Machine learning has a great amount of meanings to different people. It has been around for years and now it is getting the attention it deserves. The technical definition of machine learning is “the concept that a computer program can learn and adapt to new data without human interference. Machine learning is a field of artificial intelligence that keeps a computer’s built-in algorithms current regardless of changes in the worldwide economy.” Many people think of artificial intelligence when they hear machine learning, and those people are not wrong but they are not right either. Essentially machine learning entails programming but instead of using code, one uses images and videos to build a model. A training set is selected, and the programmer chooses what imitates a positive or a negative for that specific set. Then they assign a specific model to use. The data one works with effects the model type and the size and quality of the sets. Additionally, it impacts the outcome. The process and technology itself is extremely complex and if something does not work correctly, the operation begins all over again. On the other hand, Artificial Intelligence depends on feedback as it interacts with the world and continuously adapts to new changes. There are layers to the overall aspect of Artificial Intelligence, which can easily be explained by this graphic:
Each level is a subset of the one before it, meaning one cannot properly function without the other. Deep learning carries out the machine learning process using an artificial neural net that is composed of levels arranged in a hierarchy. The particular network will learn something simple at the primary level, and then it sends the data to the following level. At the next level, the simple information is mixed with more data and becomes a little more complex. Then, it is sent to the third level. This process repeats itself at every level as the hierarchy becomes more and more complicated. Deep learning is essential for machines to be able to learn without a human labeling every single type of information that comes in, which is supervised learning. For instance, an endless amount of personal data is collected from social media accounts, hardware and software service agreements, app permissions and cookies. Businesses use this information for many reasons and it can be extremely valued. Not every element of the data is labeled which means it cannot be used to teach the machine learning programs that depend on supervised learning. To have a programmer oversee every data set that comes through would be extremely time consuming and expensive. Luckily, deep learning can help. It excels at unsupervised learning where the data is not labeled. Overtime, with enough information, deep learning can determine what is what, which helps machines learn without a human programmer.
Shockingly enough a children’s animation movie illustrated the abilities a robot could possess through machine learning. The Incredibles, a Pixar movie introduced in 2004, provides a good visualization of how machine learning and artificial intelligence work and the speed at which it processes real time information. Over time, it learns how Mr. Incredible fights and is able to determine ways to beat him. The villain, Syndrome, designed these killer robots, The Omnidroid, to eventually eliminate Mr. Incredible. However, over the years he put his prototypes up against other Supers in order to perfect his creation. As Mr. Incredible goes up against The Omnidroid the first time, it learned how he moved and predicted what he would do next. By doing so, it was able to delay Mr. Incredible’s victory. Even in 2004, moviemakers implemented this advanced technology in a children’s movie in order to show its future capabilities. A machine being able to learn in real time. As you watch the video below, pay attention to how Mr. Incredible initially jumps over The Omnidroid and how it then predicts him to jump over again.
Google has successfully combined deep and machine learning into a new program called AlphaGo. Go is an ancient Chinese board game which is allegedly the most complex and subtle of all board games. This is not as typical as chess, where a computer can calculate every possible move thousands times fast than a human. In Go, the possible moves are endless, which means no computer can handle trying. However, researchers in the UK were able to teach a computer program to play Go, but it also has the ability to learn from its mistakes. It used deep and machine learn to study which moves left it vulnerable and which ones lead to victory. In the beginning, there were a lot of mistakes made, but this was a good thing because the program was learning. The more it played against humans, the more it learned. As time went on, it played against more skilled players, and in addition with those games, it played matches against itself. When the researchers put AlphaGo up against itself, the program became much better as it was able to compete against a player at the same skill level. Eventually they put AlphaGo against the number one player in the world, Ke Jie, and beat him in five games out of five. After that, there was no one else left to beat in the world. AlphaGo was able to learn from itself and become the greatest Go player in the world, and it is just a computer program. If it is already the best, how much better could it get? AlphaGo Zero.
The opportunities to use machine learning are endless and highly effective. If in 2004 a children’s movie can display its basic capabilities then it is astounding to see how far it has come since then in the real world. The news of AlphaGo and AlphaGo Zero are surprising because it was not expected that this technology would advance this rapidly. These machine learning programs is truly a breakthrough breakthrough and it will continue to improve in the future.