136 replies to this topic

[MuonNeutrino]

xhrit, on 05 April 2013 - 09:03 AM, said:

Do you know how AI works? It has a big list of things to "ALWAYS DO" based on a list of things "IT DEPENDS" on.

This is a set of instructions that is always rigidly adhered to.

Any sufficiently advanced AI is unbeatable by humans.

AI is based on "ALWAYS DO A THING" philosophy, and is unbeatable by humans.

Sufficiently advanced AI is indeed often unbeatable. However, in a game like this if it is unbeatable it is generally because it is far, far faster than a human can ever hope to be, does not have to deal with the clumsiness of a user interface, and never fails to perform the action it intended to. None of those have anything to do with the philosophy behind how it makes decisions. It's like someone running an aimbot in a game of counter-strike - they win because the millisecond the tiniest inch of their enemy shows around the corner they put bullets into it. Their decision-making is irrelevant. The games where decision-making skills are actually what make AI unbeatable are rigidly circumscribed ones like chess or turn-based strategy.

Edited by MuonNeutrino, 05 April 2013 - 02:55 PM.

[xhrit]

MuonNeutrino, on 05 April 2013 - 02:54 PM, said:

Sufficiently advanced AI is indeed often unbeatable. However, in a game like this if it is unbeatable it is generally because it is far, far faster than a human can ever hope to be, does not have to deal with the clumsiness of a user interface, and never fails to perform the action it intended to.

Machine like reflexes, near instantaneous response time, and total clarity of action.

Sounds like Mushin to me.

MuonNeutrino, on 05 April 2013 - 02:54 PM, said:

None of those have anything to do with the philosophy behind how it makes decisions. It's like someone running an aimbot in a game of counter-strike - they win because the millisecond the tiniest inch of their enemy shows around the corner they put bullets into it.

And how many people here have been accused of haxs in some game or another because their skills were so advanced they were mistaken for a machine?

MuonNeutrino, on 05 April 2013 - 02:54 PM, said:

Their decision-making is irrelevant. The games where decision-making skills are actually what make AI unbeatable are rigidly circumscribed ones like chess or turn-based strategy.

Given enough time to collect enough information, unbeatable AI can be created for every game. After only one month of development students were able to create a Starcraft AI that used the exact same rulesets as human players; no cheating or shortcuts, but using the same data available to humans.

Overmind AI is capable of high-level strategic planning, managing diverse groups of unit types, and information collection and management. The highest rated starcraft players in the world were able to defeat the student created AI in a close match. Average players were gang ***** and left dead in a ditch. In a few years there will be agents that can consistently compete against the highest levels of human competition; just like how chess AI was soundly defeated by humans... until enough data about the game was mapped. Once all variables are known, the AI will be nearly invincible against meatspace intelligence.

Do you know the origin of the word "computer"?

It was was invented in the 17th century to describe humans.

Human computers.

We are all just machines.

Edited by xhrit, 05 April 2013 - 03:42 PM.

[Smk]

The JM6-A with like 450 damage. I have several mechs with more total damage in less games.

[MuonNeutrino]

xhrit, on 05 April 2013 - 03:24 PM, said:

Machine like reflexes, near instantaneous response time, and total clarity of action.

Sounds like Mushin to me.

BS, frankly. Humans can train themselves to try and speed reactions, reduce mistakes, and improve precision. They will still never even remotely *approach* a computer's ability to do the same thing, let alone equal it. Human response time is ultimately strictly limited by physiological factors to a value FAR slower than what a computer can do. A human can train to improve their hand speed. A computer requires no training at all to send dozens of commands simultaneously at a speed far greater than the fastest motion the human body is capable of. A human may practice to less frequently hit the wrong key or have their hand slip on the mouse. A computer requires no practice at all to *never* hit the wrong key or have their hand slip.

And none of that has anything to do with AI decision-making, which was the original point.

Quote

And how many people here have been accused of haxs in some game or another because their skills were so advanced they were mistaken for a machine?

Irrelevant. All that proves is that a: people are poor losers, or b: people are bad at identifying hacks, or c: there can exist skill gaps such that the more skilled player has an advantage similar in magnitude to that bestowed by a machine. All of those are already known, and none of them have anything to do with AI decisionmaking.

Quote

Given enough time to collect enough information, unbeatable AI can be created for every game.

Almost certainly true. However, that does not prove that said unbeatable AI is unbeatable because of its decision making skills.

Quote

After only one month of development students were able to create a Starcraft AI that used the exact same rulesets as human players; no cheating or shortcuts, but using the same data available to humans.

The AI may not be cheating in the sense of breaking the rules of the game, but it is most certainly taking full advantage of mechanical abilities that are utterly beyond the capacity of human players. Just take a look at some of those videos of the AI using mutalisks against templars. It is completely and utterly impossible for any human player to manipulate mutalisks like that - in particular, note the instantaneous scatter in every direction to avoid the initial storm volley. The computer is issuing commands to *individual mutalisks*, and doing so at such speed that they all appear to be commanded simultaneously. I don't care *how* good you are, no human can do that. It is simply not possible.

The AI is certainly making reasonable strategic decisions. It wouldn't be able to compete at all in a game like starcraft if it wasn't, no matter *how* insane its micro may be. However, it's not doing anything groundbreaking. Mutalisk harass may be a tactic that is especially well suited to the computer's advantages, but it's nothing new. I think I can fairly confidently state that its true advantages, as usual for AI in real time games, come mostly from its literally inhuman speed and multitasking ability.

I'm not arguing that unbeatable AIs don't exist. I am arguing against the apparent assumption in your original post that they are necessarily unbeatable because of their decision making philosophy. In chess, yes. In a turn-based strategy, yes. In starcraft, or MWO for that matter? No.

Edited by MuonNeutrino, 05 April 2013 - 05:06 PM.

[Rayah]

254

Well this is awkward... to be fair it's a Hunchback 4P though.

Ops! Just kidding, it's actually 375 in an Atlas AS7-D.

Edited by Rayzor, 05 April 2013 - 05:15 PM.

[Nonsense]

This AI stuff is irrelevant anyway. He's implying that if he uses AI-like decision making, he'll be a better player. Then he tells us that students made an AI that uses THE SAME DECISION MAKING AS HUMANS was able to beat humans. Why is this significant?

Current AI beat humans because they don't have to deal with UIs, because they have perfect memories and calculation abilities. They can't innovate...they have to be reprogrammed in the face of new things.

Pure calculation doesn't make you adaptable, and adaptability is what makes you good in a changing environment. Apparently this guy wants to argue that not adapting is best. OK DUDE, NICE TRY.

[Atlas32]

xhrit, on 05 April 2013 - 04:01 AM, said:

It only takes a maximum of 378 points of damage to kill all 8 mechs on the enemy team, solo, regardless of their weight class or armor load. Often you will be able to wipe an entire enemy team with even less damage, since most of the time you will not be facing a group of eight 100 ton assault mechs with maxed out armor values.

If you consistently score more then 400 points of damage per match you really need to work on your aim, so that you don't spread so much damage around the enemy mech. Target the single weakest spot on the enemy mech then focus all your attacks against it.

Look at it this way : put ten rounds from an AC/20 into ten different locations and you will not kill even the lightest stock trial mech.

On the other hand, ten rounds from an AC/20 into the same location will kill any custom mech no matter how big it is.

Completely wrong.....It is good to do more damage regardless if it's to JUST the CT. It helps your team by giving them less work to kill each enemy. Tearing off components is equally as important because it reduces the threat level of an otherwise hard to kill mech.

It's not about getting the kill, it's about helping the team overall. In theory, sure, you should be hitting in the CT every shot to kill, but most people take damage to other parts before dying (XL engines).

[Nonsense]

Atlas32, on 05 April 2013 - 06:10 PM, said:

Completely wrong.....It is good to do more damage regardless if it's to JUST the CT. It helps your team by giving them less work to kill each enemy. Tearing off components is equally as important because it reduces the threat level of an otherwise hard to kill mech.

It's not about getting the kill, it's about helping the team overall. In theory, sure, you should be hitting in the CT every shot to kill, but most people take damage to other parts before dying (XL engines).

3 people have tried explaining this to the guy...we'll see if he gets if he ever posts again.

[xhrit]

Nonsense, on 05 April 2013 - 05:54 PM, said:

Pure calculation doesn't make you adaptable, and adaptability is what makes you good in a changing environment. Apparently this guy wants to argue that not adapting is best. OK DUDE, NICE TRY.

Environments are not ever changing. There is a limited set of circumstances that govern every dynamic, and once those are measured there will always be an optimal solution.

"Randomness" does not mean "anything can happen".

The outcome of a random d6 roll is always going to be a whole number between 1 and 6.

The outcome will never be 7.

And once you factor in all variables such as construction, balance, control, and style... you can roll a 6 every time using controlled dice throws.

Even in TT I could reliably land shots on specific mech components.

Edited by xhrit, 05 April 2013 - 06:18 PM.

[xhrit]

Atlas32, on 05 April 2013 - 06:10 PM, said:

It is good to do more damage regardless if it's to JUST the CT. It helps your team by giving them less work to kill each enemy.

It is fitting that nonsense likes this post, since it is a totally nonsensical post.

If damage is work then more damage to kill enemy means more work to kill enemy.

less damage to kill enemy means less work to kill enemy.

end logic.

[xhrit]

Nonsense, on 05 April 2013 - 06:15 PM, said:

3 people have tried explaining this to the guy...we'll see if he gets if he ever posts again.

Go count how many people have tried to explain that having high damage numbers just means you are spamming fire instead of shooting to kill.

[Commander Kobold]

before missile nerf my catapult c1 had a rough 430dmg/match haven't touched it since I tried the "hotfix" though

Edited by Omni 13, 05 April 2013 - 06:38 PM.

[Levi Porphyrogenitus]

552.7 per match with my D-DC. I've got over a 3.3 KDR in that mech, at least since the detailed stat tracking was added.

Second place goes to my Catapult C4 with 476.2 per match and over 2.3 KDR.

I lack data for most of my mechs since I finished mastering them before the detailed stat tracking was implemented, but my Catapults have more games after that point, and my D-DC is the one I tend to play more when trying to make money or play with friends.

[Skoad]

On my Atlas-D I generally get 400-700 per match. Except for the times I get a completely fail team(which seems to be a large majority of the games as I play solo.. no friends =/).

Edited by Skoad, 05 April 2013 - 07:39 PM.

[Nonsense]

xhrit, on 05 April 2013 - 06:17 PM, said:

And once you factor in all variables such as construction, balance, control, and style... you can roll a 6 every time using controlled dice throws.

Even in TT I could reliably land shots on specific mech components.

10/10 troll.

You should go to vegas, not play battletech.

[Noobzorz]

200 some with the pathetic and overrated pile of **** that is the DRG-1C. It is actually my lowest damage per match of any mech I have (except for my commandos, which I am far, far more effective with from a W/L and K/D standpoint) by a large, large margin, and it isn't even the first DRG I bought, as I started on the 5N. Mastering it was a real chore.

I guess what I'm trying to say is the DRG-1C is not very good, and if you are playing well in a 1C, then I would suggest that you could do better still in another variant.

[Noobzorz]

xhrit, on 05 April 2013 - 06:25 PM, said:

It is fitting that nonsense likes this post, since it is a totally nonsensical post.

If damage is work then more damage to kill enemy means more work to kill enemy.

less damage to kill enemy means less work to kill enemy.

end logic.

Ceteris paribus more damage is better. Full stop. No discussion necessary.

If you want to suggest that more damage implies sloppy aim and crappy, misdirected offense, then you should say that very explicitly. Unless we have more information, the notion that more damage somehow implies inefficiency (which it may) is irrelevant.

If you are suggesting an endogeneity problem that necessitates that the ceteris paribus case virtually never occurs (and I'll grant you, that sounds quite plausible), then you should say that. Your claim as it stands is insufficient to rebut Atlas32's.

[Nonsense]

Aww man, you just got lawyered, statisticianed, economisted, or philosophy-studented. I don't know which, but...bamf.

Sometimes I wish I had studied these things so I'd be better able to categorize peoples' logical errors and explain them both to myself and the person in question, but that would sort of be ironic, wouldn't it? Get better at segmenting and categorizing something so I can more effectively criticize people for having an affinity (to a fault) for segmenting and categorizing everything?

If more people understood stochasticity intuitively, we wouldn't be having this bizarre conversation.

[xhrit]

Noobzorz, on 05 April 2013 - 07:51 PM, said:

If you want to suggest that more damage implies sloppy aim and crappy, misdirected offense, then you should say that very explicitly.

I have said almost that exact thing many times in this thread. Here is an excerpt from my very first post in this thread:

xhrit, on 05 April 2013 - 04:01 AM, said:

If you consistently score more then 400 points of damage per match you really need to work on your aim, so that you don't spread so much damage around the enemy mech.

Every other post I have made to this thread since that time has been an attempt to rephrase my initial statement.

Noobzorz, on 05 April 2013 - 07:51 PM, said:

If you are suggesting an endogeneity problem that necessitates that the ceteris paribus case virtually never occurs (and I'll grant you, that sounds quite plausible), then you should say that.

I do believe I said that too, except I refrained from expressing the sentiment in too verbose a manner, for it has become clear that half the people with whom I am arguing with can barley speak English.

Nonsense, on 05 April 2013 - 08:04 PM, said:

Aww man, you just got lawyered, statisticianed, economisted, or philosophy-studented. I don't know which, but...bamf.

Except he more or less agreed with my sentiment by saying the ideas behind my statements are not only entirely possible but also quite plausible.

Nonsense, on 05 April 2013 - 08:04 PM, said:

If more people understood stochasticity intuitively, we wouldn't be having this bizarre conversation.

I am a Determinist. I believe it is an objective fact that randomness does not in fact exist. Everything is governed by the law of cause-and-effect; given a enough data set, you can predict any effect. My tactics are driven by mathematics.

Edited by xhrit, 06 April 2013 - 04:34 AM.

[John MatriX82]

So far nearing 160 matches on my beloved STK3F with over 580 dmg a match:


The build relies on srms, I thought damage should have gone down a lot after the SRM nerf (of course I've employed this thing also before), but I've checked and it's steadily growing a bit, probably because I downgraded the ERLL that I used to give some range to the thing to normal LL to compensate the damage loss of the srms.