Google's DeepMind AlphaGo beats world champion Lee Sedol

[UndrgroundRider]

Not sure if anyone else watched the live stream tonight, but Google's DeepMind AlphaGo AI beat 18-time world champion Lee Sedol in an even game. It's a HUGE milestone for AI technology. A year ago anyone who was anyone who knew about Go and AI would have told you we're at least 10-20 years away (or never) from a computer beating top rated humans.

I think more importantly than that, was the fact that this was not an algorithmic approach like Deep Blue which famously beat Gary Kasparov at chess in the 90's. DeepMind instead uses a neural network which allows it to actually learn and progressively improve its ability. AI neural networks are based on our current understanding of how human brains function. They simulate the same sort of neuronal interactions within the computer. For the technology to have advanced this far is nothing short of amazing.

Background:

"The number of possible combinations of the board is more than the number of atoms in the universe."




Match 1:


Match 2:


Match 3:


Match 4:


Match 5:

[Dumbass17]

What?

[Feruk]

Well said... Dumbass17

[taemo]

watched this couple of weeks ago


didn't know it was happening know

[UndrgroundRider]

Originally posted by Dumbass17
What?

Did I stutter?

[04_dc5]

2-0 after todays match.

Probably going to be 5-0 for the machine?

[OU812]

Skynet is born!

[n1zm0]

2018: Google DeepMind contracted by US DoD for work on drones.

2020: All drones in the US DoD arsenal are fully autonomous on decisions, engagement and battle tactics.

2031: Google DeepMind becomes self-aware.

Jokes aside this is pretty crazy, no algorithms, just a computer thinking like a human would from adaptive learning, albeit initially only for a thousands of years old East Asian board game. It can only lead to amazing things in the future for AI as a whole, unless Google hogs it all.

The way it's going, I think the computer will win all the rounds.

[Sugarphreak]

...

[pheoxs]

What amazes me most about this and how it uses a neural network is not that it beat him, its that the computer came up with moves that no one ever though to try before, and although it didn't make any sense at the time, once the game played out experts are now analyzing AlphaGo's moves and discovering better strategies that no one had previously come up with.

The really interesting thing is they can't figure out logically how AlphaGo arrived at these strategies because no one had used them before, just that it did indeed figure it out.

[g-m]

Larry Page read it Sun Tzu every night before bed

[lilmira]

Can it make a kickass sandwich?

[sputnik]

Way to go Ultron/Ava/Jarvis!

[RealJimmyJames]

Originally posted by Dumbass17
What?

[RealJimmyJames]

Originally posted by Sugarphreak
It won't be long before Google and Amazon have delivery centers complete with drones that completely run on AI and robotics

I would assume that he primary objective of these types of AI’s will be to maximize profits through efficiency. However it will be interesting once the AI systems become aware of each other, and realize that they can increase profit margins by disrupting or eliminating the competition. It might give rise to a commercial AI cold war.

Once the drone-AI-net figures out what percentage of packages constitute acceptable losses, I assume the drones will start kamikaze-dive-bombing the competition, and the war will go hot. It only makes sense for the drones to have some kind of "defensive" weaponry at that point.

I see this ending well.

[Modelexis]

Originally posted by n1zm0
no algorithms, just a computer thinking like a human would from adaptive learning

"AlphaGo's algorithm uses a combination of machine learning and tree search techniques, combined with extensive training, both from human and computer play. The system's neural networks were initially bootstrapped from human game-play expertise. AlphaGo was initially trained to mimic human play by attempting to match the moves of expert players from recorded historical games, using a database of around 30 million moves."

So yes, it does use algorithms.

[UndrgroundRider]

Wow. AlphaGo wins the third straight game in a row to win the match. Crazy complicated fighting too. AlphaGo's tenuki when Lee Sedol was still invading.

[UndrgroundRider]

Originally posted by Modelexis
So yes, it does use algorithms.

You're not wrong, but most people consider algorithms to be rigidly defined functions and processes. The power of AlphaGo comes from its neural network, which is software simulating neurons as they would work in a human/animal brain. Initially this neural network is not very good, but it learns by playing/reviewing many games. This is the main difference between a traditional algorithm and the neural network, that the neural network learns and advances over time.

[Sugarphreak]

...

[HiTempguy1]

Originally posted by UndrgroundRider


You're not wrong, but most people consider algorithms to be rigidly defined functions and processes. The power of AlphaGo comes from its neural network, which is software simulating neurons as they would work in a human/animal brain. Initially this neural network is not very good, but it learns by playing/reviewing many games. This is the main difference between a traditional algorithm and the neural network, that the neural network learns and advances over time.

There is a demo on youtube of them using a neural network to create an AI that can play Super Mario Bros and Super Mario World.

Watching it learn is fascinating (and I find it a bit easier to follow since there are only the inputs of the controller and relatively few options for fighting).