AlphaGo Defeats World's Best Human Player

AI Go Program Performs Better Than Expected

Anything that doesn't fit anywhere else below.

Moderators: Calilasseia, ADParker

Re: AlphaGo Defeats World's Best Human Player

#61  Postby tuco » Jan 24, 2019 4:39 pm

DEEPMIND - STARCRAFT II DEMONSTRATION

When we last checked in with DeepMind, Oriol Vinyals stepped on to the BlizzCon 2018 stage to share the exciting progress their AI had made in StarCraft II. The AI, or agent, was able to perform basic macro focused strategies as well as defend against cheesy tactics like cannon rushes.

On January 24, at 19:00 CET, head over to StarCraft’s Twitch channel or DeepMind’s Youtube channel to learn what developments have been made. You won’t want to miss it!


https://news.blizzard.com/en-gb/starcra ... onstration


---
For those unfamiliar with Starcraft, as a reminder of challenges: http://www.rationalskepticism.org/gener ... l#p2390017
tuco
 
Posts: 14774

Print view this post

Ads by Google


Re: AlphaGo Defeats World's Best Human Player

#62  Postby newolder » Jan 24, 2019 8:20 pm

Phew!

AlphaStar 10 1 Humans

But the human (Mana) won the final live game. Yay ground apes!
Geometric forgetting gives me loops. - Nima A-H
User avatar
newolder
 
Name: Albert Ross
Posts: 5959
Age: 8
Male

Country: Feudal Estate number 9
Print view this post

Re: AlphaGo Defeats World's Best Human Player

#63  Postby tuco » Jan 24, 2019 8:27 pm

Just ended.

5:0 and 5:0 against pro players, in an exhibition match, lost to a pro, revenge.

One of the guys from DeepMind said its comparable to Kasparov losing to Deep Blue or Seedol losing to AlphaGo. One of the pros said, it actually learned new strategies from it. They talk quite a bit of time about DeepMind between the matches. 40k people watching live on Twitch.

--
edit: no comment
Last edited by tuco on Jan 24, 2019 8:35 pm, edited 1 time in total.
tuco
 
Posts: 14774

Print view this post

Re: AlphaGo Defeats World's Best Human Player

#64  Postby tuco » Jan 24, 2019 8:28 pm

newolder wrote:Phew!

AlphaStar 10 1 Humans

But the human (Mana) won the final live game. Yay ground apes!


Yeah I rooted for MaNa :) and I dont understand the game at all.
tuco
 
Posts: 14774

Print view this post

Re: AlphaGo Defeats World's Best Human Player

#65  Postby newolder » Jan 24, 2019 8:37 pm

3-d space invaders with millennial names for the combating force elements? :dunno:

The "Taxi driver" project I've seen mentioned in some of the DeepMind reports will have an interesting nay crashing start, I'll wager. :lol:
Geometric forgetting gives me loops. - Nima A-H
User avatar
newolder
 
Name: Albert Ross
Posts: 5959
Age: 8
Male

Country: Feudal Estate number 9
Print view this post

Re: AlphaGo Defeats World's Best Human Player

#66  Postby tuco » Jan 24, 2019 8:42 pm

No rules.

edit: Also I expect a millennial to object. The did not go for chess, I guess, because it was done and because this was more challenging.
tuco
 
Posts: 14774

Print view this post

Re: AlphaGo Defeats World's Best Human Player

#67  Postby newolder » Jan 24, 2019 9:25 pm

At a charge of £x per km and without crashing, get from A to C via B then on to E via D etc. and return to A at the end of a shift. Anyone sick in the vehicle got charged an extra fiver when I was at Uni. in Liverpool (max. 5 passengers). Those are the rules as far as I recall.
Geometric forgetting gives me loops. - Nima A-H
User avatar
newolder
 
Name: Albert Ross
Posts: 5959
Age: 8
Male

Country: Feudal Estate number 9
Print view this post

Ads by Google


Re: AlphaGo Defeats World's Best Human Player

#68  Postby newolder » Jan 24, 2019 9:28 pm

tuco wrote:No rules.

edit: Also I expect a millennial to object. The did not go for chess, I guess, because it was done and because this was more challenging.


LeelaChessZero (LC0) is a Neural Net. chess engine (based on the original AlphaZero codes) currently destroying the number 31 seeded engine in the TCEC cup competition here. :smile:
Geometric forgetting gives me loops. - Nima A-H
User avatar
newolder
 
Name: Albert Ross
Posts: 5959
Age: 8
Male

Country: Feudal Estate number 9
Print view this post

Re: AlphaGo Defeats World's Best Human Player

#69  Postby tuco » Jan 24, 2019 9:35 pm

Competition can be amazing. And lots of cores?
tuco
 
Posts: 14774

Print view this post

Re: AlphaGo Defeats World's Best Human Player

#70  Postby newolder » Jan 24, 2019 9:44 pm

Details at the TCEC site - I don't know the exact specs. but LC0 & other NN engines use GPUs whilst "traditionally coded" engines use a 44-core CPU block. The hardware comparison between the two methods is a keen topic of discussion in the chatbox - Should equality of electrical power use be the aim? for example...
Geometric forgetting gives me loops. - Nima A-H
User avatar
newolder
 
Name: Albert Ross
Posts: 5959
Age: 8
Male

Country: Feudal Estate number 9
Print view this post

Re: AlphaGo Defeats World's Best Human Player

#71  Postby tuco » Jan 25, 2019 7:31 am

For competitions, it makes lots of sense.

---
edit: this is the DM team write up on the event: https://deepmind.com/blog/alphastar-mas ... rcraft-ii/
tuco
 
Posts: 14774

Print view this post

Re: AlphaGo Defeats World's Best Human Player

#72  Postby tuco » Jan 25, 2019 12:44 pm

newolder wrote:
tuco wrote:No rules.

edit: Also I expect a millennial to object. The did not go for chess, I guess, because it was done and because this was more challenging.


LeelaChessZero (LC0) is a Neural Net. chess engine (based on the original AlphaZero codes) currently destroying the number 31 seeded engine in the TCEC cup competition here. :smile:


“Traditional engines are exceptionally strong and make few obvious mistakes, but can drift when faced with positions with no concrete and calculable solution,” he says. “It's precisely in such positions where ‘feeling’, ‘insight’ or ‘intuition’ is required that AlphaZero comes into its own."


https://deepmind.com/blog/alphazero-she ... gi-and-go/

I noticed yesterday that Oriol Vinyals was talking about AlphaStar using terms like "perceive" or "obsessed" and now ‘feeling’, ‘insight’ or ‘intuition’. Just curious, is this how we should think about AI nowadays or is he just using common language to be understood?
tuco
 
Posts: 14774

Print view this post

Re: AlphaGo Defeats World's Best Human Player

#73  Postby newolder » Jan 25, 2019 1:03 pm

I'm no expert in AI and machine learning and I agree that the experts - at DeepMind at least - seem to use those words to describe their efforts. Since TPUs (tensor precessing units) are massively parallel and our own brains show parallelism too, I'm not too surprised by the convergence of descriptive language here. I was reading about feedback in neural networks recently and when a neural net. with feedback extracts faces from randomised input data it seems pareidolia is an emergent feature of such hardware that would be ridiculously difficult to hard code.
Geometric forgetting gives me loops. - Nima A-H
User avatar
newolder
 
Name: Albert Ross
Posts: 5959
Age: 8
Male

Country: Feudal Estate number 9
Print view this post

Re: AlphaGo Defeats World's Best Human Player

#74  Postby GrahamH » Jan 25, 2019 1:32 pm

newolder wrote:I'm no expert in AI and machine learning and I agree that the experts - at DeepMind at least - seem to use those words to describe their efforts. Since TPUs (tensor precessing units) are massively parallel and our own brains show parallelism too, I'm not too surprised by the convergence of descriptive language here. I was reading about feedback in neural networks recently and when a neural net. with feedback extracts faces from randomised input data it seems pareidolia is an emergent feature of such hardware that would be ridiculously difficult to hard code.


I'm not an expert either, but from what I do know of ANNs I think "intuition" is a useful term to use. The networks have stronger or weaker resposnses to patterns in the data but it's not analytical or explicit or conventionally algorithmic. We could say "they know it when they see it" but the machine itself, or it's human programmers wouldn't know quite why it's seen it, or what features of pattern it saw. So we could call it "ineffable" like human intuition.


Having said that there are some tools for mapping states in the network to sensory patterns. Google Dream demonstrates something of what's inside and generative networks can reveal how parts of a network relate to patterns by stimulating those nodes and seing how the generated output changes.


Why do you think that?
GrahamH
 
Posts: 19431

Print view this post

Re: AlphaGo Defeats World's Best Human Player

#75  Postby GrahamH » Mar 08, 2019 1:42 pm

This is somewhat related to how human-like ANNs are or can be.

Why do you think that?
GrahamH
 
Posts: 19431

Print view this post

Previous

Return to General Science & Technology

Who is online

Users viewing this topic: No registered users and 1 guest