Originally posted by Meowhead

'On August 7, 2008, the computer program MoGo running on 25 nodes (800 cores, 4 cpus per node with each core running at 4.7 GHz to produce 15 Teraflops) of the Huygens cluster in Amsterdam beat professional Go player Myungwan Kim (8p) in a nine stone handicap game on the 19x19 board on the KGS Go Server. MoGo won by 1.5 points. Mr. Kim used around 13 minutes of time while MoGo took around 55'

 

"On July 28, 2010, at the 2010 European Go Congress in Finland computer program MogoTW played European professional 5 dan Catalin Taranu 19x19 Go. MogoTW had a 7 stone handicap. The computer won. MogoTW is a joint project between the MoGo team and a Taiwanese team.

On a small-sized 9x9 go boards, the computers are already faring very well playing almost evenly against their master level human opponents. "

It is with very nice computers though.  The progress being made is pretty fast though, so that's why I give it 5-10 years before a computer can hold its own against the masters.

Originally posted by Meowhead

Originally posted by tehm

On the human side maybe...

The only thing allowing humans to win at ALL right now are the time constraints they place on the tournaments and the fact that they invariably limit the machines running the programs to quad core "desktop style" machines.

If you either expand the time limits to hour/move OR allow the go programs access to "real" hardware then humans would already be losing left and right.

 

<.<

 

Monte-Carlo method is ridiculously good at finding solutions to basically ANY board-game style game which has terminal positions which can be evaluated. That includes Go.

'On August 7, 2008, the computer program MoGo running on 25 nodes (800 cores, 4 cpus per node with each core running at 4.7 GHz to produce 15 Teraflops) of the Huygens cluster in Amsterdam beat professional Go player Myungwan Kim (8p) in a nine stone handicap game on the 19x19 board on the KGS Go Server. MoGo won by 1.5 points. Mr. Kim used around 13 minutes of time while MoGo took around 55'

... that's what wicked-ass quad core desktop style machine.  What the heck kind of computer do you have on YOUR desktop?

MoGo also uses Monte-Carlo method.  In fact, it uses a lot of AI methods, and an improved version of the Monte-Carlo method.  It had a =9= stone handicap.  ... and used 3 times as much time as the human being.

Yeah, computers could win if they kept hitting the human with a giant stick in the crotch, but any time you start handicapping the human players OVER the computer players to let the computers win, are they computers really winning? :P  Why should the computer have any less of a time handicap than a human?

MoGo by the way won the Go Computer Olympiad, with AIs duking it out against each other, in a contest to see which is the best AI.  Not even a contest involving humans at all. :P

That was right after the switch in algorithms...

By July 2010 a computer had beat the best European master in the world at 19x19 at 7 stone.

By the Tokyo cup (Dec 2010) a computer beat Aoba(4p) at 6 stone using only 6 Xeons.

In February of this year a computer beat Chou (5p) at 0 stone.

Handicapping in Go isn't like giving pieces away in Chess. It's part of the game. At amature ranks handicap is essentially how the rating system works even. 

TL;DR It's not the power of the computers that are increasing. It's the algorithms of the programs getting better. They're starting to get crazy good. You can't base your information on stuff from even 3 months ago, much less 3 years ago.

Originally posted by nordic135

Originally posted by gobla

Originally posted by nordic135

It's not tracking a bird. It's tracking a pixel pattern. I know that doesn't sound like a meaningfull diffirence to you but if you're really interested in AI ( I = Intelligence ) then it does matter a LOT.

Read my post above on how optimised the human brain is.

No, it was real life.  At Boeing.  It was a bird.

A computer doesn't know what a bird is.

A computer gets input. Most likely pixels. It could also be a sonic image or something like that.

It gets pixels or some other form of data. Some human specifies a pattern in that data. The computer then follows that pattern as long as it doesn't change in too extreme a fashion.

You're humanising computers. You're attaching human reasoning patterns to a CPU. It doesn't work that way. A computer doesn't aim. A computer doesn't know what a bird is, let alone how to track one.

Following that pixel pattern is a fine piece of AI. It's an impressive thing for a computer to do. But it doesn't come close to what people do. Computers can't beat people. Not if they're both playing the same game. Not yet. Maybe in a few hundred years.

Humans track pixel patterns too.  What you see of the world is pixel patterns.  Your eye only has a finite resolution.  It is the same thing as you tracking a bird as it flies throught the sky, except that the decision making of whether to track it or not is done by a human for the computer, and its done at a distance that you couldn't even see the bird without optical help.  And yes, it does know how to track a bird/object (it didn't have to be a bird, it could track a person or car just as easily if that's what you chose to lock onto), it just doesn't know how to make the decision of whether to track it or not.

Ofcourse my vision is pixels ( or the neural equivalent ). The diffirence is that I know it's a bird. I know it's flying. I know it's in the sky. I know I'm tracking it.

That computer isn't tracking a bird as it flies through the sky. It doesn't have the context of bird, sky, tracking and flying.

That computer is doing some ( for us ) basic pattern recognition of a small group of pixels in a larger group of pixels.

It's what our eyes are constantly doing when looking at the world around us. Not just for one thing, but for hundreds of things. It's how we recognise everything in the world around us. It's how you read the text on your screen. You're doing pattern recognition on each individual letter. At least after you've done pattern recognition on your screen to determine which pixels are letters and which aren't. Now start counting every single letter on your screen. Add to that everything else you recognise. You're no doubt seeing countless buttons which you recognise as buttons. You're seeing taskbars, images, links etc. You know what everything means. All those thousands of little components on your screen. You're tracking them all. If your mouse moves there's not a single doubt in your mind where it goes. When you scroll and all those thousand components go up you've no trouble determining that they went up.

Now find me a tracking computer able to track thousands of patterns like that at the same time in real time.

Originally posted by gobla

 

Ofcourse my vision is pixels ( or the neural equivalent ). The diffirence is that I know it's a bird. I know it's flying. I know it's in the sky. I know I'm tracking it.

That computer isn't tracking a bird as it flies through the sky. It doesn't have the context of bird, sky, tracking and flying.

That computer is doing some ( for us ) basic pattern recognition of a small group of pixels in a larger group of pixels.

It's what our eyes are constantly doing when looking at the world around us. Not just for one thing, but for hundreds of things. It's how we recognise everything in the world around us. It's how you read the text on your screen. You're doing pattern recognition on each individual letter. At least after you've done pattern recognition on your screen to determine which pixels are letters and which aren't. Now start counting every single letter on your screen. Add to that everything else you recognise. You're no doubt seeing countless buttons which you recognise as buttons. You're seeing taskbars, images, links etc. You know what everything means. All those thousands of little components on your screen. You're tracking them all. If your mouse moves there's not a single doubt in your mind where it goes. When you scroll and all those thousand components go up you've no trouble determining that they went up.

Now find me a tracking computer able to track thousands of patterns like that at the same time in real time.

I don't think you get it.  Yes, the computer AI isn't good at everything.  It can't process all kinds of human thought, but it is good at certain things, and those things that you optimize the AI for can be and very often are better than humans at those things.  Even Luke Skywalker has to cheat and use the force when trying to hit the thermal exhaust port of the DeathStar with his proton torpedos.  The rest of us have to use the computer, because it can aim better.  

I'm kidding obviously, but pilots use computers to track their targets, why is that do you think?

Originally posted by nordic135

I don't think you get it.  Yes, the computer AI isn't good at everything.  It can't process all kinds of human thought, but it is good at certain things, and those things that you optimize the AI for can be and very often are better than humans at those things.  Even Luke Skywalker has to cheat and use the force when trying to hit the thermal exhaust port of the DeathStar with his proton torpedos.  The rest of us have to use the computer, because it can aim better.  

I'm kidding obviously, but pilots use computers to track their targets, why is that do you think?

I also use a shovel to dig holes in the ground. That doesn't make that shovel intelligent, strong, fast or better.

Tracking computers achieve what we need them to achieve. A pilot with a tracking computer performs better then one without. Not denying that.

That tracking computer is used exactly because it's not as fast and efficient as the human brain. If it was then it'd have developed consciousness and all sorts of things you really don't want your tracking computer to have. Because a tracking computer is so extremely slow and inefficient it's able to track that single target and not do anything else.

Originally posted by gobla

I also use a shovel to dig holes in the ground. That doesn't make that shovel intelligent, strong, fast or better.

Tracking computers achieve what we need them to achieve. A pilot with a tracking computer performs better then one without. Not denying that.

That tracking computer is used exactly because it's not as fast and efficient as the human brain. If it was then it'd have developed consciousness and all sorts of things you really don't want your tracking computer to have. Because a tracking computer is so extremely slow and inefficient it's able to track that single target and not do anything else.

No, it's because the tracking computer is so fast and efficient at what it is designed to do, track targets.  So much faster, more efficient and more accurate than a human that we don't trust the human to target anymore and let the computer do it.  

It doesn't have consciousness, but that's not what it's designed for.  It couldn't even if we wanted it to right now, but that's beside the point.

Edit: In fact even if we could we would still use AI that is designed for one specific task for many tasks because it would stil be much more efficient and faster.

Originally posted by nordic135

No, it's because the tracking computer is so fast and efficient at what it is designed to do, track targets.  So much faster, more efficient and more accurate than a human that we don't trust the human to target anymore and let the computer do it.  

 

It doesn't have consciousness, but that's not what it's designed for.  It couldn't even if we wanted it to right now, but that's beside the point.

Let's take a diffirent angle.

We've got the human brain with hardware and software so advanced that nobody understands it.

Then we've got computers with hardware and software so basic that a lot of people understand it.

Now the reason why that tracking computer is so "fast" and "efficient" is because it really is so very slow and inefficient.

If it was as fast and efficient as the human brain we wouldn't be able to program it and it'd be useless. But because computers are horribly slow and inefficient we can understand them and program them.

If we were able to program neurons do you honestly think we'd still be using CPUs at all?

Computers aren't better at anything at all then brains. The only reason they appear to be is because we understand them and thus are able to view and change their inner workings.

Brain hardware is better then computer hardware. Brain software is better then computer software. Brain hardware and brain software combined is a LOT better then computer hardware and software combined. It's just that developing brain hardware and brain software takes a few million years of evolution ( or alternatively the hand of god if that's what you believe ).

So we settle for the practical solution. Computers. Not becaue they're better, faster or more efficient. But because they're worse, slower and far less efficient. Because they're easy.

 Just because the computer can't do all the things a human brain can do does not make it slow and inefficient.  In fact computers are much much much much much much faster at doing a lot of things than even the best human brain is capable of.  It may not have the overall processing power of the brain, or even close to as good overall intelligence, but in specific areas the computer can and does dominate the human brain.  

Originally posted by nordic135

 Just because the computer can't do all the things a human brain can do does not make it slow and inefficient.  In fact computers are much much much much much much faster at doing a lot of things than even the best human brain is capable of.  It may not have the overall processing power of the brain, or even close to as good overall intelligence, but in specific areas the computer can and does dominate the human brain.  

A human brain takes a few square centimeters in size. It's capable of more calculations, processes and activity then a classroom full of computers.

Where exactly are computers faster and more efficient?

The only area in which computers are better is that they're easier. Easy to understand and easy to program.

The problem with better AI might be this. Once an npc is smart enough to know when to interrupt you, and heal itself, and so and so and so, it's really just a numbers game isn't it? Can you heal and dps faster than it can?

Faction champions was one ofthe most difficult encounters in Wrath-WoW because you were faced an arena team of like 6 npc enemies with a combination of almost anything, never using the same more than once (you wouldn't face a team with two healing druids, but you could face a resto and moonkin--dps) and almost any role of any class (there weren't tanks).

There were a few minor exploits but it was really about laying good dps on a set order of targets, they cc'd you, interrupted, and would focus your clothies at random. They also had 15x (10 man normal) - 140x (25 man heroic) more health, and healed much harder.

The only substantial exploit was that npc pets could take priority over npc team members occupying some of their healer's cast times, often making the difference between the target dying, or regaining a chunk of health. Some groups used this, some did not. Sick several pets on one or two enemy pets, rolling things like autogrowl as well.

In regular mode they were tauntable but they changed targets as they saw fit, and often and in heroic mode, nothing but the pets were tauntable.

Originally posted by Eronakis

The argument should be null about whether it can or can't be done. It's all subjective based on the players preference. This thread is for the players who want to get away from the linear AI into something more adaptive, strategic and dynamic. Can a Ai be implemented as such? Don't know yet, because it hasn't been done before. Why? Because of the server descrepencies, lack of budget and time and possibly technology.

Here is the simple answer, I've worked with professors who research AI, and does consulting for EA and other Canadian companies.

He's done a lot of work with Bioware and here is the main, glaring reason.

Nobody wants it. AI isn't a selling feature worth investing in. Current AI is handled by 1 developer generally, maybe 2 and it might just be a side job of 1 developer.

Assuming you pay a developer 80k thats 40-120k development cost per year on AI

To get better more interesting AI you would need more people, specialized people. AI developers in other industries pull very large salaries, and if you wanted a team of 3-4 guys working on it, you'd be looking at 300-400k+ a year in development costs.

Is a better AI going to sell the extra 100,000 copies needed to make it profitable? Proably not, players would much rather see better graphics, physics or longer stories.

Currently, AI is not a finacially viable area to push in game development, this may change. For example you've all no doubt seen F.E.A.R which moved videogame AI from the 1940's all the way into the 1970's. Thats now the standard, but has anybody really pushed AI to see if there is a market? No. There may or may not not be a market for it, but nobody wants to take the financial risk.

I simply can't believe the AI to be moderately challenging is difficult. As I said above, faction champions demonstrated enough to be a challenge. It would just end in long grueling duels and would be a punishing experience. It basically comes down to if/else statements for mobs.

if health > Low_Health% then

-- loop from 1 to 50 as loop_int

-- -- if mobClassAttacks[loop_int] cd = 0 then

-- -- -- do mobClassAttacks[loop_int]

-- -- -- break from the loop

-- -- end if

-- end loop

else

-- If Heal_CD = 0 then

-- -- do Heal

-- else

-- -- interrupt / stun / defensive cd while waiting for Heal to come off cd

-- end if

end if

(I hate to break it down into a simple and terrible representation of programming but it really is not difficult to expect, it just wouldn't be any fun to grind on. Add in the ability for the mobs to check their health and all others engaged in combat nearby and soloing two or three becomes well...predictable.

It's also true that many mobs wouldn't be able to heal and have defensive cool downs and interrupts, but I can name a few games where classes a few classes have all 3, or did when I played.

And then of course, there's the whole issue of threat. Why is the the boss attacking the tank when giving the dude over there a quick swipe, the one with the white magic hands, would end the fight.

PS, kinda wanted to write this in lolcode

Until I see a Starcraft AI that is remotely as skilled as a halfway decent human player, then I'll believe that computer opponents can actually compete at a game that requires creative strategy or tactics. The fact is that computer scripts by definition are entirely predictable, so no matter how "adaptable" you make an AI eventually players will learn that adaptation and how to break it. Computers will never have that spark of inspiration or genius that humans can have. One a computer's strategy is learnt, it's nothing more than a virtual punching bag.

Perhaps a computer could at least be an effective source of entertainment for moderately skilled players by attempting to emulate the actions used by human players which have defeated it. Eventually it may develop a network of strategies designed to counteract almost every conceivable character build players may attack them with. This would certainly at least make it a new experience for each character, keeping the game interesting as players try new tactics.

Originally posted by Munki

Originally posted by Eronakis

The argument should be null about whether it can or can't be done. It's all subjective based on the players preference. This thread is for the players who want to get away from the linear AI into something more adaptive, strategic and dynamic. Can a Ai be implemented as such? Don't know yet, because it hasn't been done before. Why? Because of the server descrepencies, lack of budget and time and possibly technology.

Here is the simple answer, I've worked with professors who research AI, and does consulting for EA and other Canadian companies.

He's done a lot of work with Bioware and here is the main, glaring reason.

Nobody wants it. AI isn't a selling feature worth investing in. Current AI is handled by 1 developer generally, maybe 2 and it might just be a side job of 1 developer.

Assuming you pay a developer 80k thats 40-120k development cost per year on AI

To get better more interesting AI you would need more people, specialized people. AI developers in other industries pull very large salaries, and if you wanted a team of 3-4 guys working on it, you'd be looking at 300-400k+ a year in development costs.

Is a better AI going to sell the extra 100,000 copies needed to make it profitable? Proably not, players would much rather see better graphics, physics or longer stories.

Currently, AI is not a finacially viable area to push in game development, this may change. For example you've all no doubt seen F.E.A.R which moved videogame AI from the 1940's all the way into the 1970's. Thats now the standard, but has anybody really pushed AI to see if there is a market? No. There may or may not not be a market for it, but nobody wants to take the financial risk.

Very intresting post. I thought as much. My idea is this. You have class design in one spectrum and combat mechanics in one spectrum. Combat mechanics are a separate entity than class mechanics but compliment each other. Say the combat mechanics give the player a dozen options to win a fight instead of spamming your 1 2 and 3 hot keys for your abilities. The way I envision a more intelligent AI is, simply it would adapt within the realm of the combat mechanics and detect what type of archetype that class is based on their abililties used within a fight. Lets take this for an example.

Say for melee combat mechanics you have 3 stances, High for an offensive strike, Low for defending such as parrying and Middle for striking and defending, but is not efficient with either defending or striking. For simplicity, let's say that PC is approaching a hut with 1 guard. The NPC is in Middle stance posture. This means, the guard is not looking to defend nor strike. The PC may not know the strategy of this particular guard. The PC decides to strike in High Stance position to have a higher chance to perform an offensive strike.

Strike is successful because NPC did not have a successfull Parry. PC decides to go again in High stance and strike again with another successful strike. NPC AI says (dependent on their hierarchy of intelligence) detects the patter than this PC is going to be using offensive strikes. So the NPC will now become conserveritve and engange in Low stance postition for defense until they have an attack of opportunity.

If the PC would of went with Low stance after the offesnive strike, it might of thrown the NPC's AI off to not detect a particular strategy. The AI would detect strategy based on angel of weapon and the collision of weapons hitting, NPC body, or weapon or sheild.

That is what I am envisioning in a very basic form.

Originally posted by gobla

First, just so we're on the same line, human behaviour is based on nothing more then if-then statements. If one of my neurons gets enough signals through it's dendrites then it will fire a signal through it's axon. That's the basis of all human behaviour. One big system of electrochemical triggers.

Wow, talk about wrong.  That is how the mechanics of the body works, any body.  From human to dog to computer.  What you do with those signals is what makes intelligence, artificial or otherwise.

I hit my thumb with a hammer.  All the right nerves fire and I feel pain.  There are several responses I could have though.  1) Scream and throw the hammer. 2) Drop the hammer and cuss. 3) Put the hammer down and treat the injury.  Our thoughts (cognition or intelligence) is what determines how we react.

Originally posted by Erolat

Wow, talk about wrong.  That is how the mechanics of the body works, any body.  From human to dog to computer.  What you do with those signals is what makes intelligence, artificial or otherwise.

I hit my thumb with a hammer.  All the right nerves fire and I feel pain.  There are several responses I could have though.  1) Scream and throw the hammer. 2) Drop the hammer and cuss. 3) Put the hammer down and treat the injury.  Our thoughts (cognition or intelligence) is what determines how we react.

And what do you think causes those responses? What do you think that cognition is? What do you think intelligence is? If not the unbelievably complex and advanced combination of a 100 billion neurons working in tandem then what then?

I believe in causality. Meaning that everything has a cause and an effect. All my thoughts have a cause and an effect. All my actions have a cause and an effect. I don't believe that I'm somehow magically excluded from the laws of physics.

The moment I drop that belief in causality and start believing that my thoughts have no cause and are somehow magically "free" then, logically, I also have to accept that all things could possibly be without cause. And all rational thought goes out the window as all logic is based on cause and effect. All science is based on cause and effect.

If you want to believe that we're more then this amazing combination of trillions of cells working in perfect unison then that's fine. But be consistent and also throw science and rationality out.

Originally posted by Erolat

Originally posted by gobla

First, just so we're on the same line, human behaviour is based on nothing more then if-then statements. If one of my neurons gets enough signals through it's dendrites then it will fire a signal through it's axon. That's the basis of all human behaviour. One big system of electrochemical triggers.

 

Wow, talk about wrong.  That is how the mechanics of the body works, any body.  From human to dog to computer.  What you do with those signals is what makes intelligence, artificial or otherwise.

I hit my thumb with a hammer.  All the right nerves fire and I feel pain.  There are several responses I could have though.  1) Scream and throw the hammer. 2) Drop the hammer and cuss. 3) Put the hammer down and treat the injury.  Our thoughts (cognition or intelligence) is what determines how we react.

 And every single one of those responses is the result of an action potential, a bio-electrical stimulas travelling down an axon of a nerve that received enough stimulus to generate said potential. 

That's an if-then statement, if the nerve receives said stimulus, an actional potential generates that triggers said action.  The only question is which series of nerves will fire, that determines the response.

Venge

Originally posted by gobla

If you want to believe that we're more then this amazing combination of trillions of cells working in perfect unison then that's fine. But be consistent and also throw science and rationality out.

"Fun" isn't really a big factor in this thread, is it?

All the people asking for skilled AI need to go play 20 games of Chess against the best Chess AI in the world, then play 20 more games against an AI matched with their own skill, and tell me which AI was more fun: the "smart" or the "stupid"?

It's important for things to be dynamic/unpredictable enough to keep players learning, but it's cheaper (and therefore more efficient) to implement this unpredictability mostly through mob abilities.  It's also far clearer exactly how the mob is different (this one casts a snaring frostbolt vs. the one who cast a DoT fireball.)

Especially when it comes to this talk of mobs running away or healing (and let's hopefully not even mention CC), all these things have long track records of pissing off players and generally not being fun to fight against unless it's a clear combat puzzle to unravel (Oh I can interrupt the heal!)

If we're suggesting giving players completely new toolsets from existing MMORPGs to deal with specific (and very visual) actions that mobs do, then that's great.  As long as we're not suggesting mobs are completely random or completely smart (because neither of those things makes mobs more fun.)

Originally posted by Axehilt

"Fun" isn't really a big factor in this thread, is it?

All the people asking for skilled AI need to go play 20 games of Chess against the best Chess AI in the world, then play 20 more games against an AI matched with their own skill, and tell me which AI was more fun: the "smart" or the "stupid"?

It's important for things to be dynamic/unpredictable enough to keep players learning, but it's cheaper (and therefore more efficient) to implement this unpredictability mostly through mob abilities.  It's also far clearer exactly how the mob is different (this one casts a snaring frostbolt vs. the one who cast a DoT fireball.)

Especially when it comes to this talk of mobs running away or healing (and let's hopefully not even mention CC), all these things have long track records of pissing off players and generally not being fun to fight against unless it's a clear combat puzzle to unravel (Oh I can interrupt the heal!)

If we're suggesting giving players completely new toolsets from existing MMORPGs to deal with specific (and very visual) actions that mobs do, then that's great.  As long as we're not suggesting mobs are completely random or completely smart (because neither of those things makes mobs more fun.)

I was speaking along the lines based on a combat mechanic design I have. Go back and read my posts in this thread and you should convay an overall intention.

Fun is very subjective. WoW is not fun at all to me, but on the other hand, WoW is fun to you. Smart AI can happen. I wasn't taking about a Winston or whatever they call that robot that can guess jeopordy answers or whatever. I think one has to comprehend what linear combat mechanics mean in order to see what I am talking about.