It took Alpha Zero 4 hours of training to become world's best in chess...

It took Alpha Zero 4 hours of training to become world's best in chess. But in speedrunning people use a lot of unusual strategies and exploits/glitches which makes me think it would be harder for it than a strict game like chess. If we could challenge Alpha Zero to beat the speedrun world record of any game what game would it be?

Other urls found in this thread:

wiki.teamliquid.net/dota2/Patches
twitter.com/AnonBabble

Mario Bros. 3

Any game that gets broken by TAS

What about Dragster?

Super Mario Bros

It can't play 70 thousand games of any actual video games without spending years.

Boardgames have simple rules that can be quickly simulated.

It wouldn't make sense to challenge Alpha Zero to this at all.

There's something fucking unsettling about this image, can't really put my finger on it.

versus a gimped stockfish

Dota 2 had a similar projects where it only requires a few weeks of training to beat the best dota players in the world

Most 2D sidescrollers are not complicated. It's visually differentiating the hazards from the background and knowing how to navigate. That's all that required and it could easily be adapted to a neural net such as AlphaGo. It would probably take some time to learn what is what.

No, stop, this is misleading stupid bullshit.
It beats the best players in the world...
>in a 1v1
>with everything that the computer can't handle disabled
>locked to one specific character that doesn't get played often
>the locked character pretty much just rewards micromanagement
>the locked character is extremely straightforward in the made-up 1v1 gamemode they picked
>the AI pretty much just exploits super-fine micromanagement that's impossible for a human to achieve and would be impossible for the AI too if it didn't have 0 ping
and OH WAIT
IT STILL LOST ONCE THE PLAYERS GOT USED TO THIS RETARDED MADE UP GAMEMODE

That, and again, Alpha Zero played thousands of chess games in four hours. Maybe millions.

A Mario game would take the time a Mario game takes. On top of the many many more rules.

well they said that they are going to play 5v5 next year. but we'll see how ithat goes

Cant you speed up a game like Super Mario Bros to make it practise faster?

It's unlikely to go well.

Besides, how frequent are balancing updates?

Chess is a really simple game with a finite amount of legal moves and solutions.

I suppose.

You could likely run it and several thousand times the speed now.

Depends entirely on the game. Something like Mario would be trivial, as long as the computer understood the function of warp pipes.

It's a very simple game where the goal is simply to hold right and B as much as possible. I doubt it'd even need a day of replaying frame data with real time rewinding to break the current human WR, but breaking the current TAS record might be hard. Give it a week and maybe it'll find a new strat for that.

This shit is coming on in leaps and bounds, and it's absolutely bone-chilling. We'll probably all be dead in 20 years when the Googlebots come round.

wiki.teamliquid.net/dota2/Patches

Strong general-purpose AI is decades away; a self-aware Skynet may be impossible. Perhaps brains are the only computers capable of self-awareness. If so it would require grafting neural tissue to a computer system to make it self-aware.

Weak AI that convincingly simulates a human is probably only 20-30 years out, depending, and will see it's first consumer application in OS's and digital assistants. Essentially, Knight Rider.

Portal

Chess isn't that hard to master. The reason why that robot could beat chess is because the robot could think of a million patterns.

>Perhaps brains are the only computers capable of self-awareness.
Who even types shite like this? Sure, the potential for neuronal growth and connection development is a facet still poorly understood, but even if it requires a few trillion transistors to emulate a single human brain, even if it takes an entire building's worth of high tech servers to achieve that kind of response in real time, what's stopping a multi billion dollar company like Google from just connecting their super computer to the one next door at Amazon?

Organic computing is another field entirely, and it's conceivable that the minds of world ending level intelligence will be grown rather than built. But AI is the ember that starts the fire that consumes all.

Eh.

We still don't know why a neuron fires.

Thinking we can achieve something remotely similar to actual sapience with computers and algorithms as we construct them now is a little optimistic.

The robots claiming victory. The retards over at Boston Dynamics bullied them so much we'll be lucky if they don't just wipe us all out.
I for one, welcome our robot overlords and beg for their mercy.

For me as a visual designer it's just the way the fingers and the Google logo almost touch.

>but even if it requires a few trillion transistors to emulate a single human brain

Pretty sure someone said something similar back when all they had to work with were gears and shafts.

Seems like a lot of people are missing what I think is the point of the thread.

I think the next step for a gaming robot like this is to tackle a game with limited play space. It could easily dominate fields such as Tetris with perfect algorithmic thinking and timing. There's no human strategy that could out perform a robot in Tetris.

Then you probably need to read up on basic physiology. Automated or sponatneous neuron firing can be explained by something as "basic" as the heart's sinus node.
Memory and consciousness, however, is another story. Very hard to explain without someone finding holes in your hypothesis.