Dawn of the Planet of the Robots

Over the spring break, as I was indulging myself with delicious food in Chicago, my heart was following an epic competition between robots and mankind in Seoul, Korea. Initially, I was planning to write about the mysterious rolling rocks in Death Valley for this week’s post. However, upon witnessing the live battles between AlphGo and Lee Sedol, I instead decided to explore one of the most intriguing yet contentious realms of modern science and technology: artificial intelligence.

 

What is artificial intelligence?

Credit: iqworkforce.com

Coined by John McCarthy in 1955, artificial intelligence, or AI, is “the intelligence exhibited by machines or software.” In academia, it is “the study and design intelligent agents” that can respond to environmental changes and take action to maximize the chances of success. AI is one of most fast-growing industries in the 21st century; its research include reasoning, knowledge, learning, communication, recognition, etc. In other words, we strive to create a machine that can think on its own.

Over the past few decades, AI has been able to beat top human players in many different chess games. By relying on an incredibly large database that contains almost every single possible scenario, AI can easily outplay human in many of those abstract strategy games. However, despite the rapid advancement of AI, there was one impregnable fortress that AI failed to break through; it was the game of Go.

 

What is Go?

In the game of Go, players try to control as much territory as possible by surrounding opponent’s pieces. Credit: popsci.com

Also known as an “encircling game”, Go is a popular Eastern board game that originated from ancient China back in the 4th century. Similar to Western chess game, Go has very few rules but demands creative mind, as well as strategic thinking. To win Go, a player needs to surround as much territory as possible on the board. It sounds pretty simple, right? Interesting enough, what separates Go from many other chess games is the unfathomable amount of different circumstances and critical moves. This number is absolutely humongous, approaching 10^765, while the universe is only comprised of 10^80 atoms. Therefore, it is impossible for a computer to just memorize every single step in Go. Meanwhile, although not knowing everything about Go, we human are able to use our intuition and intellectual thinking to our advantage. Because of this, over the years many researchers had claimed that computer programs would never reach the level of top human player at Go, until the advent of AlphaGo in 2016

 

What was AlphaGo?

AlphaGo Logo. Credit: gardinerchess.com.au

Lee Sedol is a professional Go player of 9-dan rank (highest possible in Korea) and is one of the strongest Go players in the world. Credit: gocommentary.com

 

As you might hear from the news, last week, in a series of five matches between humanity and artificial intelligence, AlphaGo, a computer program developed by Google DeepMind, succeeded in defeating legendary Korean 9-dan Go player, Lee Sedol, with a decisive score of 4 to 1. The victory of AlphaGo marked a major milestone for the study of artificial intelligence, as no computer program had been able to master the game of Go before. In game, to everyone’s surprise, AlphaGo played many “unprecedented, creative, and even ‘beautiful’ moves.”  Even Lee Sedol noted, during the press conference after Game 1, that AlphaGo had at the beginning made “one unusual move that no human Go player would have made,” which ultimately secured AI’s victory.

Surprisingly, the secret behind AlphaGo’s immense power was neither strong central processor nor comprehensive database. Rather, AlphaGo was programed to stimulate how human brains work. At first, AlphaGo learned to play Go by relying on a database of around 30 million moves that were used expert players from recorded historical games. After reaching a certain level of proficiency, it was trained further by “being set to play large numbers of games against other instances of itself, suing reinforcement learning to improve its play.” In other words, AlphaGo improved by playing against itself! Furthermore, it implemented “a combination of machine learning and tree search techniques.” By resorting to a “value network” and a “policy network” similar to the basic functioning of human brains, AlphaGo was able to reduce the infinite number of different circumstances involved in the game of Go, thereby selecting the best possible plan of actions that even Lee might overlook.

 

Will we reach technological singularity in the near future?

 

Undoubtedly, throughout history technological breakthrough has always been a major driving force behind many societal changes, and the victory of AlphaGo was just another indication of artificial intelligence entering our daily life. While we are enjoying our iPhones, whose processors are said to be even stronger than the entire NASA’s computational power during the mission to moon, let’s step back and think about the limit of AI for a moment. Will machines become so intelligent that they start progressively redesigning themselves, to a point where we can no longer understand what is going on? This hypothetical situation is called technological singularity, and it is a contentious question.

This graph demonstrates the point where machines take over human. Credit: futurebuff.com

One side of the argument states that no computer or machine will ever achieve human intelligence, while others believe that “the definition of intelligence is irrelevant if the net result is the same.” To be honest, I think given the technological explosion in the 21st century, there is no reason for us to be pessimistic about the future of AI. In my opinion, robots and machines are, in essence, extensions of our basic human body’s functioning. For example, cars were invented so that we could “walk” faster. The quest for AI is nothing more than an intellectual exploration of the limit of human brains.

But who knows? Only time can tell.

 

AlphaGo vs. Lee Sedol: https://en.wikipedia.org/wiki/AlphaGo_versus_Lee_Sedol

Google Official Blog: https://googleblog.blogspot.com/2016/03/what-we-learned-in-seoul-with-alphago.html

Technological singularity: https://en.wikipedia.org/wiki/Technological_singularity

One thought on “Dawn of the Planet of the Robots

  1. rmb5769

    It is certainly always amazing to think about the steps technology has taken in just a short amount of years. The increase in artificial intelligence is indeed exponential, and growing more rapidly than ever. I thought it was very interesting how a game built on human intuition took the longest to program. I believe we are not too far from technological singularity. Perhaps, we will see this shift in our own lifetimes. It will be interesting to follow

Leave a Reply