itself in a matter of seconds, change its own code to do whatever it wants, and can beat human players at one of the world's most complex games, but what does this mean for the future of artificial intelligence? Based on opinions from many developers, including Musk, if we don’t take action to properly teach our machine learning projects, our future will quickly get out of hand. People are working to guide the future of AI, and are learning how to control machine learning algorithms.
Research is being done to further understand what more complex learning systems are, and what they're fully capable of doing. As Bill Gates said during his Reddit AMA, "First the machines will do a lot of jobs for us and not be super-intelligent. That should be positive if we manage it well. A few decades after that, though, their intelligence will be strong enough to be a concern." Although today's machine learning programs can learn and create new ways of thinking on a variety of topics, they are still more like supercharged Excel spreadsheets than thinkers.
It's amazing to see young students with such a thoughtful perspective on AI and Machine Learning. We're optimistic that the interest and passion of students like Matt will ensure a future where human and artificial intelligences work together to empower amazing outcomes.To learn about some of the most recent developments we've been working on in AI, take a look at the Microsoft Research Blog.
You can find all of our edtech tools and training at microsoft.ca/education. Access free courses and resources through the Microsoft Educator Network. If you'd like to access free training at your school, you can request a session with one of our trainers or Microsoft Learning Consultants.