Match 5 15 Minute Summary - Google DeepMind Challenge Match 2016

Google DeepMind · Intermediate ·📰 AI News & Updates ·10y ago

Key Takeaways

The video discusses the final match of the Google DeepMind Challenge Match 2016 between AlphaGo and Lee Sedol, with a focus on the game's strategic moves and territory control. The match is analyzed by Michael Redmond and Chris Garlock, providing insights into the game's progression and AlphaGo's strengths.

Full Transcript

[Music] hi Chris Garlock here with Michael Redmond for the match highlights in the fifth and final game of the Google deep mine challenge Michael leas at all had one game four snatching Victory from the jaws of defeat and came into this final game uh with a with a pretty clear game plan can we explain what the plan was and how it worked there were already signs of it I think from game number two okay um but uh at first uh Le was just playing it with white um kind of an amashi strategy like taking territory and then invading the moyo um which is usually played with white but uh in this game he was playing it with black and I was sort of I did foresee that sort of I I was thinking maybe he would do it but since it's usually not a strategy that's played with black I was sort of um not quite sure what who would play for territory as he did we can already see that black has taken territory in two corners um playing a opening that is very oriented towards territory um or I thought still maybe he would try a fighting style opening which didn't really work in the first game right and it didn't work in the third game either so um I think in the fourth game he really decided that um he could take territory and um exploit the weakness maybe of alphao by um invading the moyo and getting into a complicated fight there so um after all it was the game plan in this game too um I was sort of suspecting it but I wasn't sure so I was interested to see him actually using this plan um it's a bit unusual for white to ignore this but white is playing for the whole board so it's maybe a valid way for white to play White plays the H this is the strongest uh attack and with this move white is basically already thinking of just sacrificing Stone so black extends and white plays there now white has a base here um black has a nice territory here locally it's quite a success for black white did get the extra move here which is going to influence the area here and white has sente so white put one stone here um in return black has a very nice territory here um there's the point that it's it he already had a solid territory so he's just building on a solid territory which is not the most efficient thing but I would be happy to be playing with black yeah and you like this because of all the territory yeah and then black slides um this is also a natural move um and white pincers this is a pincer um to make influence in this area with this behind it so this is a very natural move if White had played in the corner like this it would have been um just perfect for black to extend here making a stable group uh in the place where White had hoped to make some well if not territory a moo at least um so white plays here at this point uh leis played this move which I I sort of Doubt actually it's a it's hard to say which is better but since the ladder is good for black I would have probably just played here and if white plays here black can connect underneath and the honey the ladder here is good for black so white will either play here or sometimes you see players play this um but in that case black gets this leaves a lot of potential for these two black stones to move out later in the game um there's still the invasion in the through three point here so I would be happy to play with black in this opening in the actual game um black gave some territory back to White by playing this and pulling back this move B basically makes um Mii or um Mii or interchangeable points so either that move which is where white plate or if white plays on this side um black if black will play here white extends and again this this shape is kind of a double extension from the Star point and it's sort of inefficient when black jumps into the three three point so this is more or less what black was aiming at so it was correct for white to play on this side and allow black to cut and in this shape black has slightly more thickness compared to to this shape which is a common josei black has more thickness towards the center um but white has these extra Stones here which sort of form some extra territory here so black has lost some territory in return for this double uh this line of stones here MH um which you're facing towards which you're facing towards the center so he's he's sort of um afraid of a center moo a little bit it's not really consistent with his plan to take as much territory as possible so that's why I sort of question this move but it is a kind of a balanced play here then white plays on top here um this is a natural follow-up move for this whole shape the throwing away because it's um sort of planning to squeeze using the AY of these Stones maybe not the connection but at least um squeeze black reduce Black's territory while building on the center so it's a very natural move so black resists a little bit and push through once yeah and now black points out to the left that reduces White's potential in this area of the Border great deal now something odd happens something odd happens um this cut he cut first Cuts first um the whole sequence here um just to um pointed out in the beginning the whole sequence here will um end up with black being very very strong in this area so the drawback of white sequence is that White's going to make very black very solid here and that makes a difference when white plays moves like this or moves like in the game white chose this one in either case there's an an entry point here okay and so um I would calculate White's loss on this side to be about five points when white plays starts playing moves like this here and here and here and white connected PL a h and we can see the black position getting stronger and stronger and none of this actually works it doesn't work actually why it's gaining a couple of points on this side a few points on this side so it depends on really how on how you count the gain that black has on this side which I think is pretty bad but um if if for instance if this becomes Black's territory or if nothing U special happens here then there is a slight gain for white on this side so it is um it sort of depends on how the game continues in some cases this is not going to be a loss at all um the general idea is that White's lost so much on this side that usually it's bad for white but we can't really be 100% sure yet at this point so sometimes this doesn't have such a um terrible impact on the game and there was a little concern when this was being pled during the game that maybe alphao is going on tilt a little bit again well I didn't think so well the the local moves sort of make sense and even if you assume that it was a genuine mistake it would be a mistake um such as the mistake in the center um after the white move 78 in the fourth game right um where maybe alphao didn't have the capacity to to read one of the moves that was a low profile so um of course the programmers will be able to dig into the program and find that out later and then after this white place here mhm um this is a standard sequence here throwing one stone away like this just to get the extra forcing moves on the outside to reduce Black's territory black this is more or less forced and white can play an Atari on the top and then Atari on the side so by sacrificing this one stone white gets some extra forcing moves and a nice wall there now at this point black has something in the vicinity of 40 points here okay about 10 points here more than 10 points there so well well slightly over 60 points it's a lot of cash it's all it's all um solid territory it's not going to well this is actually if you only count the very solid territory slightly less than 40 points but in either case is he's black is in the vicinity of 60 points so liso's plan is kind of working he's got white to create this big moyo and he's now going to jump in I guess and this is not really territory yet because black can jump in here uh White has some territory here and some territory here all of White's territory um doesn't really add up to more than 30 points at this point so black has a huge um advantage in territory and he played here which was probably a mistake um this turned out to be a bit deep one would hope that white would be answering something like this in which case black could jump and this would really uh completely reduce White's potential in its whole area and even if White made some territory in the corner here it would be good enough for black to wi and this works nicely with solid gr and we're sort of seeing if you assume that white is continuing to try to take territory um you can see that the center is just disappearing quite nicely right this wouldn't be too easy right um actually um what black should have done was black should not have been so greedy maybe and played something closer to the center which would be not as effective when white compared to the diagram I just showed the variation I just showed but it's much more safe CU white really it's more difficult for white to break in through there um if white and next Black does have a very very nice move here which puts pressure on the corner if white plays something like that this move alone without even following it up um is already reduced White's territory here and it's very difficult for white to successfully attack now because black can sort of just dodge away treat it very lightly and um it's a a move on the top for a move below very difficult for white to surround it now MH so black could even uh black has a choice then he could play another move towards the center or he could jump in here um which is a very big territorial move mhm um so this would be I think this is a slight Advantage for black um this move it looked really natural um but if you assume moves like this or moves like this it's easy for black but the move white played this capping move was very very good um it's a move that um actually uh human players would play too it's not as if it's um especially a computer move mhm um but it's just perfect for this um it's it's an example of alphago's strength so anyone who is afraid that alphao would sort of go berserk at this point um should be um relieved of that um worry by seeing this move and I think your analysis and commentary was that alphao really just plays this whole sequence it's very nice sort of good style you might say um it's um very natural moves and white didn't even try to kill black uh for instance at this point if White had played here this would have been a more serious attempt to kill the black grou um and it would um entail some danger because after this there would be the cut here later and white stones in this area are not so strong that means that black could possibly use that to sort of move out into the center and this move would be rather small as far as territory is concerned so it was probably a wise way for white to play just to play on the top and and and sort of make this the width of this area thicker black plays the key point and yes and now black is alive white continues to sort of squeeze black and now it's completely now uh there's no doubt that this black group is alive and white plays on top again again it's good that you see that white is not um connecting uh connecting is a bit heavy actually once white has forced black to live here uh you can say that these two stones are not really all that important it's not that White's um going to give them up too easily sure um but in some cases white can afford to throw them away so um they're not as important as they were before MH um and white plays on the outside oh actually I think black played once here first and then lived but it's the same thing um white plays on the top the point is white gets a big this is already looking almost like territory I think if we added one white move it would be territory of close to 50 points and so black had to come in and white still had potential on the side in this area and this territory is very difficult to count actually because it it can it could it decreases with every time black plays a move there um but it's also it takes some time for black to get rid of the whole territory so it's it becomes a very difficult endgame to execute and very close already at this point and I think towards the end it was a half Point difference at some point but um at least it'll lost a little bit um alphao after its win was clinched I think it did play some questionable moves but again it's a question of it was winning anyway so maybe it was just playing the safer route so this this game from here and you were counting and and the score was really interesting because there was a lot of moves there were some swaps uh but in the in the end the score didn't really change for a really long time not much all the time it was something like seven eight points on the board so it was really really close right uh it was it was a great uh we thought we're actually going to have a countable game yeah well it was countable didn't count they just a bother because it was I I think that the they were saying it was like a point and a half was points so uh well yet another uh amazing game uh I want to say thank you to you Michael for you know all of your just incredible analysis uh these game commentaries they were all just exciting and tough games appreciate that uh and congratulations uh again to the entire alphao team uh and to Lisa doll this was an amazing week uh really appreciate the attention to go I think it's really going to as you said open up a new era uh for the game for players and uh for artificial intelligence so we look forward to that uh so I look forward to next time I see you across the board um and thanks everybody for watching hope you enjoyed thank you [Music]

Original Description

15 Minute Summary by Michael Redmond 9 dan professional and Chris Garlock on today's nail-biting fifth and final match between DeepMind's program AlphaGo take on the legendary Lee Sedol (9-dan pro), the top Go player of the past decade, in a $1M 5-game challenge match in Seoul. In October 2015, AlphaGo became the first computer program ever to beat a professional Go player by winning 5-0 against the reigning 3-times European Champion Fan Hui (2-dan pro). That work was featured in a front cover article in the science journal Nature in January 2016.
Watch on YouTube ↗ (saves to browser)
Sign in to unlock AI tutor explanation · ⚡30

Playlist

Uploads from Google DeepMind · Google DeepMind · 20 of 60

1 RL Course by David Silver - Lecture 8: Integrating Learning and Planning
RL Course by David Silver - Lecture 8: Integrating Learning and Planning
Google DeepMind
2 RL Course by David Silver - Lecture 1: Introduction to Reinforcement Learning
RL Course by David Silver - Lecture 1: Introduction to Reinforcement Learning
Google DeepMind
3 RL Course by David Silver - Lecture 2: Markov Decision Process
RL Course by David Silver - Lecture 2: Markov Decision Process
Google DeepMind
4 RL Course by David Silver - Lecture 5: Model Free Control
RL Course by David Silver - Lecture 5: Model Free Control
Google DeepMind
5 RL Course by David Silver - Lecture 6: Value Function Approximation
RL Course by David Silver - Lecture 6: Value Function Approximation
Google DeepMind
6 RL Course by David Silver - Lecture 4: Model-Free Prediction
RL Course by David Silver - Lecture 4: Model-Free Prediction
Google DeepMind
7 RL Course by David Silver - Lecture 3: Planning by Dynamic Programming
RL Course by David Silver - Lecture 3: Planning by Dynamic Programming
Google DeepMind
8 RL Course by David Silver - Lecture 10: Classic Games
RL Course by David Silver - Lecture 10: Classic Games
Google DeepMind
9 RL Course by David Silver - Lecture 7: Policy Gradient Methods
RL Course by David Silver - Lecture 7: Policy Gradient Methods
Google DeepMind
10 Google DeepMind: Ground-breaking AlphaGo masters the game of Go
Google DeepMind: Ground-breaking AlphaGo masters the game of Go
Google DeepMind
11 Match 1 - Google DeepMind Challenge Match: Lee Sedol vs AlphaGo
Match 1 - Google DeepMind Challenge Match: Lee Sedol vs AlphaGo
Google DeepMind
12 Match 2 - Google DeepMind Challenge Match: Lee Sedol vs AlphaGo
Match 2 - Google DeepMind Challenge Match: Lee Sedol vs AlphaGo
Google DeepMind
13 Match 1 15 min Summary - Google DeepMind Challenge Match
Match 1 15 min Summary - Google DeepMind Challenge Match
Google DeepMind
14 Match 3 - Google DeepMind Challenge Match: Lee Sedol vs AlphaGo
Match 3 - Google DeepMind Challenge Match: Lee Sedol vs AlphaGo
Google DeepMind
15 Match 2 15 Minute Summary - Google DeepMind Challenge Match 2016
Match 2 15 Minute Summary - Google DeepMind Challenge Match 2016
Google DeepMind
16 Match 3 15 Minute Summary - Google DeepMind Challenge Match 2016
Match 3 15 Minute Summary - Google DeepMind Challenge Match 2016
Google DeepMind
17 Match 4 - Google DeepMind Challenge Match: Lee Sedol vs AlphaGo
Match 4 - Google DeepMind Challenge Match: Lee Sedol vs AlphaGo
Google DeepMind
18 Match 4 15 Minute Summary - Google DeepMind Challenge Match 2016
Match 4 15 Minute Summary - Google DeepMind Challenge Match 2016
Google DeepMind
19 Match 5 - Google DeepMind Challenge Match: Lee Sedol vs AlphaGo
Match 5 - Google DeepMind Challenge Match: Lee Sedol vs AlphaGo
Google DeepMind
Match 5 15 Minute Summary - Google DeepMind Challenge Match 2016
Match 5 15 Minute Summary - Google DeepMind Challenge Match 2016
Google DeepMind
21 DQN SPACE INVADERS
DQN SPACE INVADERS
Google DeepMind
22 DQN Breakout
DQN Breakout
Google DeepMind
23 Asynchronous Methods for Deep Reinforcement Learning: Labyrinth
Asynchronous Methods for Deep Reinforcement Learning: Labyrinth
Google DeepMind
24 Asynchronous Methods for Deep Reinforcement Learning: MuJoCo
Asynchronous Methods for Deep Reinforcement Learning: MuJoCo
Google DeepMind
25 Asynchronous Methods for Deep Reinforcement Learning: TORCS
Asynchronous Methods for Deep Reinforcement Learning: TORCS
Google DeepMind
26 Differentiable neural computer family tree inference task
Differentiable neural computer family tree inference task
Google DeepMind
27 StarCraft II DeepMind feature layer API
StarCraft II DeepMind feature layer API
Google DeepMind
28 DeepMind Health – Partnership with the Royal Free London NHS Foundation Trust
DeepMind Health – Partnership with the Royal Free London NHS Foundation Trust
Google DeepMind
29 DeepMind Health – Michael Wise – a patient's journey
DeepMind Health – Michael Wise – a patient's journey
Google DeepMind
30 Streams – a platform for a digital NHS
Streams – a platform for a digital NHS
Google DeepMind
31 DeepMind Lab - Nav Maze Level 1
DeepMind Lab - Nav Maze Level 1
Google DeepMind
32 DeepMind Lab - Stairway to Melon Level
DeepMind Lab - Stairway to Melon Level
Google DeepMind
33 DeepMind Lab - Laser Tag Space Bounce Level (Hard)
DeepMind Lab - Laser Tag Space Bounce Level (Hard)
Google DeepMind
34 Exploring the mysteries of Go with AlphaGo and China's top players
Exploring the mysteries of Go with AlphaGo and China's top players
Google DeepMind
35 Demis Hassabis on AlphaGo: its legacy and the 'Future of Go Summit' in Wuzhen, China
Demis Hassabis on AlphaGo: its legacy and the 'Future of Go Summit' in Wuzhen, China
Google DeepMind
36 The Future of Go Summit: AlphaGo & Ke Jie match 1 moves analysis
The Future of Go Summit: AlphaGo & Ke Jie match 1 moves analysis
Google DeepMind
37 The Future of Go Summit: AlphaGo & Ke Jie match 2 moves analysis
The Future of Go Summit: AlphaGo & Ke Jie match 2 moves analysis
Google DeepMind
38 The Future of Go Summit: Pair Go moves analysis
The Future of Go Summit: Pair Go moves analysis
Google DeepMind
39 The Future of Go Summit: AlphaGo & Ke Jie match 3 moves analysis
The Future of Go Summit: AlphaGo & Ke Jie match 3 moves analysis
Google DeepMind
40 Emergence of Locomotion Behaviours in Rich Environments
Emergence of Locomotion Behaviours in Rich Environments
Google DeepMind
41 StarCraft II 'mini games' for AI research
StarCraft II 'mini games' for AI research
Google DeepMind
42 Trained and untrained agents play StarCraft II full 1vs1 game
Trained and untrained agents play StarCraft II full 1vs1 game
Google DeepMind
43 DeepMind open source PySC2 toolset for Starcraft II
DeepMind open source PySC2 toolset for Starcraft II
Google DeepMind
44 ICML 2017: Test of Time Award (Sylvain Gelly & David Silver)
ICML 2017: Test of Time Award (Sylvain Gelly & David Silver)
Google DeepMind
45 Ke Jie and DeepMind's Go Ambassador Fan Hui review the 3rd AlphaGo vs Ke Jie game
Ke Jie and DeepMind's Go Ambassador Fan Hui review the 3rd AlphaGo vs Ke Jie game
Google DeepMind
46 Ke Jie and DeepMind's Go Ambassador Fan Hui review the 1st AlphaGo vs Ke Jie game
Ke Jie and DeepMind's Go Ambassador Fan Hui review the 1st AlphaGo vs Ke Jie game
Google DeepMind
47 Ke Jie and DeepMind's Go Ambassador Fan Hui review the 2nd AlphaGo vs Ke Jie game
Ke Jie and DeepMind's Go Ambassador Fan Hui review the 2nd AlphaGo vs Ke Jie game
Google DeepMind
48 AlphaGo Zero: Discovering new knowledge
AlphaGo Zero: Discovering new knowledge
Google DeepMind
49 AlphaGo Zero: Starting from scratch
AlphaGo Zero: Starting from scratch
Google DeepMind
50 Defining principles for tech companies in the NHS: DeepMind Health's Collaborative Listening Summit
Defining principles for tech companies in the NHS: DeepMind Health's Collaborative Listening Summit
Google DeepMind
51 A systems neuroscience approach to building AGI - Demis Hassabis, Singularity Summit 2010
A systems neuroscience approach to building AGI - Demis Hassabis, Singularity Summit 2010
Google DeepMind
52 Retour de Rémi Munos en France et ouverture de DeepMind Paris
Retour de Rémi Munos en France et ouverture de DeepMind Paris
Google DeepMind
53 Grid cells - Caswell Barry, UCL
Grid cells - Caswell Barry, UCL
Google DeepMind
54 DeepMind Health Research and Moorfields Eye Hospital NHS Foundation Trust: What our research shows
DeepMind Health Research and Moorfields Eye Hospital NHS Foundation Trust: What our research shows
Google DeepMind
55 DeepMind Health Research and Moorfields Eye Hospital NHS Foundation Trust: A Patient's Story
DeepMind Health Research and Moorfields Eye Hospital NHS Foundation Trust: A Patient's Story
Google DeepMind
56 Deep Learning 3: Neural Networks Foundations
Deep Learning 3: Neural Networks Foundations
Google DeepMind
57 Deep Learning 5: Optimization for Machine Learning
Deep Learning 5: Optimization for Machine Learning
Google DeepMind
58 Deep Learning 8: Unsupervised learning and generative models
Deep Learning 8: Unsupervised learning and generative models
Google DeepMind
59 Reinforcement Learning 1: Introduction to Reinforcement Learning
Reinforcement Learning 1: Introduction to Reinforcement Learning
Google DeepMind
60 Deep Learning 2: Introduction to TensorFlow
Deep Learning 2: Introduction to TensorFlow
Google DeepMind

This video provides an in-depth analysis of the final match of the Google DeepMind Challenge Match 2016, highlighting AlphaGo's strengths and strategic moves. Viewers can learn about the game's progression and gain insights into AI decision-making in complex scenarios. The video is a valuable resource for those interested in AI and game analysis.

Key Takeaways
  1. Watch the video to understand the game's progression
  2. Analyze AlphaGo's strategic moves and territory control
  3. Learn about the moyo strategy and its application in the game
  4. Understand the concept of sente and its importance in Go
  5. Study the game's analysis and insights provided by Michael Redmond and Chris Garlock
💡 AlphaGo's strength lies in its ability to make natural and good style moves, which allowed it to gain a significant advantage in territory control.

Related Reads

Up next
how to make passive income with chatgpt AI and audible ?
SuccessPursuitZone
Watch →