any good source to follow the SC1/SC2 bot development? V interested in following this

No way not beats me at lemmings I never last a lemming yet.

the openai team said they wanted to compete at the 5v5 level next year for dota2 which is just bat**** insane to me if they even attempt it. i keep trying to read articles/updates on their progress but i can never find anything.

AlphaGo zero is badass as ****

They took so much time to develop the previous version of AlphaGo, where the algorithm used millions of human games to learn. It was regarded as very impressive but they used a lot of expert knowledge to develop the algorithm.

And now one year later they are like: Yea we dont actually need any prior knowledge of the game, heres a version that starts from scratch that destroys anyone

both deepmind and openAI are crushing it and moving fast. Wonder what game/problem is next :P

i really wish AI can solve the GTO solution for world peace without annihilating us.

like what if it can come up with a solution so creative, we're just like duh. implement it, and boom world peace.

like right now, seems like GTO strategy is, unfortunately, loading up on weapons and threatening mutual destruction.

but that's because the genesis of this strategy is based on competition and survival that humanity had to start with. our current GTO strategy is an extrapolation of that initial survival instinct.

however, what if AI can start with an open slate (ie AlphaGoZero), without survival/competition being the genesis of its strategy, and figure out a completely new way to resolve our current situation.

just dreaming of course, but would be cool.

nice take

Sent from my SM-G930F using Tapatalk

I've read the same thing as well, that the GTO solution in nuclear arms race is keeping up the balance of terror. If you treat nuclear arms race as a prisoner's dilemma problem, you want to avoid your opponent ability to obtain a dominant position. (Both stack up weapons is equivalent of both betraying, and not allowing situation of other confessing/other staying silent to happen.)

I am currently stronger my neighbour. I want to maintain this position, because I like being dominant. I also want to protect my family. My neighbour decided to craft a sword. He is almost finished. When he finishes his sword, he will be stonger than me. What is my GTO solution? I tried talking him out of it, he didnt listen. So I crafted a sword too. What do you think happens next? The mother****er builds a second sword!!. I built a 2nd one too.

Fast forward, we both have 100 swords now. My neighbour stopped crafting swords because he realized a few extra swords wont make him dominant anymore. We are at peace now, however there is much tension in the air. My neighbour is slightly crazy. He believes im slightly crazy too. I try talking to him every day: "Lets destroy the swords !! This situation is dangerous, and crafting swords takes away all our valuable resources!! I destroy 1 of my swords, now you destroy 1 too! We can go from (100,100) to (0,0) like this: (99,100), (99,99), (98,99), (98,98).............(0,0).

My neighbour wont listen, because he knows that in the (0,0) situation I will be dominant.

WOPR solved this problem for us in 1983:

What is the solution to this problem?

U change the rules of the game.

We need to find a way to make sure the (0 swords, 0 swords) situation is a stable equilibrium. U can do this by creating incentives such that crafting a sword is no longer a good idea. These incentives come in the form of punishments.

To handle these punishments we install a 3rd party that we name the government

This has traditionally been the main problem of free markets. Sometimes they lead to bad prisoner dillema outcomes. And this is the role of every government in the world. A friendly dictator that ensures people and companies dont get stuck in these bad prisoner dillema outcomes, by punishing those that try to push towards one.

Murder and rape are GTO for some people in a society that doesnt punish them. So we have a government that punishes those people, and now murder and rape are not GTO anymore for most of these persons. (Unfortunately not everyone plays GTO :P)

In the game of country versus country we have organisations like the United nations that try to impose punishments on those countries that try to push towards bad equilibria. We dont have a world government yet tho that can enforce these rules.

A really good outcome for the world would be the case where we install a friendly dictator AI that plays this role.

Unfortunately in the short term the AIs are gonna be in control of unfriendly competitiors and we might see this arms race all over again, countries that want to become dominant will race towards developing the AI, a technology that is way harder to control than nuclear weapons.

This is the Elon musk doom scenario:

"China, Russia, soon all countries w strong computer science. Competition for AI superiority at national level most likely cause of WW3 imo."

Sidenote for people who are into cryptocurrency, blockchains are basically a new way of solving this problem and that is why they are so interesting. U have a bunch of unfriendly players in a network. they get stuck in bad prisoners dillema outcomes cuz everyone is trying to become the dominant player in the network. Those players can instead put a blockchain in charge, basically a neutral government that ensures a certain incentive system is carried out perfectly.

A really good book to read if you're interested in a game theoretical modelling of the ongoing race to AGI.

Kristol interview from yesterday

Garry Kasparov on Artificial Intelligence, Technology and Politics, and AlphaZero Chess

Love Kristol's first question... to paraphrase the long-winded opening... "Are AI enhanced computers gonna just kill us all "

I suggest AlphaGo on Netflix for your viewing pleasure.

^^

Just watched. Amazing. Thanks for sharing.

Watching that made me even more curious about these new algorithms and how they will impact future. The opening scene with breakout game was such a great way to start the documentary.

Kasporov had it right. Its not binary where this tech is either good or bad, either machines take over the world and kill us or they don't.

These new algorithms are just tools and will benefit many things in our lives.

Any tool can be used for good and bad.

Welcome to the beginning of the beginning.

Just watched, awesome documentary.

Completely agree with you.

I look forward to seeing AlphaZero at the bum stakes on ACR.

AI can't learn dialects. Kurzweil's onthology is childish and naive...
Therefore I dought that your wish will ever come to fruition.

+1 great documentary!

What is the alphago zero estimated elo?

People always underestimate computer capabilities. When I was a kid (I'm 62) I remember all of the arguments about whether a computer could beat a human in chess. The consensus was that it would never happen.

That would be Mr Blue Sky thinking.

Bumping for AlphaZero, which crushes AlphaGo Zero at Go, Stockfish (even with it's opening book) at chess, and Elmo at shogi.

"In chess, AlphaZero defeated Stockfish, winning 155 games and losing 6 games out of 1,000 (Fig. 2). To verify the robustness of AlphaZero, we played additional matches that started from common human openings (Fig. 3). AlphaZero defeated Stockfish in each opening, suggesting that AlphaZero has mastered a wide spectrum of chess play. The frequency plots in Fig. 3 and the timeline in fig. S2 show that common human openings were independently discovered and played frequently by AlphaZero during self-play training. We also played a match that started from the set of opening positions used in the 2016 TCEC world championship; AlphaZero won convincingly in this match too (26) (see fig. S4). We played additional matches against the most recent development version of Stockfish (27), and a variant of Stockfish that uses a strong opening book (28). AlphaZero won all matches by a large margin (Fig. 2)."

The above paragraph is from the DeepMind manuscript that was just published in Science, you can download it from the DeepMind website: https://deepmind.com/research/publications/

Here's a couple of other good reads:

Essay in the NY Times today


Blogpost on DeepMind's website

Finally, here's a link to a 2+2 thread on AlphaZero crushing Stockfish where it didn't have it's opening book, so many complained it was not at full strength.

Cliffs: AlphaZero rules! The most amazing thing to me is that it somehow lost 2 out of 500 games of chess while playing the white pieces...