Do you believe in magic? Well, you should cause you are just about to witness some. Allow me....
The simple answer: Machine Learning. Machine Learning is application of Artificial Intelligence that gives computer systems the ability to learn and become intelligent without being explicitly programmed. That’s right, no more old school ‘hard-coding’ a computer to do stuff. Instead, they learn and do it themselves. I am going to demonstrate this idea with one branch of Machine Learning known as Reinforcement Learning.
Reinforcement Learning allows computers to have goals. And this goals are what pushes the computer to get better and better. Let’s think of this goals in form of points, where making the right choices leads to the highest points and the computer obviously aims to get the most points it can get. The algorithm interacts with its environment and corrects and adjusts itself based on its results of interaction; that is, it aims to get better after each interaction. It learns from its mistakes. Just like us in schools, after each fail or pass in a CAT or exam, we adjust ourselves accordingly, sometimes we study more, we go on with the method we’ve been studying with or change the study method altogether.
I’m going to give you two real world examples where machines learned and became really good, even better than humans.
1. Learning to play Atari games.
Remember the early Atari Games, specifically Atari Breakout? Where you had to move a bar of some kind hitting a ball that hits the ceiling that awards you points?
Well, Google’s Deepmind was put to the test, of learning to play the game. The AI is given sensory input and ordered to maximize the points on the screen. Note that no domain knowledge is used. This means that the AI doesn’t know what the controls do or the concept of the ball.
After around 10 minutes, the AI the algorithm tries to hit the ball, but its not yet perfect and it frequently misses the ball.
Two hours later, it’s really good and it gets returns good points, which is amazing but most humans can do this too.
After around four hours, is where the real magic happens. The AI algorithm learns a strategy where it can maximize its points by drilling a whole in the ceiling and letting the ball bounce on the upper red and orange side first since it returns most points. I know, blew my mind too. Here’s an algorithm that had never played Breakout before and had no knowledge of the controls and ball, but it discovers a brilliant way of playing the game that no human had ever thought of, IN JUST FOUR HOURS!!!!!! Dzaaaayuum.
The other example is where AlphaZero AI developed by Google was able to learn Chess in four hours without human help by just playing itself, then proceeded to beat the reigning computer chess program in a 100-game tournament: 28 wins, 72 draws and no losses. If that’s not scary enough, it used moves and tactics never seen before, in the 1500 years history of the game.
Google’s AlphaGo also beat the world champion Lee Sedol, 4-1, again by moves and tactics never seen before. The most interesting fact about Go is that in this game, after the first two moves, there are 10^170 possible moves (please correct me if I’m wrong) more than the number of atoms in the planet, 10^80.
I know, it gets more and more interesting.
Computers learning to play computer games and board games doesn’t sound that magical.
In truth, by use of Reinforcement Learning plus the other forms of Machine Learning: Supervised Learning and Unsupervised Learning, computers are learning to do awesome stuff.
From Virtual Personal Assistants like Siri and Cortana, to Email Spam and Malware Filtering, to Medical Diagnosis like Cancer detection, to Product Recommendation in shopping websites, and Friend Recommendation in social media sites....Artificial Intelligence is really impacting our lives positively.
And as I always say, there is more, much much more.
This is the property of the HalfBloodPrince.
gracieabrams
germansingergirl
2087
yousharknotpass
titsay
blakekathryn
baratheonpilled
discoholicmusic
kianamaiart