Log in register
piqd uses cookies and other analytical tools to offer this service and to enhance your user experience.

Your podcast discovery platform

Curious minds select the most fascinating podcasts from around the world. Discover hand-piqd audio recommendations on your favorite topics.

You are currently in channel:

Technology and society

Roshawn Terrell
AI and Machine Learning Researcher For
View piqer profile
piqer: Roshawn Terrell
Monday, 12 November 2018

Why Self-Taught Artificial Intelligence Has Trouble With The Real World

Listen to this podcast from Quanta Magazine, covering the current state of the art of artificial intelligence, and why current techniques still have trouble with the real world. 

The first notable achievement was probably when IBM's Deep Blue computer beat Garry Kasparov at chess in 1997. They did this in part by building their algorithm based off of centuries of chess wisdom. There have been many chess engines since then, with the most powerful being Stockfish, capable of analyzing over 70 million moves per second. Chess was beat, though only through brute force calculation.

However the game of Go is a much tougher nut to crack. As simple as the rules may be, Go is a game of profound complexity. There are an astonishing 10 to the power of 170 possible board configurations – more than the number of atoms in the known universe – making Go a googol times more complex than chess. You literally wouldn't have enough matter in the universe to build a computer capable of calculating all the possible moves. Which is why it's so fascinating that the human brain can handle the game so well.

Only recently, with the power of algorithms modeled on the human brain, have we been able to conquer Go. 

The problem with all of these games, though, is that they are games of 'perfect information,' which means the players see everything that's happening on the board at all times. The real world, however, is mostly comprised of 'hidden information,' where you don't have all the variables, and yet you still must make a decision. 

Currently, Deepmind – the company that built the algorithm that conquered Go – is setting its eyes toward a game known as StarCraft II, a real time strategy game of immense complexity in which players must make decisions based on hidden imperfect information all the time. The battlefield in the game is covered in a 'fog of war,' allowing the enemy to obscure their positions.

If a game like StarCraft II can be beat, then we come one step closer to creating true AI. 

Why Self-Taught Artificial Intelligence Has Trouble With The Real World
One vote

Would you like to comment? Then register now for free!