Skip to main content

AlphaGo AI


The AI, AlphaGo was created by scientists looking to improve artificial intelligence. Its purpose was to independently learn the ways of the complex Chinese board game, Go, from absolutely no knowledge of it at all. The AI ended up defeating the best Go players in the world with ease. A new version of the program, called AlphaGo Zero, faced off against its original self in 100 games of Go. The new program went undefeated, winning 100-0.

Image result for alphago defeats lee
Pictured: AlphaGo versus Lee Sedol

Image result for go (game)
Pictured: the ancient Chinese game of Go

AlphaGo observed and analyzed games of Go, then played against itself as well to improve. It used and analyzed data gathered from human games to learn. AlphaGo Zero eliminated the human data, instead learning and gathering data from playing games against itself. AlphaGo Zero was given only the rules of the game, and taught itself by playing games against itself, beginning with completely random moves and eventually making well thought-out, efficient moves. Zero is better than the past version in every way. Zero’s predecessor took 30 million training games and several months to learn all it knows. Zero only played 4.9 million training games in an astonishing three days to achieve the same level of knowledge the original had.

AlphaGo had two individual neural networks that would work to determine the best moves to make based on data gathered from human players. The moves would then be tested by playing self-games to  chose the most effective move to make. AlphaGo Zero had one combined neural network with a simpler search algorithm. It did not rely on human data and did not play testing games against itself. This updated network allowed the AlphaGo Zero AI to perform new, never-before-seen moves that “are now redefining how Go is played.” 

Although the program had to play many more games to reach its level of expertise than the number of games a human would have to play to become an expert, the AI reached a more adept level much quicker than it would have taken a person.

Scientists could keep updating/modifying this kind of technology and apply it to other situations besides board games. Application to more complex things without strictly set rules/guidelines could lead to great new scientific breakthroughs, discoveries, and inventions. The self-learning technology can quickly advance knowledge and learn at a fast pace. If these programs are left to learn new things on their own, they could be capable of figuring/discovering that humans had not thought of.
One concern of mine is that if this artificial intelligence technology becomes more advanced and a regular part of our daily lives, it would be taking over human purpose. It's mental efficiency would transcend that of a mere humans, think for itself, and then what? 
 

Comments

Popular posts from this blog

Physics of Black Holes...Or Lack Thereof

Isabella Jacavone To comprehend how the universe works, we must dwell into the most basic building blocks of existence; matter, energy, space, and time. NASA's  Physics of the Cosmos program involves cosmology, astrophysics, and fundamental physics intended to answer questions about the elusiveness of complex concepts such as black holes, neutron stars, dark energy, and gravitational waves. In this blog post, I'd like to elaborate on a subject that is very intriguing  to me; Black holes. And more specifically, what would happen if we got near one. A black hole is anything but a hole, but rather an immense amount of matter compacted into an extremely small area. A black hole is caused when, hypothetically, a star four times more massive than our sun collapses into a sphere no bigger than 600 square km. To put that in perspective, that's about the size of New York City. B lack holes were predicted by Einstein's theory of general relativity, which showed that when a

Physics Behind Drone Flight

A drone flies by using its downward thrust and forcing air in a particular direction in order to sustain a certain speed as well as a specific height. In this video my friend and I had been flying a drone at exactly 4 mph which converts to 1.788 m/s. In this project, we will be determining the forces acting upon the drone in order to sustain a consistent flight in terms of velocity and height while excluding the effects of air resitance. The drone is flying at an angle of 28˚, this is found by extending the tilted axis of the drone to the horizontal and finding the angle with a protractor. From this angle we will be able to calculate the downward thrust and the acceleration of the drone that allows it to maintain its height and velocity during flight. When the mass of the drone is taken it results in 734 grams or .734 kilograms, which will also be used for the calculations within the project. The freebody diagram pictured above will alow us to derive the force equations f

Aerodynamics of a Golf Ball

One may wonder how a small golf ball can travel at incredibly high speeds for such long distances.  While the swing of the club is a major component, the structure of the golf ball is quite important.  Unlike a baseball or tennis ball, a golf ball has dimples all over it (usually 336 dimples).  These dimples allow the golf ball to travel without facing much air resistance.  This diagram shows how air travels around the golf ball. The dimples on the golf ball also prevent drag that would occur in the wake region, resulting in further distance.  Also due to the contact with the club during the swing, the golf ball has backspin during its entire flight.  This diagram shows the motion of the golf ball mid flight with the lift force of F. There are hundreds of different types of golf balls that a player can choose.  Some show little affect to a player's game while others can alter their performance completely.  Personally, I prefer Callaway Supersoft golf balls, but it is entirely