Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Sergarr]

It's extremely unlikely, though.

One doesn't need to have a perfect knowledge of a game to beat somebody 100 times out of 100, they just need to be really better.

EDIT: I like this part from here:

"Queen and pawn versus queen. Tablebases have shown that this can be won in many more positions than was thought, but the logic of the moves is presently beyond human understanding (Nunn 1995:265)."

So, yeah.

[notquitethere]

Beating AI in a game is often about exploiting specific aspects of their stupidity.

I'm not sure what you mean. You agreed that it was impossible for humans to beat computers and then gave a description on how humans go about beating computers.

I was moving on to a broader point about AI in computer games, rather than chess specifically.

But no, it hasn't gotten to the point where it's impossible for a human to beat a computer. Computers have become extremely good, but not perfect at chess. Making such a statement would require chess to be solved

Not impossible sure, but go read the wiki page I linked: it looks like the best chess computers are now better than chess grandmasters.

[alexandertnt]

It's extremely unlikely, though.

Not impossible sure

Impossible does not mean difficult. They are two very different words whose difference is crucial when discussing AI (unless you like being punched in the face by AI researchers )

One doesn't need to have a perfect knowledge of a game to beat somebody 100 times out of 100, they just need to be really better.

All your computer would have to do is make moves completely at random to have a chance at beating someone 100/100 times...

[Sergarr]

It's extremely unlikely, though.

Not impossible sure

Impossible does not mean difficult. They are two very different words whose difference is crucial when discussing AI (unless you like being punched in the face by AI researchers )

Last time I checked the world-class chess champion was unable to win even one time against a chess AI, only forcing 4 draws and losing 2 times. That means that it's practically impossible to win against a chess AI. Unless you're smarter than a world-class chess champion.

One doesn't need to have a perfect knowledge of a game to beat somebody 100 times out of 100, they just need to be really better.

All your computer would have to do is make moves completely at random to have a chance at beating someone 100/100 times...

logic.exe not found.

"really better" than someone means not making random moves. It should be obvious, yo.

[alexandertnt]

logic.exe not found.

"really better" than someone means not making random moves. It should be obvious, yo.

One doesn't need to have a perfect knowledge of a game to beat somebody 100 times out of 100, they just need to be really better.

My point: They need no such thing to accomplish that. This is to further highlight the point that extremes like "difficult" and "easy" are not the same as "guaranteed" and "impossible".

Last time I checked the world-class chess champion was unable to win even one time against a chess AI, only forcing 4 draws and losing 2 times. That means that it's practically impossible to win against a chess AI. Unless you're smarter than a world-class chess champion.

This shows the AI is extremely difficult. It does not in any way whatsoever prove that the AI is impossible to beat.

[Sergarr]

logic.exe not found.

"really better" than someone means not making random moves. It should be obvious, yo.

One doesn't need to have a perfect knowledge of a game to beat somebody 100 times out of 100, they just need to be really better.

My point: They need no such thing to accomplish that. This is to further highlight the point that extremes like "difficult" and "easy" are not the same as "guaranteed" and "impossible".

I was sure that if you're not significantly better than your opponent at playing the game, he's going to win a few times, but apparently not.

Last time I checked the world-class chess champion was unable to win even one time against a chess AI, only forcing 4 draws and losing 2 times. That means that it's practically impossible to win against a chess AI. Unless you're smarter than a world-class chess champion.

This shows the AI is extremely difficult. It does not in any way whatsoever prove that the AI is impossible to beat.

It's practically impossible to beat, for a human. Of course against a perfect-playing opponent current Chess AI would lose 100 times out of 100 - but we don't have those. We have humans and Chess AIs - and Chess AI are significantly better than humans, to the point where the best humans are unable to win (only able to force some draws). Given that beating somebody in Chess usually involves winning at least once, it's save to say that Chess AI can win against any human currently existing 100 times out of 100.

[i2amroy]

While we haven't actually "solved" chess, (as in, from the very start with all the pieces) what we have "solved" is all non-trivial endgames with no more than the two kings and five other pieces. A computer with access to that will be able to either force a draw or win regardless of what you do unless you can force the win before that point. Since computers are good enough at thinking ahead, it's not really possible to do that to the top chess algorithms.
http://en.wikipedia.org/wiki/Solved_game

IIRC there was a sort of exhibition contest where they had a weaker computer compete against a bunch of humans. The smartphone (which was nowhere near our top chess computer) ranked at about the grandmaster level, losing a small handful of games but never enough to stop it from winning the contest itself.

[UXLZ]

You're taking the definition of literal impossibility and applying it to a conversation about practical impossibility.

Strictly speaking, there is almost nothing in the entirety of existence that is literally impossible save things that are mutually exclusive, and even then they're impossible really only due to our human understanding of them.

That is to say, there are things that have a chance of one on ten to the googolplex to the googolplex (1 / (10)^(10^100)^(10^100) that strictly speaking aren't impossible, but the chance of it actually occurring is so low it may as well be zero. Positive Zero, so to speak.

[Shadowlord]

Beating AI in a game is often about exploiting specific aspects of their stupidity.

Ignoring the chess miscommunication...

For example, in Hazeron near the end of my time playing it I discovered a peculiarity in the AI: If I manned the helm on my warship and spun my ship, enemy ship helmsmen would fly huge circles around my ship (the radius of which were determined by how slow or fast I was spinning, IIRC). This permitted me to defeat them easily (by designing my ships with a greater weapons range, etc).

One possible explanation would be that they were attempting to continue to aim at one shield face, and when I rotated it away from them, they figured they had to fly around until they could see it again.
Another possible explanation would be that they were predicting my future position based on my velocity (and 0 acceleration), but not taking into account the fact that I never went anywhere because of the way velocity vectors rotate with ships in hazeron (so that I was travelling in a small circle).

The shield facing one sounds better (less like programmer technobabble). The position prediction one is probably correct, though.

Either way, they flew in circles given the right stimulus so it was possible (and likely still is) to defeat entire fleets of AI ships (or player-commanded ships with AI helmsmen) using a single player-helmed ship.

[alexandertnt]

I was sure that if you're not significantly better than your opponent at playing the game, he's going to win a few times, but apparently not.

How does that even logically follow from my argument?

Given that beating somebody in Chess usually involves winning at least once, it's save to say that Chess AI can win against any human currently existing 100 times out of 100.

What? No. Demonstrating that an AI won any finite number of times does not make it in any way safe to say that chess AI will win 100% of the time. Statistics do not work like that.

You're taking the definition of literal impossibility and applying it to a conversation about practical impossibility.

I am using the definition of the word used in Game Theory. This strikes me as being quite appropriate to use in a discussion about game systems and AI.

Strictly speaking, there is almost nothing in the entirety of existence that is literally impossible save things that are mutually exclusive, and even then they're impossible really only due to our human understanding of them.

This is just not true. For example, against a perfect tic-tac-toe AI, it is impossible for you to win. Tic Tac Toe is a Solved Game.

That is to say, there are things that have a chance of one on ten to the googolplex to the googolplex (1 / (10)^(10^100)^(10^100) that strictly speaking aren't impossible, but the chance of it actually occurring is so low it may as well be zero. Positive Zero, so to speak.

Prove to me that the "possibility of a human beating an AI is incredibly low". Without a complete solution, there may always be unknowns you just don't know about yet e.g. there may be some unutilised anti-computer tactic that has yet to be discovered, or some flaw in the heuristics used in deciding a move we have yet to find.

These potential strategies that may or may not exist is the reason why saying it is impossible to say it is impossible to beat Chess AI.

[Shadowlord]

The possibility of a human beating an AI is very high if the human uses an EMP.

[Sergarr]

alexandertnt, by your standards, all empirical evidence we have is null and void. We can't say that, for instance, an apple thrown by a human standing on Earth will fall down, because it has only been proven a finite amount of time, and we'll never be able to experimentally test all possible positions, apples and throw methods. That's one truly absurd proposition.

The lack of any human victories against any moderately powerful chess AI in the recent decade proves rather handily that the modern good (not intentionally gimped) chess AI cannot be beaten by a human (as in, he cannot force a victory).

Any anti-AI strategy would have to be so complicated that no human would be able to make even one turn a day without help from a computer - and that would rather defeat the whole point. If it was simple enough to be found by a human, it would've been already found - because people tried hard in doing that. Even with utilizing anti-CPU methods, no victories were made, only draws.

Given that the Chess AI is also evolving - and doing it faster than humans, there's practically zero chances that the situation will reverse in the future, unless humans learn to download more brains.

[alexandertnt]

alexandertnt, by your standards, all empirical evidence we have is null and void.

What? Empirical evidence is incredibly useful to test hypotheses. For example, the evidence that Chess AI wins games against the strongest human components is very useful evidence in support of a hypothesis like "Chess AI is better at playing chess than humans". That's a fundamentally different statement then "Chess AI cannot be beaten by humans".

The lack of any human victories against any moderately powerful chess AI in the recent decade proves rather handily that the modern good (not intentionally gimped) chess AI cannot be beaten by a human (as in, he cannot force a victory).

Does the fact that "all deaths in the last decade have not been mine" prove that "I cannot die"?

[NobodyPro]

But seriously they look like they're basically a new specie in everything but the name. I wonder why they chosen to name them Kobolds.

IIRC the Kobolds of myth are invisible household spirits, nothing to do with the reptilian DnD race.

[alexandertnt]

Stupid generic fantasy races
I guess this complaint includes more than just video games, but I absolutely hate it when a fantasy game limit itself to the few standard fantasy races, and even more when they are given their even more generic personalities. What I really hate about it is not their concept but rather the fact that they were so massively overused. Even worse to me when a game try to pretend to be imaginative by adding a single new race which is presented as "different" or "unique" while being basically a human with one or two main differences and having a culture that is blatantly a rip-off from a real life civilization (Minor variation of another fantasy races also counts). I'd love to actually see some creativity from the genre that is supposed to be all about creativity.

Seeing standard races being interpreted differently is a step in the right direction, but I don't think it's enough.

Also posting again to state how much I agree with this.

I am peeved by the fact that different races are almost always just humans but stout (dwarf), or tall (elf) etc. And that these are now "the" races in fantasy. How boring!

Other cultures are almost always just some generic ancient culture. So you end up with tall blonde Aztec's, for example.

Also the fact that people from those other races almost uniformly fit a single culture. So every single stout person will be honourable, hardworking and devoted to a single God, for example, and exceptions will be rare if there are even any.