AlphaZero Don't Believe the Hype
The games between AlphaZero and Stockfish are going to revolutionise our planet. Artificial intelligence is creating more jobs in our tech industries and shaping how we view the world. Google is a pioneer but others are joining the party, big brains and big money. Healthcare is improving as labs develop new drugs and economists will better know the impact of certain referendums. Soon advancements may supersede our senses and we may not know whether online video-game players are real, or if YouTube comments section are genuine people.
AlphaZero is a tech company with a London office and backed by Google. When the project AlphaGo defeated Ke Jie this caused a disturbance in the computer community. Then it beat 2nd reigning champion Lee Seedol, but has yet to play number one World Champion Lee Changho. Never was a programme expected to win in a human game where lying is a critical component of play.
Alpha also flexed it’s muscles against leading Starcraft player in a bloodbath of war, Elon Musk jumped with investments just to keep an eye on it.
This powerful super-computer-algorithm has now launched into the chess community, and they are welcoming this entity. AlphaZero completely crushed Stockfish in 2017 showcasing one-hundred games, now in 2018 they have one-thousand high-quality games available. Stockfish is the worlds leading chess calculator, and it kept getting lured into faulty lines. And I think I know how this happened.
In Go there is clear win or lose and a draw is impossible kind of. Chess is slightly different and a draw is very common against two equally matched opponents. I don’t think AlphaZero is the end all it’s made out to be and here’s why:
1. AlphaZero profits from confusing Stockfish by offering a short-term advantage it’s happy to accept knowing it knows no better. Alpha understands there are blindspots coming up.
2. Stockfish goes for the win and this is a losing strategy. Stockfish searches for the highest evaluation (i.e. greatest chances) and pursues that sequence which makes no sense up against a super-computer. It should play for a draw, trade down, simple chess. Whose complications are really going to come out on top?
3. This position explains AlphaZeros success. No computer would think to move the Knight in this pattern for it to break free, it requires too much computation. Weak engines suggest moving the King back and forth endlessly, no Knight moves at all! Except AlphaZero would know what to do, which is why it often forces a fixed position similar to this. Here explains image.
I have a feeling that the Alpha team are skeptical to showcase a big tournament until they know a win against the World Chess Champion Magnus Carlsen is possible. If he plays the usual safe moves then a draw could happen, the machine would have to pose real challenges as if every move is a new puzzle. If a human chose the Stockfish moves against AlphaZero then the outcome will obviously be a draw, possibly even a win.
GrandMaster Hikaru Nakamura has shared his sceptic eye on the engines capability.
The algorithm is mostly hype but I’m proud to see the chess world gain some much needed exposure. What DeepMind need to do next is release more interesting games.
AlphaZero is a tech company with a London office and backed by Google. When the project AlphaGo defeated Ke Jie this caused a disturbance in the computer community. Then it beat 2nd reigning champion Lee Seedol, but has yet to play number one World Champion Lee Changho. Never was a programme expected to win in a human game where lying is a critical component of play.
Alpha also flexed it’s muscles against leading Starcraft player in a bloodbath of war, Elon Musk jumped with investments just to keep an eye on it.
This powerful super-computer-algorithm has now launched into the chess community, and they are welcoming this entity. AlphaZero completely crushed Stockfish in 2017 showcasing one-hundred games, now in 2018 they have one-thousand high-quality games available. Stockfish is the worlds leading chess calculator, and it kept getting lured into faulty lines. And I think I know how this happened.
In Go there is clear win or lose and a draw is impossible kind of. Chess is slightly different and a draw is very common against two equally matched opponents. I don’t think AlphaZero is the end all it’s made out to be and here’s why:
1. AlphaZero profits from confusing Stockfish by offering a short-term advantage it’s happy to accept knowing it knows no better. Alpha understands there are blindspots coming up.
2. Stockfish goes for the win and this is a losing strategy. Stockfish searches for the highest evaluation (i.e. greatest chances) and pursues that sequence which makes no sense up against a super-computer. It should play for a draw, trade down, simple chess. Whose complications are really going to come out on top?
3. This position explains AlphaZeros success. No computer would think to move the Knight in this pattern for it to break free, it requires too much computation. Weak engines suggest moving the King back and forth endlessly, no Knight moves at all! Except AlphaZero would know what to do, which is why it often forces a fixed position similar to this. Here explains image.
I have a feeling that the Alpha team are skeptical to showcase a big tournament until they know a win against the World Chess Champion Magnus Carlsen is possible. If he plays the usual safe moves then a draw could happen, the machine would have to pose real challenges as if every move is a new puzzle. If a human chose the Stockfish moves against AlphaZero then the outcome will obviously be a draw, possibly even a win.
GrandMaster Hikaru Nakamura has shared his sceptic eye on the engines capability.
The algorithm is mostly hype but I’m proud to see the chess world gain some much needed exposure. What DeepMind need to do next is release more interesting games.
https://deepmind.com/
Comments
Post a Comment