When do you PWN the AIs in an SP game?

[Skanky Burns]

I'd love to see Soren's efforts to make an AI for SMAC though. It would be impressive - he could probably even teach the AI to exploit crawlers too.

[Jamski]

Skanky, if you like we can have a game, and I'll be the "Ai" for you...

-Jam

[Skanky Burns]

I'll see if I can actually survive the first few turns of a Transend game, perhaps even win one against the current AI... maybe then I'll take you up on your offer.

[Snowflake]

Don't be afraid of him. He is not unbeatable.

[Skanky Burns]

Even at my skill level? May I remind the jury that in my latest game I discovered hybrid forests before tree farms? That is my skill level.

[Snowflake]

Be unpredictable. That is the key to your victory.

[RX2000]

It usually depends on the map size. If the map is small, I usually beat the crap out of anyone near me as soon as I can, so I usually have a pretty good lead from the beginning. If I'm playing on a larger map, I just work on expansion and usually fall behind the AI a bit.

[Jamski]

Don't be afraid of him. He is not unbeatable

Very true. I lost to Lazerus and the Hive once.

-Jam

[livid imp]

Originally posted by Trithemius
And vastly harder to program, as I understand it.
(I'm not an artificial intelligence expert, but some of my best friends are... :P).

I don't claim to be a AI expert either, but I am a programmer/aspiring game designer and have a general inclination how it is handled.
Look at how the RTS crowd handles it. They make a AI that is as close to perfectly efficient as possible then they use the difficulty slider to reduce the amount of clock cycles the AI uses in real time. This effectively makes the AI dumber, or slower. And that is handling multiple (sometimes hundreds of) instances in real time. Of course SMAC is much more complex than a typical RTS but it could be constantly be processing possible moves and counters during you turn. I would never expect the AI to ever come close to human intelligence. Such AI would be used for real life defense systems long before it would be used in a silly game. But I don't think it is asking too much for the AI not to blindly smash an endless stream of troops into a heavily fortified city just because it is the closest town.
Nothing in this world will get better without some demand for it to improve.

Originally posted by Trithemius
Giving the AI an advantage is a simple way of scaling the difficulty up,

That is a crutch to help a lame AI limp-a-long. Not a real solution.

Originally posted by Trithemius without needing to employ heaps of programmers all the damn time.

The game industry has been making more money than the movie industry for a few years now. To boot, programmers (even really good ones) don't make the kind of money that big name movie stars and directors make. The lack of money argument no longer holds water.
The new problem seems to be that (much like the movie industry) the more creative developers seem to get stuck in lower budget companies. While Blizzard can make big bucks just recycling Diablows and Warcraft 10 years later.

What Fraxis needed when developing SMAC is one of the number crunching psychos* around here to say things like "hey, the skunk works is completely pointless" and "make the AI crawl mins" back in beta.
For as great a game as SMAC is, there is still loads of room for improvement.


* I have the utmost respect for number crunching psychos, keep up the good work.

[Commy]

It sucks that on Transcend and thinker the AI gets benefits. And on civ, it's even worse, With the AI getting benifits on Monarch to Sid levels.

[Jamski]

What Praxis needed when developing SMAC is one of the number crunching psychos* around here to say things like "hey, the skunk works is completely pointless" and "make the AI crawl mins" back in beta.
For as great a game as SMAC is, there is still loads of room for improvement.

The only big failure in SMAC is that it is too good. By including so many options, they have opened it to exploits which the player can take advantage of, and at the same time hidden the optimum play choices so well that even the AI can't numbercrunch the best option out of it.

What other games are so replayable though?

/me psychopathically crunches some numbers *

And to think I used to tell Kody that it was no fun to do that

-Jam

[livid imp]

Originally posted by Jamski


The only big failure in SMAC is that it is too good. By including so many options, they have opened it to exploits which the player can take advantage of, and at the same time hidden the optimum play choices so well that even the AI can't numbercrunch the best option out of it.

This is true, and I don't want to seem ungrateful. SMAC has provided me with hundreds of hours of enterainment for like 40 or 50 bucks. That is one hell of a deal .
But I don't think it should stop game designers and programmers from pushing the AI envelope. I don't ask for perfection, just improvement

[Santiago_Claus]

SMAC-X is too easy because of bad AI? You want really bad AI, look at the original Civ. True, Civ was much harder than SMAC-X, but only because the AI cheated so much.

I lost interest in Civ when I realized after a few games that the "race to Sufferage" was a crap shoot. SMAC-X allows a better player to do better in the game, but in Civ it was pure luck.

You could expand and terraform and spy and research and keep your citizens happy in an ancient peacetime Republic whilst struggling to explore the world in sailboats that caused unrest as soon as they left port and skip the Horseback Riding tech branch just to claw your way to industrialization that much sooner. You could micromanage to the nth degree, do EVERYTHING right, and it didn't mean sh*t. If the AI was feeling generous that day, it might just hand Sufferage to the Babylonians. One might as well devote oneself to working out a system to beat a slot machine.

No wonder the Trailer Park Chariot Rush was so popular: like a system that really would beat a slot machine, it was an exploit against a rigged game.

SMAC-X may have grown too easy to beat, but even so it's still interesting to play. That in itself represents a huge accomplishment in game AI. Rejoice, gamers: be thankful that "AI" has advanced beyond the capricious random number generator.

[Sikander]

The reason that we don't have great AI in SMAC is that it is simply beyond the capabilities of our society and our tech at this point. A simple example would be to compare chess, a simple game that only recently and with large computing and programming resources has become a challenge for great players to beat the AI. SMAC is many orders of magnitude more complicated than chess. The board is much larger, the terrain is variable and can change as the game goes on. Instead of alternating movements of one piece, players instead can move all their pieces in the same turn, and in some cases move a piece, move another and then move the original piece again. Combat is not a known quantity, so one has to play the odds rather than simply moving somewhere and capturing a piece. Pieces can be manufactured during the game. The map is unknown at the beginning of the game and throughout the game large portions of it are unknown as far as what units may be occupying a square. The rules also change as the game moves on, as new techs are researched, SPs built etc.

So a true chess like AI isn't going to be a reality for a game like SMAC for a very long time. Instead we have scripts which are damned good for what they are, but repetitive predictable and uncreative.

[Skanky Burns]

Chess also has only 64 squares which can be occupied by only one unit. This greatly reduces the processing power needed to evaluate moves, yet still requires a supercomputer to do so. Even the smallest SMAC map has many times that number of tiles, each able to hold many units. An exponential increase in complexity, an exponential increase in the processing power needed.