Sort-of Yes.  Better AI is a good thing, of course.  But development resources are limited, and I expect a complete and fun game above all else.  I can certainly understand how somebody can be disappointed by AI that's not as good as possible, but "as good as possible" is very vague, and puts priority on one aspect of game development at the cost of others.  I would think - you would know better than I, so feel free to tell me I'm completely wrong - but I would think that there comes a point in developing an AI, where diminishing returns become an issue and improvements to the AI become less and less pronounced. So, the logical direction for the developers is to use their available resources to work on other areas of the game, rather than make better AI.

Whatever the reason, the decision as to how much to develop any one area of a game is determined by a developer allocating their resources - time, manpower, money - to the various components of a game, in order to develop the best game possible.  I don't see it as an issue of a developer not trying as hard as they should, but rather prioritizing.  I don't always like developers' priorities, as you apparently don't like Stardock's priorities here.  But it is their game, and their decision. 

I'm sure they want a very good AI, but they must recognize their limitations.  I expect a challenge in any game I play, but I don't necessarily expect that challenge to always come from superior AI.  Challenge that comes from some other obstacle - such as bonuses to computer players - is still a challenge, and one that I have to figure out how to overcome.  In an perfect game, this wouldn't be necessary, but all we can really do is try the best we can to do the most with what we have.

Sorry, but that is horribly...  horribly... horribly innacurate.  I'm not sure how anybody could possibly believe there are more board positions in chess than atoms in the universe.  I'm sure there are more atoms in a single chess piece than there are positions in chess.  I'm really not trying to troll or anything, but come on...  wow

Current estimates place it at somewhere around the number of atoms in 6x10^11 to 6x10^17 metric tons of water, depending on who you ask. Rather a bit more than a chess piece, but a far cry from the whole universe. It's the possible combinations of moves to reach said positions that is larger.

However, a tremendous number of the positions could only possibly arise through cooperation by both players with explicit the goal of creating said position, and thus aren't relevant in the context of an AI chess opponent.

 I think it would be wise to reword the difficulty levels more transparently. Traditionally in gaming. when you click hard you assume your playing against the AIs best, then you have settings like ridiculously hard and so forth. Also from a captain obvious point of view, normal would be literally normal stock AI.

As far as deep blue beating Garry Kasparov it is a very questionable if the machine acted alone being they kept it in a separate room out of site. There is allot of fishy stuff about that match and IBM had allot to loose in share holding if they lost. 

Blue_Oyster I understand where you are coming from, I have played a many games where the AI is just to easy unless you give them handy caps, I long for the day where the AI is a challenge with out giving them handy caps or boosts, I'd love a AI that takes advantage of every bit of the game but with a game like this it is quite hard sense most things are randomly generated, other then the home world.

You said you are a Artificial Intelligent programmer, so you know more about it then me and might know how to do it but as far as my knowledge goes I do not know or understand it to a point, To my understanding it is just a list of scripts that if a certain criteria is meet it does one thing or another.

Surprisingly, there are a vast number of possible configurations of chess positions.  The way to get this into your head is: 

Break out a chess board and put a pawn in the lower left corner.  Wait one second then move the pawn one space to the right.  Repeat this process until the pawn touches all 64 squares of the board. 

Now, add another pawn in the lower left corner and one pawn in the square directly to the right.  Move the second pawn one square to the right and wait one second.  Repeat that process until the second pawn touches all the remaining squares THEN move the first pawn one space to the right and the second pawn into the space vacated by the first pawn and repeat the whole process for the second pawn until it has touched all remaining 63 squares.  Do this until the first pawn has been moved into all 64 squares... this is going to take a while; exactly 64^2 seconds or 4096 seconds. 

Of course no one is really going to do this but, you start to see the enormous number of possible configurations that exist.  Breaking all the actual movement rules and just configuring the pieces in all possible positions is 64^32(assuming a unique identification of each piece) ...  a vast number. 

Oh yes, I understand the math.  64 squares, 16 possible pieces per side, 9 different movement patterns available for those pieces per side, pieces will be eliminated, so on and so forth.  Start doing all the multiplying and the numbers start getting pretty big pretty quick.  But there is mostly certainly a very FINITE number of possible moves and positions on a little 64 square chess board with 32 pieces.  Whatever that final number is, it is definately calculable, and a number most people would be able to at least wrap their mind around somewhat.  Now try counting how many hydrogen atoms there are just in our relatively small sun.  Think trillions and your are thinking too small, but we will just go with trillions.  Now take into account there are millions of our suns just in our galaxy.  Now take into account there are millions upon millions of galaxies just within our "sight range" of the known universe.  You don't have to go very far before you completely blow away any possible number of positions and moves available for a chess game.

It just baffles me sometimes how small the human race thinks, when there is so much out there in the universe

...but such an exercise isn't very meaningful in the context of a chess AI. Using illegal board positions would be only detrimental to its performance. That's why folks are only interested in the legal positions, which are considerably more challenging to accurately estimate.

If you feel you can craft an AI that is invulnerable to its patterns being learned by players and exploits to get around its patterns, good luck, I'd like to see it.  Its unfortunate but the reality is that, beyond a certain point, the AI cannot truly learn and adapt to a human playstyle, and thus giving the AI booster cheats will make a game more challenging than not having them and extend the challenge of a game.

My answer is, no I am not disappointed. I have always known it would be this way

You can't blame a weaker player (AI) for bringing a gun to a knife fight. It is cheating, however it presents a difficult, but winnable challenge for the highly confident and skilled knife fighter. If any particular knife fighter is unwilling to accept the challenge and abhors the cheating, the answer is simple, avoid confrontations with the gunfighter.

Throughout the viable life of GCII, Brad attempted to make the "normal" AI better. Anyone who didn't care for the AI's higher level bonuses could keep playing on normal with the expectation that someone was working on making it smarter.

the bar is low because all the strategy game developers have set it low.

until customers expect competent AI, they won't ever get it because developers are lazy (translation: seeking profit instead of quality)

chess has good AI because chess is ridiculously easy and simple, but also because tons of smart people have put a lot of effort into it

strategy game AI is usually some half-assed part-time job done by someone long before the game and its rules are complete. usually, the programmer doesn't even know how to play the game optimally, so there's no chance for the programmed AI to do so

See my previous post in this topic to understand why "laziness" or greed is not the prevailing factor here.

There is no longer any question that Deep Blue beat Kasparov.  Looking back with hindsight, and taking into consideration further advances in the field in 18 years since 1997, even he admits it now.  My favorite quote from the 2nd match came after Kasparov's loss in game 2, on the morning following his claim of cheating.  He met with his team of seconds, who had been up all night analyzing, and one of them said to him, "It's OK, Garry, we found some mistakes [that you made]".  Ironically, this calmed Kasparov down, because he grasped that he lost because of his mistakes, not because the machine was superior.  (In fact, Deep Blue blundered in that game 2 due to a bug in evaluating king safety, and threw away its win when it left itself open to a perpetual check, but Kasparov failed to find it.)

Since then, chess programs/machines have moved way past that, and we've learned that optimized "brute-force" search (minimax with alpha-beta pruning), running on specialized chess co-processors with gobs of memory, scales in Elo strength roughly linearly with hardware, for playing strengths from 2500 to ~3100.  WC Magnus Carlsen is currently at mid-2800s, and only about three human GMs have ever cracked 2800.  So chess machines simply out-machine humans, the way internal combustion engines simply outdrive us.  It hasn't destroyed human chess as a fascinating sport.  Chess is at least as noble as golf -- and there will always be both

• In 2002-2006, the Hydra consortium http://en.wikipedia.org/wiki/Hydra_%28chess%29 made a sequel to Deep Blue, using a 32-node cluster with FPGA chess co-processors and 64 GB total memory (which is only 2 GB per node, so don't get all excited), over a fiber optic link out of Abu Dhabi.  The 32-node version searched ~150 million moves/second, similar to Deep Blue, but used better heuristics to prune branches, enabling average search depth of 18-ply vs. Deep Blue's 12-ply.  It beat up on some strong human GMs in the 2nd tier of top GMs: Adams (5.5/6), Kazimdzhanov (1.5/2), Ponomariov, and Khalifman (0.5/1).  Like all chess machine projects, it was cancelled due to lack of funding, interest, or both, and no longer exists.
• Stock chess programs (which come in free, paid, and probably GM-subscription tiers) have exhibited a similar gradual improvement in their hardware-to-playing strength curves over the last few decades.  You can just grab the free versions of Fritz, Stockfish, or Fruit with one search and a few minutes, and it will beat you on your laptop.  Hardcore dudes run 8-core PCs with 32 GB RAM (remember that some of us here in GC3 are at 16 GB already :)), and at that level even those standard programs approach GM strength -- so much so that a GM's homework these days is to hole up in a room with such a PC and his subscriptions to 3-5 engines, and thoroughly analyze some line he's chosen.
• Chess programs, including machines such as Hydra, still do suffer from being ... idiot savants.  They can have bugs, especially in their static eval functions, and be simply wrong.  (Hence a GM never trusts just one of them, and always amasses an ensemble of a few or several, and lets them all vote.  CEOs amass boards of directors, and presidents amass cabinets, for much the same reason.)

The other problem of sustaining a monster chess machine is that ... you realize you're a genius in a room full of geniuses, and each of you could earn millions doing something else.  So every such team has soon broken up

I say Gilmoy has brought this surreal thread to a suitable conclusion.

I actually wrote a 3-hour response with my reply to the OP (plus some ... sidetracks).

When I clicked "Post Reply", it took me to a login page.  Logging in did not post it.

Going back in my browser cache took me back to the previous reply I edited (the one above).

So this damn Forum software ate my post.  I'll know to never trust it again.

One of the things Brad has talked about is that the data-mining enabled by steam will allow AI to learn from the best human players. He mentioned that previously he could only program the AI to play as well as he could play, but now he can make it play as well as the best players. The idea of chess AI is not completely inapplicable either as one of the main strengths of chess AI is that it has a huge bank of previous games in its memory, and so can look at a situation and find the same or an analogous situation and understand how certain moves might play out for each player. This is the thing that might now be possible in a 4x. If the AI can be programmed to collect information and then draw good analogies between it's current game and previous games played by expert players it could play at or near the highest level a human can play without cheating. I think we are a ways off, but it is not impossible to create an AI that not only is challenging, but adapts automatically as players find exploits and the game expands through expansions/dlc.

This seems like a non issue brought up by an "artificial intelligence programmer" with no burden of proving his legitimacy in the first place.

As the stormies would say.. Move along, move along.

I wouldn't ask somebody to reveal their identity online.

But even though I disagree that it's a big problem for this game, I think there is legitimate discussion about AI, to what extent they should develop their AI, and the benefits that brings to the game vs. the cost of that development (not just monetary) 

Plus, the conversations in this thread are fascinating.

Guys / Gals,

                Look the AI is stupid compared to really good human players can do in 2015.  Get over it.  At least this means that a future Skynet has a long way to go before mass producing human killing terminators.  Everytime my printer screws up I think to myself that AI has a long way to go to having a fully functioning T800 not screw up somehow. 

The thread is fascinating and I have enjoyed reading it. However...

It is mildly irritating to see persistent, specious, arguments from someone claiming to have expertise in a specialized field without any demonstration of that expertise beyond self-damning, and outrageous declarations that are patently incorrect.

It's no big deal really. This is the inter-web and it takes all kinds. Carry on.

actually your printer is self aware and is purposefully trying to sabotage you. think about it how come it only causes problems when you're in a rush?

OMG  .. that is So true ..

Chess programs don't actually do that.  What you described is case-based reasoning, and it is an AI technique suitable for some ill-defined domains, but not chess.  (Real chess programs do come with large opening books, which are simply canned lines fed into it, representing the best theoretical knowledge acquired by humans and/or computers over centuries, but they don't reason over their books.)

Chess falls under the umbrella of perfect-information two-player turn-based games, which includes checkers, xiangchi, shogi, jetan, and klin zha kinta, but not Stratego (hidden knowledge) or Monopoly (random die rolls).  For this category of games, the uber-method is adversarial search, more specifically minimax with alpha-beta pruning (and many chess-specific heuristics, e.g. board-encoding representations to detect repetitions, which become loops in your search tree).  It's ultimately just a brute-force expansion of an exponential search space.  At the cut-off depth, you apply a static eval function to the board to boil it down to one score.  Then you bubble these up the tree, and play I-cut-you-choose at each level, flipping back and forth between the two players' perspectives, so that each of them, at every level, can only choose the least-worst of their opponents' best subtrees.  It's pure computation, all math and data structures, absolutely nothing smarter than that.  To use your clock time most efficiently, you search the full tree to depth 1, then to depth 2, then ..., then to depth i, which is called iterative deepening.  It's trivial to prove that the total time for the first N-1 depths combined is a tiny fraction of the cost to search at depth N, for all N (remember: exponential trees are weird that way, like the puzzle with the lily pads), so this simple scheme is optimal for any finite time limit.

Any "chess knowledge" from human experts must be encoded into the static eval function.  This is where they can be wrong and have bugs appear -- not a bug in the minimax algorithm, but a bug in "how it plays this chess-like game".

• A common flaw in strong commercial chess programs is that they will not accept a draw when they think they're ahead in material.  They would rather make a suicidal pawn move, allowing a catastrophic breakthrough, than just settle for a draw.  Exploit: Start hundreds of games, resign instantly if the pawn structure is not "lockable".  When you finally lure it into a "lockable" pawn structure, arrange the pawns so that all 16 pawns survive, blockade each other, and have no captures.  Trade off most of the pieces, then just stall behind your pawn wall.  American GM Hikaru Nakamura went through a phase in 2007-2008 where he goofed off doing this, playing hundreds of games on ICC (a popular chess server that attracts high GMs).
• The 50-move draw rule means: after 50 moves without a pawn move or capture, draw.  Hence, Nakamura times it perfectly to dangle an exchange (his Rook = 5 pawns for the program's Bishop or Knight = 3 pawns) at the 50-move limit.  The program always accepts, and thereafter is sure that it's "winning".  On the 50th move after this capture, the program will push a pawn rather than stall.  Nakamura wins the ensuing suicidal pawn-racing, breaks through with his King, eats all of the computer's pawns from behind ... and then has fun underpromoting to ridiculous piece mixes.  He has:
• Crafty (computer) vs. Nakamura, 2007, with six promoted knights and nothing else: http://www.chessgames.com/perl/chessgame?gid=1480850
• Rybka (computer) vs. Nakamura, 2008, with five promoted bishops and nothing else: http://www.chessgames.com/perl/chessgame?gid=1497429
• He got blacklisted by each programs' author for abusing ICC courtesy rules when playing vs. these programs.  He must have lost 500 games to get these 2 trophy wins.  Think of him as a beta tester of chess programs, and it's only a little less weird

Computer chess power does not translate readily to any other domain, because it's based entirely on minimax search, and most problems just can't be shoehorned into a simple tree of positions with a simple algorithm crawling over it.

I feel stupid now.  Instead of saying "its not rocket science" they should say "its not computer AI programming".