Free AI web copilot to create summaries, insights and extended knowledge, download it at here

Abstract

.</p><blockquote id="19ad"><p><i>“</i>Even if I become the number one, there is an entity that cannot be defeated.” — Lee Sedol asserted.</p></blockquote><p id="f6ea">In late 2017, DeepMind revealed a newer, improved version of AlphaGo. Named AlphaGo Zero, it picked up the game from scratch. Learning by playing millions of games against itself, AlphaGo Zero thrashed its little brother in a Go-off. With a score of 100 to 0, its skill level is considered beyond human capabilities.</p><h1 id="44d3">AlphaZero</h1><p id="1068">Similar to AlphaGo, AlphaZero was also developed to dominate the field of an ancient strategic board game. Only this time, it’s chess. In a game of 64 squares and 32 pieces, the aim is to checkmate the opponent.</p><p id="a38a">AlphaZero used a similar concept to AlphaGo — reinforcement learning and self-play. In just 9 hours, AlphaZero played 44 million games against itself. Essentially the Doctor Strange of chess, AlphaZero inspects tens of millions of positions per second to accurately determine the best move.</p><p id="3e14">In 2017, AlphaZero played Stockfish (the strongest chess engine at the time) in a 100-game match. The match ended 28, nil to AlphaZero (and 72 draws). Some said the match was one-sided, with AlphaZero using Google’s “supercomputer” and Stockfish using the hardware of a laptop. Hence a rematch with 1000 games. With more advanced hardware and a lar

Options

ger database, the odds were even. Or so they thought. AlphaZero triumphed in 155 of those 1000 games, conceded six times, and drew the rest.</p><p id="557a">Humans simply stand no chance against the computer. Current world champion Magnus Carlsen has a rating of 2800 while AlphaZero is reportedly rated over 3500.</p><h1 id="1482">AlphaStar</h1><p id="c294">StarCraft 2 is a real-time strategy game, a simulation where players gather resources to outgun or defend against opponents. Again, DeepMind’s artificial intelligence team created AlphaStar to crush all StarCraft players’ dreams.</p><p id="0c51">AlphaStar learns by playing against itself as well. Due to the bajillion strategies and possible actions in the game, AlphaStar has yet to “solve” StarCraft. However, it beat top player Grzegorz “MaNa” Komincz and teammate Dario “TLO” Wünsch in a series of 10 matches while in testing. After further training, AlphaStar ranked above 99.8% of all active players and earned a spot among the top 200 regional players.</p><h1 id="f0e1">Final Thoughts</h1><p id="320d">If there’s one thing we can learn from these AIs, it would be their self-learning abilities. It goes to show how crucial it is to not only learn from others but yourself. We should be excited to see what DeepMind has in store for the following years. Maybe “AlphaMine” would be able to complete Minecraft in a few minutes.</p></article></body>

The AI That Defeated Humans

Why we don’t stand a chance

Artificial intelligence could be one of humanity’s most useful inventions. — Google’s DeepMind

In 2016, Google’s artificial intelligence firm — DeepMind, set out to conquer the games of Go, Chess, and Starcraft. To say that they succeeded is an understatement.

AlphaGo convincingly beat the world’s greatest Go players. AlphaZero defeated the strongest chess engine at the time. And AlphaStar outplayed all but the very best StarCraft players.

AlphaGo

Believed to be the oldest board game played to this day, Go is a two-player abstract strategy board game. The objective of the game is to surround more territory than the opponent. Every captured stone and vacant space is a point, the player with the most points wins.

AlphaGo learned the game of Go by studying a set of over 100,000 human games. It then defeated 18-time world champion Lee Sedol in 2016 and world number one Ke Jie in 2017.

“Even if I become the number one, there is an entity that cannot be defeated.” — Lee Sedol asserted.

In late 2017, DeepMind revealed a newer, improved version of AlphaGo. Named AlphaGo Zero, it picked up the game from scratch. Learning by playing millions of games against itself, AlphaGo Zero thrashed its little brother in a Go-off. With a score of 100 to 0, its skill level is considered beyond human capabilities.

AlphaZero

Similar to AlphaGo, AlphaZero was also developed to dominate the field of an ancient strategic board game. Only this time, it’s chess. In a game of 64 squares and 32 pieces, the aim is to checkmate the opponent.

AlphaZero used a similar concept to AlphaGo — reinforcement learning and self-play. In just 9 hours, AlphaZero played 44 million games against itself. Essentially the Doctor Strange of chess, AlphaZero inspects tens of millions of positions per second to accurately determine the best move.

In 2017, AlphaZero played Stockfish (the strongest chess engine at the time) in a 100-game match. The match ended 28, nil to AlphaZero (and 72 draws). Some said the match was one-sided, with AlphaZero using Google’s “supercomputer” and Stockfish using the hardware of a laptop. Hence a rematch with 1000 games. With more advanced hardware and a larger database, the odds were even. Or so they thought. AlphaZero triumphed in 155 of those 1000 games, conceded six times, and drew the rest.

Humans simply stand no chance against the computer. Current world champion Magnus Carlsen has a rating of 2800 while AlphaZero is reportedly rated over 3500.

AlphaStar

StarCraft 2 is a real-time strategy game, a simulation where players gather resources to outgun or defend against opponents. Again, DeepMind’s artificial intelligence team created AlphaStar to crush all StarCraft players’ dreams.

AlphaStar learns by playing against itself as well. Due to the bajillion strategies and possible actions in the game, AlphaStar has yet to “solve” StarCraft. However, it beat top player Grzegorz “MaNa” Komincz and teammate Dario “TLO” Wünsch in a series of 10 matches while in testing. After further training, AlphaStar ranked above 99.8% of all active players and earned a spot among the top 200 regional players.

Final Thoughts

If there’s one thing we can learn from these AIs, it would be their self-learning abilities. It goes to show how crucial it is to not only learn from others but yourself. We should be excited to see what DeepMind has in store for the following years. Maybe “AlphaMine” would be able to complete Minecraft in a few minutes.