In a recent publication in Nature, Google Inc GOOG 2.13% GOOGL 2.19%‘s DeepMind artificial intelligence unit detailed the ability of its Q-network (DQN) algorithm to not just learn how to play video games from scratch, but determine how to beat the games in a matter of hours.
The paper detailed the algorithm’s performance on 49 Atari 2600 video games, including Breakout, River Raid, Boxing and Enduro.
How DQN Works
DQN works by using only the raw screen pixels, the set of available actions and the game score as input. DeepMind founder Demis Hassabis was impressed by the algorithm’s ability to master the wide range of games utilizing the same tuning parameters for each.
One of the unique properties of DQN that aids with its learning is its neural-like networks, including networks for reinforced learning.
“Foremost among these was a neurobiologically inspired mechanism, termed ‘experience replay’, whereby during the learning phase DQN was trained on samples drawn from a pool of stored episodes- a process physically realized in a brain structure called the hippocampus through the ultra-fast reactivation of recent experiences during rest periods (eg sleep),” Hassabis and co-author Dharshan Kumaran explained in a blog post.
DQN Versus Human Players
The algorithm performed with varying degrees of success on the range of games.
While DQN managed to perform as well as a human player on about half of the games, it struggled compared to human players in games such as Space Invaders because of the long-term strategy the game requires.
Will DQN Enslave Us All?
DeepMind was initially funded in part by Tesla Motors Inc TSLA 1.68% CEO Elon Musk, who later admitted he recognizes the potential danger in artificial intelligence. “There have been movies about this,” Musk explained, “you know, like Terminator.”
According to DeepMind, the human race has nothing to fear from DQN. They envision the algorithm serving as an aide to humans by completing complex tasks such as planning a fun vacation.
Read this article and all my other articles for free on Benzinga by clicking here
Want to learn more about the stock market? Or maybe you just want to be able to look sophisticated in front of your coworkers when they ask you what you are reading on your Kindle, and you’d prefer to tell them “Oh, I’m just reading a book about stock market analysis,” rather than the usual “Oh, I’m just looking at pics of my ex-girlfriend on Facebook.” For these reasons and more, check out my book, Beating Wall Street with Common Sense. I don’t have a degree in finance; I have a degree in neuroscience. You don’t have to predict what stocks will do if you can predict what traders will do and be one step ahead of them. I made a 400% return in the stock market over five years using only basic principles of psychology and common sense. Beating Wall Street with Common Sense is now available on Amazon, and tradingcommonsense.com is always available on your local internet!
