Our dear friend the AI

Matrix Games Forums

Forums Register Login Photo Gallery Member List Search Calendars FAQ

My Profile Inbox Address Book My Subscription My Forums Log Out

Our dear friend the AI

View related threads: (in this forum | in all forums)

Logged in as: Guest

Users viewing this topic: none

Printable Version

All Forums >> [Current Games From Matrix.] >> [World War II] >> Uncommon Valor - Campaign for the South Pacific >> Our dear friend the AI

Page: [1] 2 next > >>

Message

<< Older Topic Newer Topic >>

Our dear friend the AI - 7/26/2002 10:42:34 AM

pasternakski

Posts: 6565
Joined: 6/29/2002
Status: offline

After a couple of months of serious discussion of this game, it occurs to me that a healthy thing might be an open forum on how the computer enemy conducts itself. I don't intend this to be an excoriation of the AI, but rather, a positive, thoughtful examination of our experience with the opponent the fine designer folk have crafted to make our admiralship as miserable as possible. I hope that suggestions will emerge to make our task as HIs (human intelligences) more difficult in ways that will be satisfying to us as gamers. I hate it when games try to defeat you through cheating and unfair advantages to AS (artificial stupidities). I believe that one of the most satisfying game experiences is defeating an AI that is truly wily, competent, and prepared (and I know that these are as rare as hens' teeth, but UV is exploring all kinds of new territory in game design, why not improved AI design, as well?).

I posit the following general observations, within the UV frame of reference, as dialogue starters:

- The AI is so conservative as to make Ronald Reagan look like a flaming liberal (sorry, Mr. President, I know you're ill)
-The AI is not creative enough to mount a successful attack against a well-defended objective
-The AI cannot organize a comprehensive, strategic defense
-The AI does not know how to get the best results from specific unit types
-The AI does not know how to make the most of subtle tactical advantages in order to turn them into winning strategies

Well, I do go on, and I apologize. These are general comments, but I hope that all of you have more specific matters to address that may lead to concrete suggestions. My sense is that we may be able to pitch in and help our favorite computer wargame designers (bless 'em, they're the greatest and the onliest) refine their approaches and make our gaming experiences more satisfying than they already are (and without Matrix and 2 by 3 how satisfied would we be?). We have all spent plenty of time playing and replaying scenarios against the computer, refining our strategies and tactics and reclaiming our combat losses. How great it would be to get thoroughly creamed because the enemy was just plain good, not lucky or artificially enhanced ...

What do you think, comrade grognards?

--------------------------

I will now proceed to entangle the entire area

Post #: 1

Ai - 7/26/2002 12:52:14 PM

Rob Roberson

Posts: 387
Joined: 5/1/2002
Status: offline

I think one of the biggest faults I see in the AI is its inabilty to recognize that when there is enemy air/naval units in the area it must send warships in with its transports. In my AAR it is a turkey shoot around Guadalcanal (which I continue to let the AI keep). No long range CAP, no friendly warships, just transport after transport coming down the slot to be bombed, strafed and sank. Now the AI has managed to keep Lunga somewhat supplied, but at what costs.

The other thing, it's February 43 in my AAR...I don't remember when the last time I saw a Japanese submarine attack anything. Not sure what the deal is there.

(in reply to pasternakski)

Post #: 2

- 7/26/2002 7:00:02 PM

Sonny

Posts: 2008
Joined: 4/3/2002
Status: offline

Well, off to a good start. In the first two posts we have one guy saying the AI is too conservative and another guy saying the AI throws transports around with abandon.:D

(in reply to pasternakski)

Post #: 3

- 7/26/2002 8:21:32 PM

Q-Ball

Posts: 7336
Joined: 6/25/2002
From: Chicago, Illinois
Status: offline

I would agree with the Transport wasting, maybe the routines could be re-written to so that they are always escorted by something, either warships or some LR CAP. I would also add:
*Will not send heavily damaged ships to nearest port, ensuring their sinking after a long voyage
*Will waste valuable CV air assets in attacks against LBA.

I will comment, however, on a couple things the AI does fairly well:

*Committs CV's in large groups rather than piecemeal. Even with a bad opponent, a 7 CV IJN TF is pretty tough customer
*Will pile alot of troops into key bases.
*Sometimes will have strong campaigns to shut a base down. After daily 200+ bomber strikes out of Rabaul (#19), PM is rubble.

But overall you have to give the AI alot of advantages to give them a chance. I am playing my first game as IJN, and gave them every advantage; very hard AI, 200% ship commitment, etc., and so far it's still not too close, with at least 30 allied AP's on the ocean floor.

(in reply to pasternakski)

Post #: 4

- 7/26/2002 8:30:00 PM

IChristie

Posts: 673
Joined: 3/26/2002
From: Ottawa, Canada
Status: offline

[QUOTE]*Will waste valuable CV air assets in attacks against LBA. [/QUOTE]

This particular tactic (which I have noticed as well) will probably be much more effective with the adjustments to level bombers in v1.2.

It seems to me that this is approach is based on the historical Japanese strategy of attacking land bases with powerful CV task groups. It probably should be more effective than it has been. With the effectiveness of allied medium and heavy bombers having been decreased I am no longer looking forward to taking on Japanese Super-CV TF's from land bases.

_____________________________

Iain Christie
-----------------
"If patience is a virtue then persistence is it's part.
It's better to light a candle than stand and curse the dark"

- James Keelaghan

(in reply to pasternakski)

Post #: 5

- 7/26/2002 10:26:24 PM

Sabre21

Posts: 8231
Joined: 4/27/2001
From: on a mountain in Idaho
Status: offline

I have noticed that the AI tends to put more land forces in than the supply system can handle. Maybe as the Japs, I can almost believe this..I know many an Imperial soldier starved on Guadacanal. But to the degree that say Shortland gets reinforced it seems is well beyond what would make sense. That area becomes a turkey shoot once Lunga is secure. I haven't run into much naval opposition in that area.

I also haven't seen any attempt by the Japanese on New Guinea to move forces cross crountry. Maybe I'm not just seeing it.

I agree I don't want to see AI cheats either but it will most definitly take some ambitious coding to defeat a human player.

Andy

(in reply to pasternakski)

Post #: 6

- 7/27/2002 12:39:45 AM

pasternakski

Posts: 6565
Joined: 6/29/2002
Status: offline

[QUOTE]Originally posted by Sonny
[B]Well, off to a good start. In the first two posts we have one guy saying the AI is too conservative and another guy saying the AI throws transports around with abandon.:D [/B][/QUOTE]

Actually, I think that the two posts are compatible. The conservatism (and I don't necessarily see it as a bad thing) is evident in the AI's overall strategy, while lack of transport protection is characteristic of its tactical execution.

I tend to play conservatively, trying for one thing to steer away from the "I fought 35 battles in the first two days and lost 18 months' worth of ships, and now I can't move my troops around" trap.

(in reply to pasternakski)

Post #: 7

- 7/27/2002 1:11:32 AM

Sonny

Posts: 2008
Joined: 4/3/2002
Status: offline

[QUOTE]Originally posted by pasternakski
[B]

.............

I tend to play conservatively, trying for one thing to steer away from the "I fought 35 battles in the first two days and lost 18 months' worth of ships, and now I can't move my troops around" trap. [/B][/QUOTE]

But you are not the AI - it does play like that if you are there to give it a battle. Not saying it is wrong, just that it will slug it out with you without worrying about conserving anything for the future. Of course my opinion may be slanted since I had never even looked at the Japanese side until I started a (PBEM) game as them last night. Don't know if the AI acts differently if it is one side or the other.:)

Aside from the fuster cluck when you capture Rabaul and the Japanese still base their ships there, one of my gripes is that the AI parks ships in ports within easy LBA range and leaves them there until you sink them. Don't know what the strategy is there unless they are waiting to ambush a bombardment or invasion group.:(

(in reply to pasternakski)

Post #: 8

- 7/30/2002 8:51:38 AM

pasternakski

Posts: 6565
Joined: 6/29/2002
Status: offline

Well, please note that the issue statements in my first message do not necessarily reflect my client's - er - my - opinion (sorry, us lawyers tend to get caught up in those kinds of BS statements). The conservatism observation was an embellishment taken from one of the designer's statements in another thread.

What I think would be helpful is small-scale suggestions on AI performance in execution of its overall campaign plan. For example, if you have turned off the "fog of war" option (as I have just to see what the heck is going on), you must have noted the "penny packeting" syndrome that characterizes the AI's use of surface ships. Good grief, when the AI is the Japanese, it looks like a whole herd of red ants crawling around all over the place. Then, whammo! You catch an un- or under-escorted troop transport convoy and whack the crap out of it.

I guess I just don't understand some of the tactical actions taken in furtherance of operational goals.

---------------------------

I will now proceed to entangle the entire area

(in reply to pasternakski)

Post #: 9

Unescorted Cruiser TFs - 7/30/2002 9:49:15 AM

denisonh

Posts: 2194
Joined: 12/21/2001
From: Upstate SC
Status: offline

One thing I have noticed in a SC17 game as the USN against the AI are surface TFs with only cruisers engaging in surace action.

With the high quality of IJN destroyers with those deadly torpedoes, it is simply stupid. 2 IJN CAs aginst 3 USN CAs, a CL and 6 DDs is a mismatch that 6 IJN DDs would equalize.

Instead, the outcome are the IJN cruisers getting hammered, and either getting picked off by subs or a/c the next day.

_____________________________

"Life is tough, it's even tougher when you're stupid" -SGT John M. Stryker, USMC

(in reply to pasternakski)

Post #: 10

- 7/30/2002 10:16:28 AM

tohoku

Posts: 415
Joined: 3/18/2002
From: at lunch, thanks.
Status: offline

I'd just be interested in knowing how the AI was coded. Whilst I haven't go my hands on a copy of the game and, ahhh, 'had a look', I can't tell from the AAR whether the thing is hard coded or not.

From what everyone has posted it seems to simply be hard coded with a randomiser for picking choices off a set list of options to pursue and few modifying hueristics thrown in to allow for tactical situations. Yes? No?

tohoku
YMMV

(in reply to pasternakski)

Post #: 11

AI - 7/30/2002 11:48:17 AM

Ron Saueracker

Posts: 12121
Joined: 1/28/2002
From: Ottawa, Canada OR Zakynthos Island, Greece
Status: offline

I think the AI development is, as we've discussed in a previous thread months prior to UV's release, a challenging task at this level of complexity. It's because of this that many, and I'm included, believe that the AI is good enough to play against while familiarizing yourself with the game, but that's about it. But, truth be told, give me an example of a good AI in a game more complicated than a sports game. You can't.

Take EA sports NHL series. There are a limited amount of situations to respond to, and a limited number of ways to counter. The game is near flawless with respect to AI. In real life, this limited number of possibilities explains why a good hockey player is somewhat overated when referred to as a "smart" hockey player. AI in Nascar sims is even easier....go fast, turn left.

I know little about how AI is approached by game designers of war games. Perhaps it needs a whole new approach by a company who concentrates exclusively on developing an AI "engine", with the hope that other companies who don't have the time and resources to develop their own can use it under license. Is this not the case with interop play engines like RTime?

Hueristics have been talked about but I know little about this. Something to do with mimicing a reasonable human decision making process through the use of prioritised weights and measures I guess. I'll have to do a little reading before I go much further.

Is it not possible to have the AI mathematically (satisfy scripted parameters) weigh the value of various areas/bases to facilitate action? Is it not possible for the AI to calculate (satisfy scripted parameters) whether or not it is suicide to enter an enemy Air Zone of Control, and utililize a threshold to determine the necessity of doing so? What about mathematically being able to determine (satisfy scripted parameters) whether or not it (AI) has strength necessary for offense, or only defense.

All these are questions I cannot answer, as I'm neither a mathematician or a programmer. Any comments? Am I totally of base?

_____________________________

Yammas from The Apo-Tiki Lounge. Future site of WITP AE benders! And then the s--t hit the fan

(in reply to pasternakski)

Post #: 12

- 7/30/2002 12:06:37 PM

pasternakski

Posts: 6565
Joined: 6/29/2002
Status: offline

[QUOTE]Originally posted by tohoku
[B]I'd just be interested in knowing how the AI was coded. Whilst I haven't go my hands on a copy of the game and, ahhh, 'had a look', I can't tell from the AAR whether the thing is hard coded or not.

From what everyone has posted it seems to simply be hard coded with a randomiser for picking choices off a set list of options to pursue and few modifying hueristics thrown in to allow for tactical situations. Yes? No?

tohoku
YMMV [/B][/QUOTE]

You know, this is a most interesting observation. The "objective choices" model is likely the best description of the truth of the design (if A, then B. If A and C, then D ...). I think that this is the state of the art for AI and has been for quite some time in simulation gaming. Can we help game design move ahead within the limitations of current technology? I believe that any enjoyer of computer simulations would like to see advances designed to model the flexibility and, therefore, competence of human gaming intelligence (oh, please, darling, beat me. Humiliate me. But, please, do it in a way that pleases me).

This brings me to one of the crucial points I had hoped to raise by starting this thread. Is there any way to "fiddle" with those choices and modifications to make the AI behave more "humanly?" I doubt that the underlying structure can be altered without throwing out the baby with the bathwater (and, in my estimation, the AI is pretty good. I am just hoping that there are ways, in which we consumers can help, to raise the bar another inch or two). In other words, is the AI so hide-bound in its choices that those choices cannot be manipulated to model more "human" (and, ipso facto, more challenging) behavior?

I can tell you this. No AI opponent I have ever seen acts in a way that would make you believe it was a human opponent. I sure wish that it could ....

-------------------------------

I will now proceed to entangle the entire area

(in reply to pasternakski)

Post #: 13

- 7/30/2002 7:52:16 PM

Sonny

Posts: 2008
Joined: 4/3/2002
Status: offline

Eventually AIs will improve as more computing power is packed into pcs.

But consider the power necessary to defeat a world champion chess player. And (as we are probably all tired of hearing) there are a somewhat limited amount of moves that can be made in a chess game (not a small number, but when you consider how many pieces and places in a chess game and then compare it to the number of pieces and places in UV...). Of course we each consider ourselves a world champion UV player so imagine how long it would take to develop a UV master AI.

Should the AI be better - a most definite yes. This is a turn based game and can afford to take time to crunch a few numbers after you hit the END ORDERS button - even with low end computers. Making it react like a human would react is not gonna happen anytime soon. Besides, which human would it react like? You? Me? Joel? Gary? :)

(in reply to pasternakski)

Post #: 14

- 8/1/2002 7:40:57 AM

pasternakski

Posts: 6565
Joined: 6/29/2002
Status: offline

[QUOTE]Originally posted by Sonny
[B]Eventually AIs will improve as more computing power is packed into pcs.

But consider the power necessary to defeat a world champion chess player. And (as we are probably all tired of hearing) there are a somewhat limited amount of moves that can be made in a chess game (not a small number, but when you consider how many pieces and places in a chess game and then compare it to the number of pieces and places in UV...). Of course we each consider ourselves a world champion UV player so imagine how long it would take to develop a UV master AI.

Should the AI be better - a most definite yes. This is a turn based game and can afford to take time to crunch a few numbers after you hit the END ORDERS button - even with low end computers. Making it react like a human would react is not gonna happen anytime soon. Besides, which human would it react like? You? Me? Joel? Gary? :) [/B][/QUOTE]

I agree with your point. I guess by "more human," I was going for "more flexibly" or something like that. I do believe that it should not be all that difficult to program some creative unpredictability into the AI without making it just weird or silly.

------------------

I will now proceed to entangle the entire area

(in reply to pasternakski)

Post #: 15

- 8/1/2002 8:59:33 AM

FirstPappy

Posts: 744
Joined: 9/12/2000
From: NY, USA
Status: offline

I always thought the AI should be somewhat user programmable. Let the player decide what strategic moves the AI should make. In UV's case they would be something along the lines of "Defend PM" "Capture Gili-Gili". There should be weights set to each. Add a randomness factor as well. You should also be able to change these settings during the course of play to avoid AI mistakes of defending lost causes.

Just my $.02

(in reply to pasternakski)

Post #: 16

- 8/1/2002 10:09:30 AM

pasternakski

Posts: 6565
Joined: 6/29/2002
Status: offline

[QUOTE]Originally posted by Pappy
[B]I always thought the AI should be somewhat user programmable. Let the player decide what strategic moves the AI should make. In UV's case they would be something along the lines of "Defend PM" "Capture Gili-Gili". There should be weights set to each. Add a randomness factor as well. You should also be able to change these settings during the course of play to avoid AI mistakes of defending lost causes.

Just my $.02 [/B][/QUOTE]

That two cents is worth at least two bucks to me, Pappy. This is a novel, highly considerable idea, in my estimation. In so many of GG's games, I have wished that I could do something to rectify the AI's approach when I've broken through and started to kick arse. I know that this drifts toward "solitaire" play, but, why not allow the human intelligence to help out the artificial stupidity once in awhile to make for a more competitive game? After all, when you're playing against your own computer, you're engaging in a sort of "mental masturbation," aren't you? Why shouldn't you have the power to make it more painful on yourself?

:rolleyes: And the onlookers turned away disgusted at the "solo sex masochism" image presented by the pariah about to be stoned ...

----------------------

I will now proceed to entangle the entire area

(in reply to pasternakski)

Post #: 17

- 8/2/2002 1:29:18 AM

byron13

Posts: 1589
Joined: 7/27/2001
Status: offline

Like Ron, I'm going to admit near ignorance on the subject, but I did do some programming years ago. The problem is that we don't realize how many calculations we make in a single game turn or how many factors we take into account with each decision. Computers are incredibly stupid, and I'll cut the AI programmers a break. To truly mimic human thought requires a programmer to recreate the human thought processes, including humans' ability to make snap value judgments, distinguish shades of gray and when a factor is relevant or not. People with deeper pockets than Matrix have not managed it. As Ron suggests, we may see a company develop an AI routine that is fairly good and could be adapted to various games, but I think adapting it would be very difficult.

How many fighters should defend an island base? You could say place half as many fighters on the island as the enemy has bombers in range of the island. Do you count torpedo bombers that may be relevant for Truk but not for Guadalcanal? How do you adjust for enemy fighters? How do you account for and distinguish between enemy fleets in the area that are clearly not a threat from those that are a threat? How do you account for the quality of your own equipment and pilots v. that of your opponent? How do you weigh the enemies' supply condition? If you do account for supply how do you account for likely future supply? What about your own supply? How do account for the fact that, while the bombers may be present, the obvious focus of the enemy is somewhere else on the map? How do you balance the needs of this island against the needs of other bases or your own total fighter stength? What percentage is placed on CAP? Is it based on the number of bombers in the area (which would result in a high percentage of CAP even though the bombers have been pasting some other target for the past two months)? Maybe you base it on the number of aircraft that have attacked the island in the past week (but that cannot account for the number of fighters on the island, i.e., 60% of 12 available fighters is a lot different than 60% of 100 available fighters). Sit down sometime and actually write down the list of If/Then statements and the interrelated concepts that are involved in deciding how many fighters should be placed on an airfield and how they will operate. No matter how long and intricate the decision path is, there will be a dozen scenarios that will render the decision path faulty if not absurd. A human can easily see the error ("yeah, but I've got five CV's in the next hex so I don't need this many fighters . . . ."), but the computer, being stupid, cannot without being told.

I am not countenancing errors such as leaving ships in a port until they are inevitably sunk or sending waves of unescorted transports into the welcoming arms of an enemy surface fleet, but I am saying that pulling off a brilliantly coordinated assault is a tough thing to orchestrate. Heck, for all the money thrown at the issue, they still can't write a program that can do something as simple as consistently and reliably identify a human face. There just ain't no substitute for a human opponent, and it will be a long, long time before that changes.

(in reply to pasternakski)

Post #: 18

- 8/2/2002 11:48:55 PM

Toro

Posts: 578
Joined: 4/9/2002
From: 16 miles southeast of Hell (Michigan, i.e.), US
Status: offline

An interesting exercise I conducted once (once was enough) was to flow-chart the decision making process I might use in generating an AI for a game. This is, after all, how a computer AI would perform. I only did this once because the flow became so incredibly unwieldy that I decided to open a beer and go watch TV instead. The variables alone for generating a "reasonable" AI decision matrix for checkers is incredible. Chess (as aluded to earlier) is even more profound. UV... god help the designers/programmers.

I agree with the gent who noted that AI will get better, but I think computing power has to take a major leap forward first. 2 gig isn't gonna do it (nor 4, or 6), not when the goal is to challenge a human. In a second, we perform not only an astronomical number of "calculations" in this game to determine the best course of action, but we also have "experience" to draw upon.

Anyway, I'm getting long-winded. I, too, would like to see the AI much better (them APs at Guadal are sitting ducks), but to effectively challenge a human, you need another human -- or maybe a Cray or two, and enough lines of code to make a room of sane programmers cry.

(in reply to pasternakski)

Post #: 19

- 8/3/2002 1:09:48 AM

Sonny

Posts: 2008
Joined: 4/3/2002
Status: offline

[QUOTE]Originally posted by Toro
[B]....and enough lines of code to make a room of sane programmers cry. [/B][/QUOTE]

To quote myself - yeah, all three of them.:D

(in reply to pasternakski)

Post #: 20

- 8/3/2002 12:32:44 PM

CommC

Posts: 467
Joined: 8/3/2002
From: Michigan, USA
Status: offline

I wouldn't be too quick to condemn the AI tactic of 1)sending unsupported transports into areas in range of enemy air or surface forces, and 2)deploying more ground troops in an area than it can reasonably supply.

There are two possibilities we may be overlooking:
1. The programmers deliberately designed it this way to model the historical practice
or
2. These tactics are driven by operational necessity.

In the critical period of 1942, and later for the Japanese, both sides were guilty of these practices. Especially in the Guadalcanal campaign. The weapons of the time favored offense over defense... it was much harder to defend any surface asset than it was to strike an offensive blow. The essence of the game, and the historical war is/was how to defend the advancing transport carrying either a landing force into enemy territory or supplies for a forward deployed ground force. Both the Japanese and the Allies struggled with this in 1942. Finally, through brute force of numbers and quality the Allies prevailed.

In 1942, neither side had enough forces that it could adequately escort every transport advanced into enemy territory. The risk of loss was frequently too great. Neither side wanted to risk valuable combat assets on a mission as mundane as escorting a supply ship, as vital as that mission appears now, in hindsight.

(in reply to pasternakski)

Post #: 21

- 8/3/2002 4:17:06 PM

Drongo

Posts: 2205
Joined: 7/12/2002
From: Melb. Oztralia
Status: offline

[QUOTE]In 1942, neither side had enough forces that it could adequately escort every transport advanced into enemy territory. The risk of loss was frequently too great. Neither side wanted to risk valuable combat assets on a mission as mundane as escorting a supply ship, as vital as that mission appears now, in hindsight.[/QUOTE]

Understand what you're saying historicaly but, whatever the reason behind the AI routines, it results in shipping suicide in UV once a player secures air dominance over an area containing AI bases. I've watched the IJN AI send convoy after convoy to its death in its attempt to routinely supply bases in Northern New Guinea when the player's bases at PM and GG are overflowing with medium bombers (its nice to get rewarded initially for your efforts but, after a while, its about as much fun as kicking puppies). A month or two of losses like that and the AI will probably have lost more than 80% of its available shipping, often without even reaching its destination. In reality, even the determination shown by the Japanese during the Solomons Campaign would not lead them to openly expose their shipping to an environment like that (once it was recognised).

IMO, whether the AI allocated escorts or not to transports would make little difference in light of the "killing grounds" that players are able to create through air dominance. The real problem is that (quite understandably) the AI threat analysis is not capable of coping with a player's ability to concentrate his air force in an area. It would be too much to expect the AI routines to come up with a carefully planned operation encompassing selected air suppression attacks on player bases, constant LR CAP over transports, CV support, etc, all timed to coincide with the approach of a large, heavily escorted transport TF to its base destination.

What is needed is for the AI to recognise where and when it is time to stop the suicidal routine convoys and to start using only Fast Transport, Tokyo Express and/or submarine supply. Not an easy programming task. Maybe the choice could be built into a player controlled toggle similar to what was discussed earlier in this thread.

_____________________________

Have no fear,
drink more beer.

(in reply to pasternakski)

Post #: 22

- 8/4/2002 3:33:24 AM

CommC

Posts: 467
Joined: 8/3/2002
From: Michigan, USA
Status: offline

Yeah, the AI continuing to send unescorted supply transports into enemy air zones of control is simply a bug that should be fixed. I don't think it should be too difficult to fix from a coding perspective.

This discussion does highlight the tactics necesary for victory, though. Having not viewed the victory conditions in detail for the various scenarios, I can't speak to this too much, yet. But I would say the plan should go something like this:
1. Establish air supperiority over an advanced postion capable of supporting and air base.
2. Invade this position and land ground troops and engineers to construct an airbase there.
3. Establish surface sea superiority around this base if possible.
4. Use this air base to establish air superiority over a more advanced area, capable of sustaining air operations.
5. Protect vital supply ships to this base at all costs
6. Repeat

This is basically the operational strategy used historically. The person who can get there with the mostest the fastest should win.

UV players that deviate from this approach will have a tough time of it.

The trick comes in in that #1 is difficult, to impossible, for either side in 42, the forces are too balanced. As the US gets more air units later, it prevails easily.

To balance the game, the Japanese must have a large superiority in starting force levels in the theater. They must strike quickly. The Allies must hold them off until it can be reinforced.

(in reply to pasternakski)

Post #: 23

consensus? - 8/4/2002 9:20:28 AM

pasternakski

Posts: 6565
Joined: 6/29/2002
Status: offline

I am beginning to sense that a common understanding is emerging from this discussion (and what a thoughtful, intelligent discussion it has become). Let's see if I can summarize a little.

I started the thread by positing a half-dozen observations (only one or two of which were my own) about the AI's performance. The general thrust was that the AI plays its cards very close to the vest as either side, but often acts in ways that throw away its forces and advantages in a rather haphazard way.

Responses have tended to be of two kinds:
-the AI is imperfect, but, due to technological and time constraints, probably cannot be improved to perform at a "human" level of competency.
-the AI can be improved at the tactical decison-making level to avoid suicidal or silly acts.

There has also been some support voiced for the idea of giving the human player some control (through toggles or other choices) over the general direction of the AI's actions in order to assist it in playing competitively. Those expressing this idea believe that the coding difficulties, though considerable, would not be insurmountable.

Most responses indicate that limited improvement is possible, but "human-like" AI will have to await further technological advance in order to be possible.

Is this a fair assessment?

I want to compliment all of you torpedo-launching, invasion-planning, dive-bombing, airstrike-forming people on your intelligence and insight. Who says that wargamers are just a bunch of unit-counter pushers or mindless keyboard punchers?

I hope that this discussion continues to grow and stimulate thought and will be of use to our friends at Matrix and 2 by 3 who produce the stuff that addicts us so much ... so that they can figure out ways to torment us further

------------------------

I will now proceed to entangle the entire area

(in reply to pasternakski)

Post #: 24

- 8/5/2002 9:56:41 AM

tohoku

Posts: 415
Joined: 3/18/2002
From: at lunch, thanks.
Status: offline

[QUOTE]Originally posted by Toro
The variables alone for generating a "reasonable" AI decision matrix for checkers is incredible. Chess (as aluded to earlier) is even more profound. UV... god help the designers/programmers.

[/QUOTE]

??!!

Draughts is dead simple to brute strength it. Chess, less so, but still do-able. That's how Deep Blue does it. Nothing fancy about it. It's about as subtle a piece of programming as a brick in the back of the head.

[QUOTE]
I agree with the gent who noted that AI will get better, but I think computing power has to take a major leap forward first. 2 gig isn't gonna do it (nor 4, or 6), not when the goal is to challenge a human. In a second, we perform not only an astronomical number of "calculations" in this game to determine the best course of action, but we also have "experience" to draw upon.

Anyway, I'm getting long-winded. I, too, would like to see the AI much better (them APs at Guadal are sitting ducks), but to effectively challenge a human, you need another human -- or maybe a Cray or two, and enough lines of code to make a room of sane programmers cry.

[/QUOTE]

What programmers need to do is learn how to programme and design beyond first order logical bounds. I doubt that UV (I've now seen a copy in action) uses anything other than that. It's little better than PW AFAIC.

The problem with designing an AI is to *stop* trying to replicate the thinking of a human. The computer isn't and never will be, so it's stupid to try making it thinking like one. It is a machine. Symbolise the problem in a manner that is to the advanatge of the machine dealing with it and let the machine work it.

The more complexing and accurately you can represent the problem and give it decision weights (and that requires the use of more complexed logics (including, if you really want it to shine, non-standard logics)) the better the AI will be.

Current AIs suck because the programmers simply aren't good or qualified at what they do: ask a programmer if they know how to implement or use, say, a genetic algorythm and 99.9% will just stare blankly back at you. Hell, most them wouldn't even know what tense logic was.

tohoku
YMMV
Too much professional interest

(in reply to pasternakski)

Post #: 25

- 8/8/2002 9:52:04 AM

CommC

Posts: 467
Joined: 8/3/2002
From: Michigan, USA
Status: offline

After reading some of the AARs posted on the forum, of humans vs the AI, I would say it looks like the AI isn't too bad...could be a lot worse.

I don't have a copy of the game yet (its on the way), so I can't comment too much on this. But at this point my view is that the AI could probably be substantially improved with some simple, bug fix type coding changes. Then, after that, further improvements are possible without a requirement for vastly increased computing power.

The secret to a quality AI running on a modest computer is programmer skill. Hiring a team with the right skillset for a short time to do this is probably the best course.

I'm guessing that right now the UV AI calculations are consuming a very small part of the available computing power in a typical modern PC. I'll bet graphics handling is consuming a lot more.

(in reply to pasternakski)

Post #: 26

- 8/8/2002 10:27:14 AM

Joel Billings

Posts: 32265
Joined: 9/20/2000
From: Santa Rosa, CA
Status: offline

Actually, the AI is taking quite a bit of computer time. In UV it's still not too bad, but in WitP, it can take several minutes for the computer to get through calculating its move. In fact, it takes so long that our test games of AI vs. AI are pretty slow. At this point, the AI could not be made significantly better without virtually a total rewrite. Gary spent several extra months trying to adjust the AI based on test results that I fed him, but in most areas, every time he made a change to improve one thing, it threw something else out of whack. More time was spent coding and testing the UV AI than on any previous AI in a Grigsby game. I'm afraid that what we have is all we're ever likely to have, at least in UV.

(in reply to pasternakski)

Post #: 27

- 8/8/2002 1:20:18 PM

pasternakski

Posts: 6565
Joined: 6/29/2002
Status: offline

[QUOTE]Originally posted by Joel Billings
[B]Actually, the AI is taking quite a bit of computer time. In UV it's still not too bad, but in WitP, it can take several minutes for the computer to get through calculating its move. In fact, it takes so long that our test games of AI vs. AI are pretty slow. At this point, the AI could not be made significantly better without virtually a total rewrite. Gary spent several extra months trying to adjust the AI based on test results that I fed him, but in most areas, every time he made a change to improve one thing, it threw something else out of whack. More time was spent coding and testing the UV AI than on any previous AI in a Grigsby game. I'm afraid that what we have is all we're ever likely to have, at least in UV. [/B][/QUOTE]

And it's pretty doggone good. The cheating that seems to have been de rigeur in early AI designs is refreshingly absent from UV.

Curiously enough, as I refine my strategies and tactics, the AI seems to play smarter. It may be because I am finally learning enough about what works that I am following paths that were anticipated by the designers, so that my moves are better anticipated by the AI than some of the goofball stuff I was pulling when first trying this thing out to see what it could do.

For future purposes, I continue to believe that there is merit in building in toggles or other techniques that would allow the human player to assist the AI at critical times, as has been suggested by other contributors to this thread.

I guess now I'll get back to planning the demise of my PBEM pal who foolishly let my Allied forces slide into Shortland in early July '42. Heh heh heh. Who says that computers are the only things capable of "artificial stupidity?"

--------------------------------

I will now proceed to entangle the entire area

(in reply to pasternakski)

Post #: 28

Oh, one more thing ... - 8/8/2002 1:24:37 PM

pasternakski

Posts: 6565
Joined: 6/29/2002
Status: offline

Joel, please pass along my two cents worth on the AI taking a long time to plan its move. I'm into this for the duration (or eternity, whichever comes first) and want, when playing against the AI, to be confronted by the most able enemy possible.

So let 'er grind away at those bits and bytes.

---------------------------------

From my highly competent command staff

(in reply to pasternakski)

Post #: 29

- 8/8/2002 6:43:53 PM

John Carney

Posts: 66
Joined: 7/1/2002
From: Tampa FL
Status: offline

I agree, that the AI can take a couple of minutes, and this would not upset me. My chess game takes about 1 min for a turn on a 1.2 G machine.
But I would also like to see an ability to adjust the AI. I still enjoy Complete Carriers at War, because you can adjust a given scenario or write a new one while adjusting the AI threat threads to respond appropriately. Unfortunately it was not workable to do a campaign with this program.
UV is the best game I own.

(in reply to pasternakski)

Post #: 30