Go has fallen to computer domination?

[Dirt]

Are the rest of the 2 games going to be played, or is the match already finished ? I would love to see Sedol lose the last 2 games...

All five are supposed to be played regardless of how the earlier games go.

[Dirt]

These poor programmers or researchers that switched from Chess to GO after it appeared that Chess had been 'solved' where should they go to now ?

Apparently StarCraft. I would be surprised if that's even as hard, but I know nothing about it.

[Rochester]

But now it's 3-0

Now complaint not know old games. Then suggest to human make a move for the computer. Demanded see the logger file!

[Leto]

These poor programmers or researchers that switched from Chess to GO after it appeared that Chess had been 'solved' where should they go to now ?

Apparently StarCraft. I would be surprised if that's even as hard, but I know nothing about it.

Did they say that? I know a lot about StarCraft, even played quite a bit of it (thousands of hours). It was released in 1998 and is the best selling real time strategy game (RTS) ever (sold over 10 million copies) and it is mainly a 1 vs 1 game and the goal of the game is to annihilate your enemy.

There are three races to play as, the Terrans (technologically advanced humans), the Protoss (alien race who are more technologically advanced than the Terrans), and the Zerg (a race of evolved organisms who operate in a hive mind-linked chain of command.

Each game starts you and your opponent on opposite sides of a map, each player starts out with a command center and a handful of workers (mineral and gas collectors). As you collect resources you gain the ability to train more workers and build structures and train soldiers. The maps contain resource areas that players should strive to control because the resources may be necessary to defeat their opponent. Games can last anywhere from 4 minutes to several hours, with the average length being around 30 minutes.

It'd be easier to understand if you watch a video of the actual game. Here's one from a recent final played in January of 2016 between a Protoss and Zerg player:
https://www.youtube.com/watch?v=q70hUy7m2JY#t=11m15s

Some games from January 2016 from the round of 16 in a Starcraft Broodwar Artificial Intelligence tournament where only AIs can compete:
https://www.youtube.com/watch?v=S7LgwN5tIng

A sequel to StarCraft, appropriately named StaCraft II, was released in 2010 with updated graphics and changes with the units. Here's a
Stacraft 2 Protoss vs Zerg final from 2015:
https://www.youtube.com/watch?v=OveyfMExs2o

And here's a Terran vs Zerg final from just a few days ago:
https://www.youtube.com/watch?v=zbwEngpKT2A

[Uri Blass]

These poor programmers or researchers that switched from Chess to GO after it appeared that Chess had been 'solved' where should they go to now ?

I wonder if humans can do better in go against machines with a bigger board(for example 29*29 instead of 19*19)

[Leto]

These poor programmers or researchers that switched from Chess to GO after it appeared that Chess had been 'solved' where should they go to now ?

Apparently StarCraft. I would be surprised if that's even as hard, but I know nothing about it.

As for difficulty I would say it is definitely harder especially if it's not cheating (able to see the entire map including the opponent). There is no "best" strategy that it can use to win every game because the opponent can choose a very risky suboptimal strategy to catch you offguard to try to win the game. Some games can last as little as 3 to 4 minutes or last several hours.

I would say it's definitely not impossible to build an AI for StarCraft and/or StarCraft 2 that would dominate the best human players, but I think it will be very challenging to do. They would need a flexible AI that has amazing scouting technique, knows what units perform well against the units that the opponent is using, must be able to adapt to the opponent switching strategies throughout the game, and must have amazing timing.

[George Tsavdaris]

These poor programmers or researchers that switched from Chess to GO after it appeared that Chess had been 'solved' where should they go to now ?

It's not that simple yet.
Yes AlphaGO seems very very strong and probably stronger than 9d pro, but this is on their 1020 CPU+200 GPU or something. But we want it on OUR computers!
Yet, their reports and interviews on the last days say that the single machine AlphaGO has already surpassed Fan Hui's strength, so i really want them to make it available to us. And definitelly a new paper for the improvements they've made between Fan Hui match and Lee Sedol(assuming it's not just self play and reinforcement learning).

Thankfully there is a new open project from the University of Rochester to move in the exact same direction as AlphaGO.
https://github.com/Rochester-NRT/AlphaGo

Hopefully many programmers will contribute to that and eventually create something of similar or greater strength than AlphaGO but for normal PCs like that we use.

[mar]

I would say it's definitely not impossible to build an AI for StarCraft and/or StarCraft 2 that would dominate the best human players, but I think it will be very challenging to do.

The nature of such games makes it impossible to think for very long. The state changes very frequently (realtime).
Also games like that can't really afford to spend more than a couple of msec per frame (on average) when thinking;
I'd say that high level "AI" is processed at a much lower rate than 60 times per second though.
So I wouldn't compare them to chess or go.

Anyway what AlphaGO achieved is absolutely amazing, to me way more than DeepBlue in 97.
I don't understand the rules of go so I had no idea who was actually winning (I had a clue thanks to the commentary).
I would've never thought Sedol would be losing 3-0 after 3 games. So Lucas was right.
So now I wonder if Sedol can win at least 1 game.

[Uri Blass]

I would say it's definitely not impossible to build an AI for StarCraft and/or StarCraft 2 that would dominate the best human players, but I think it will be very challenging to do.

The nature of such games makes it impossible to think for very long. The state changes very frequently (realtime).
Also games like that can't really afford to spend more than a couple of msec per frame (on average) when thinking;
I'd say that high level "AI" is processed at a much lower rate than 60 times per second though.
So I wouldn't compare them to chess or go.

Anyway what AlphaGO achieved is absolutely amazing, to me way more than DeepBlue in 97.
I don't understand the rules of go so I had no idea who was actually winning (I had a clue thanks to the commentary).
I would've never thought Sedol would be losing 3-0 after 3 games. So Lucas was right.
So now I wonder if Sedol can win at least 1 game.

It is not surprising for me.
See my post at page 13 when I wrote before the first game that I will not be surprised if alphago wins every game.

[Leto]

I would say it's definitely not impossible to build an AI for StarCraft and/or StarCraft 2 that would dominate the best human players, but I think it will be very challenging to do.

The nature of such games makes it impossible to think for very long. The state changes very frequently (realtime).
Also games like that can't really afford to spend more than a couple of msec per frame (on average) when thinking;
I'd say that high level "AI" is processed at a much lower rate than 60 times per second though.
So I wouldn't compare them to chess or go.

I agree StarCraft/StarCraft 2 is a completely different type of game that cannot be solved however I do think an AI can be made for it that could dominate the best human players. Since Starcraft is a reaction type of game a neural net or advanced search techniques like those in chess would be useless. Instead I think what would need to be developed is an AI with strong strategies. Probably the biggest challenge here would be teaching it to effectively scout and how to adapt appropriately to the opponent and to what's happening.

I very much enjoy watching computers dominate humans, such as Deep Blue vs Kasparov, Watson vs the Jeopardy legends Brad Rutter and Ken Jennings, and now AlphaGo vs Lee Sedol. Watching a machine destroy StarCraft/StarCraft 2 legends will make me very happy.

By the way before they move on to StarCraft I still think there's more to do in Go, the current highest rated player is the 19 year old Chinese phenom Ke Jie, who is currently rated 3622 which is already higher than Lee Sedol's peak rating of 3553 from 2010. Ke Jie has already beaten Lee Sedol in a 5 game match recently. You can see the ratinglist here: http://www.goratings.org/