Fixing things that weren't broken

[PieceOfMind]

As you can quite clearly see, all the AI does is check for conditions and react too it. In fact, civ4's AI has to ignore the unit visibility rules to act intelligently because it can't articulate that if it just saw a unit it's still nearby... if it sees something in one turn (or even part of it's turn) but not the next, as far as the AI is concerned the unit never existed. It will never approach a human's intelligence, at least not until we have quantum computers.

I think that's the second time you've mentioned quantum computers. I know this is off the topic of the present discussion, but why do you say it? Is it sort of an assumption of advancement of algorithm knowledge once such computers are common, or are you talking about a specific problem that they can solve efficiently?


By the way, I don't even know what you two are arguing anymore.

 

[deanej]

I mention quantum computers mainly because they theoretically can process information (particularly unstructured searching) a lot faster (including breaking RSA encryption, which is currently "impossible" because it takes way to long to perform the math necessary) and because it's theorized that consciousness arises from quantum mechanics.

They do offer new ways of using algorithms as well; in classical (current) computers, the only reversible operation (one that can be undone by performing it again) is the not operation. In quantum computers, all operations except measurement (reading the data in a variable) are reversible.

 

[PieceOfMind]

I mention quantum computers mainly because they theoretically can process information (particularly unstructured searching) a lot faster (including breaking RSA encryption, which is currently "impossible" because it takes way to long to perform the math necessary) and because it's theorized that consciousness arises from quantum mechanics.

They do offer new ways of using algorithms as well; in classical (current) computers, the only reversible operation (one that can be undone by performing it again) is the not operation. In quantum computers, all operations except measurement (reading the data in a variable) are reversible.

I'm aware of the basics of quantum computers which is why I'm asking, and yes I think Brian Greene is one of the most known people to have advanced that theory you mention. To be fair, I think he has only hypothesised (as many neuroscientists would) that consciousness requires quantum mechanical effects.(sorry, I guess I'm only sort of repeating you. The point is it's a different topic) For qauntum computing only a very specific aspect of the theory is taken advantage of. In particular it is probabilistic algorithms (like those that could break encryption) that have been put forward as being the main advantage (or hazard depending on your perspective) of the new technology. I think it's quite a large leap to go from those algorithms to arguing that video game AI can be revolutionised similarly.

So while maybe it'd be conceivable to have a faster pathfinding algorithm built into the game that can be utilized by computers that have a quantum computer as one of their components, I don't see how it changes anything at all about the tactical/strategic potential of the AI. Just writing it to take advantage of parallel processing on multiple cores I assume to be a pretty challenging task for a programmer.

*Basically I'm saying that quantum computers are known to be able to solve fairly particular subsets of problems in a more efficient manner (i.e. polynomial time) than conventional (classical?) computers, but this doesn't mean that they "process information faster" in general. There's quite a difference.

 

[LordTC]

You're basically asking the AI to play to your play style. The AI in this case isn't going for diplomatic. It decided it's condition at the start of the game. What you're asking it to do is check every victory condition and assess the likelyhood of achieving it every turn, switching accordingly, and doing so in a coherent matter. Even if it was implmented (which it won't be), it would result in the AI changing the victory condition it goes for very, very frequently. Either way the AI will lose, it's just a question of how much processor power it will take.

If you actually want an AI that changes strategies faster than a ferret on a sugar high, feel free to suggest the AI do this.

How about the following:

When the AI gets its 4th social policy tree, check if it is worth switching to a cultural victory.

When the UN is built check if its worth switching to a diplomatic victory.

Basically some conditions that prevent aggregious behaviour like not going for a culture win when very close to it, or losing to someone elses UN while sitting on 60,000 gold

I also think pre-determined victory conditions is pretty awful for some of the smaller maps where you might not have a single AI playing for a certain type of win.

 

[Nares]

And how often do you propose to have the AI check for the best victory?

I think the obvious point he's trying to make is that the AI could easily be programmed to assess and pursue a diplomatic victory as soon as the UN is built, regardless of its previous victory goal.

Until its built, the AI blindly follows its pre-selected victory path; presumably at least one of them will blindly pursue diplomatic victory, with a chance that none of them do and none of them build the UN (what a wonderful world that would be).

Additional checks could be added for other milestones. Science viability could be checked every time the AI reaches a new era. Conquest viability could be checked each time an AI acquires a new city by any method.

Still need more basic problems worked out, like the AI being totally schizo in its combat choices, ie bombarding the full health unit when there's a 1hp target within range, or failing to use a ranged attack when there's a target in range and instead entering alert mode or something equally useless.

 

[fed1943]

Interesting thread.
At first seems 2 people are saying the opposite...but no, they are speaking of different things.
The civ5 game and the civ5 program.
(Before you say just the program exists, the game exists as long there are players).
If the civ5 program people tried to make a game that the program can't handle, they made a mistake.

 

[deanej]

If you simply had the AI check for diplomatic when the UN is built, wouldn't you then have people saying that the AI is still not playing to win because it's not proactive about building the UN itself when it can easily get diplomatic and waits to check until someone else builds the UN?

 

[Nares]

If you simply had the AI check for diplomatic when the UN is built, wouldn't you then have people saying that the AI is still not playing to win because it's not proactive about building the UN itself when it can easily get diplomatic and waits to check until someone else builds the UN?

There will always be some level of critical thinking which players will fault the AI for not exhibiting.