Tactics, what is the point?

[tiagocc0]

People usually think that better AI is an AI that's impossible to win.
It's actually easier to make a impossible to win AI then it is to make a human-like AI, unless you have too much variables at play.
It would take longer to process and developers use heuristics to make it go faster and thus creates 'glitches' in the AI, some even use it to say they are more 'human-like'.

So when some say better AI they usually don't mean the impossible AI, but the AI that acts like it should.
Zombies acting like zombies, mercenaries acting like mercenaries and so on.

If the character acts like it should then you have as much variety from them as you could get.
Then we can move on to other things. Like the mechanics, features and rules.

 

[Midair]

I don't think it's about having superhuman AI, but having variety to keep things interesting for the player, and having AI that's sufficient and makes sense for the personality of that particular enemy. Granted, if the idea for that enemy is about superior AI, then go ahead.

Wanting various different AIs to keep things interesting is completely different from wanting sophisticated AI to provide greater challenge. This is why I think there are 2 answers to the question "what is the point of tactics?" For some people, they just want variety. They are content with a micromanagement or housekeeping style of play as long as there is variation to prevent it from getting too repetitive. Other people want a mental test, maybe not a puzzle, but something requiring clever, creative thinking like you would use in a multiplayer game against a tough opponent.

I think each desire is valid but they have very different implications for how to approach the design of tactical combat.

So when some say better AI they usually don't mean the impossible AI, but the AI that acts like it should.
Zombies acting like zombies, mercenaries acting like mercenaries and so on.

I do not think that is what most people are saying. Sure, that is a reasonable thing, but when someone says they want better AI, I think they mean more human-like and challenging AI, comparable to multiplayer.

If someone says they want a variety of enemy types, that carries an implication that they expect tactics to be mostly micromanagement and therefore want variety so it doesn't get repetitive.

 

[tiagocc0]

AI comparable to multiplayer or 'human-like' behavior is an abomination, at least for me.
It just means that the AI is unpredictable which is a very bad type of variation.

For the AI to be able to have a good type of variation it needs to 'know' every possible 'trick' and also it's role in the 'world'.
That means how to use every kind of magic/skills, weapons, terrain, tactics. In other words it can do anything an human player can.
Then we define the character of the AI, if it is a zombie you will then exclude everything it can't do like using weapons.
If it is a mercenary it could probably do anything and will be limited by it's own skills and attributes.
Then we can define the personality, like the zombie is dumb, the mercenary is greedy and so on.
Which affects things like when to run/hide/attack.

Having zombies act more human-like is a no-no.
Mercenaries act human-like by their nature yet they shouldn't however act like the player (as in multiplayer) as they are not trying to win the game, they are trying to do what they were paid for.

We shouldn't have an AI playing like the player, this might work on a multiplayer focused game where you don't have enough players to fill the slots so the AI must imitate the player.

However on a singleplayer game or on a singleplayer focused game you really shouldn't be trying to make AI that act as a player.
It's out of the context and can also lead to nasty behavior, in my opinion.

We also have several types of AI, one would be for the unit, you could have another that control the entire team.
The leader of the other team could have the same goals as the player for that mission, so the character/personality of the AI for that team leader could act on that mission the same as the player.
But that is very specific and we shouldn't be giving all AI leaders the same goal as the player in every mission.

EDIT: I'm sorry for trying to explain this, but every time I hear someone saying AI with human-like behavior or worse, AI with 'multiplayer' behavior I have a convulsion.

EDIT2: Example, in Master of Orion 2 (which I played a lot) all the races act in the same way, even if they have different racial traits. They have minor variations in how they play, but they all have the same goal which is to win the game.
So they act like in a multiplayer game, which is reasonable given there is a multiplayer option.
But how better Master of Orion 2 could be if each race had their own goals which is not to win, but to survive.
Races have a different view of the universe instead of every race using the same features to achieve the same goals.
I truly believe that multiplayer is the killer of good AI.

 

[Anomalous Underdog]

I don't think it's about having superhuman AI, but having variety to keep things interesting for the player, and having AI that's sufficient and makes sense for the personality of that particular enemy. Granted, if the idea for that enemy is about superior AI, then go ahead.

Wanting various different AIs to keep things interesting is completely different from wanting sophisticated AI to provide greater challenge. This is why I think there are 2 answers to the question "what is the point of tactics?" For some people, they just want variety. They are content with a micromanagement or housekeeping style of play as long as there is variation to prevent it from getting too repetitive. Other people want a mental test, maybe not a puzzle, but something requiring clever, creative thinking like you would use in a multiplayer game against a tough opponent.

I think each desire is valid but they have very different implications for how to approach the design of tactical combat.

If someone says they want a variety of enemy types, that carries an implication that they expect tactics to be mostly micromanagement and therefore want variety so it doesn't get repetitive.

Yes different, but why can't a game have both? I.e. different enemy personalities at the same time some of them have "human-like" AI.

Putting in smarter AI (wherever it makes sense to put it) will, of course, help any game, regardless if it has enemy variety or not.

Can you clarify? What do you mean by micromanagement/housekeeping in this context? From what I read it's like you imply that having enemy variety means the player doesn't need to be clever and creative when dealing with the enemies anymore. If the game has a variety of enemies, they *all* need to have predictable behaviour? Is that it? Having sophisticated AI means you can't have variety anymore?

If any, I think using different unit types in various combinations allow for creativity.

The idea here is role specialization. A sniper specializes in ranged attacks. A tank specializes in protecting teammates. And if that makes it sound that they will have predictable AI because each only do one thing, think about it this way: it's about team-play. Think of the whole enemy squad as one organism composed of different autonomous parts. How well (or bad) they cooperate with each other, what their squad is composed of, that's what gives the AI sophistication. If their AI is good enough, the enemy squad can adapt to any situation (with varying degrees of effectiveness) that the player throws them into.

And I don't think the enemies alone will account for the "mental test" or the variety. There are a lot more at play here, like the map's shape, terrain types, different victory/defeat conditions other than "kill all enemies", what weapons you have, heck, maybe even weather condition or time of day during battle that would affect the combat rules.



I came across a topic like this before in a podcast. They talk about breadth and depth.

Breadth: allowing different ways to win.
Depth: requiring many steps performed in a specific order to win.

And they said that puzzle games (crossword, match-3, I think they even mention card games like Klondike or whatever), some of them are heavier on depth, some heavier on breadth. Then some guy measured all games he could, and found a puzzle game (found or made one, I forgot the details) that had equal parts breadth and depth, and that players loved it.

I also think it's the reason why Deus Ex is hailed as a very good game. But I digress.

 

[Midair]

Tactics [with variations for different enemy types], what is the point?

Does spreading AI behaviors across specialized units somehow make tactics better? Not really. Chess has more kinds of pieces than Go, but this doesn't affect tactical depth. I did not mean to imply that enemy variety would weaken tactical challenge, just that it is irrelevant. I agree that variety and challenge are not in opposition, but they are not related either. You do not need one in order to have the other. So if you think the point of tactics is challenge, then enemy variety should not be relevant.

So, what is the point of tactics?

If you think it is to make the player think, then deep AI seems like a good solution to me.

If you think it is something else, then we should be discussing what exactly.

Can you clarify? What do you mean by micromanagement/housekeeping in this context? From what I read it's like you imply that having enemy variety means the player doesn't need to be clever and creative when dealing with the enemies anymore. If the game has a variety of enemies, they *all* need to have predictable behaviour? Is that it? Having sophisticated AI means you can't have variety anymore?

By micromanagement/housekeeping I mean like what you do in a sim game. It is just a matter of being attentive and looking after everything. There is no risk that your opponent is going to trick you or set a trap that will ruin your strategy. Contrast this with a multiplayer strategy game or with solving a puzzle. In both of these cases you have to think about what your opponent (or the creator of the puzzle) are thinking. I hope I have answered your other questions above.

 

[tiagocc0]

Variation has to be somewhere, you can put it into the rules (go) or into the characters (chess).

EDIT: In the end everything is a rule, be it how you move, how you win or rules given to specific characters.
The number of rules doesn't matter, you can have depth with a few quality rules or with lots of them.

Enemy variation is one way to do it. One type of rule.

 

[tiagocc0]

That is the minimax algorithm, which uses brute force (tries all the options), there are heuristics in which cuts the tree where it knows it won't get best results and also cut the depth of the tree (check 1 turn ahead, check 2 turns ahead, check 3 turns ahead..).

While this is a great initial AI, you will probably want one that is able to formulate elaborated plans.
Like in chess where losing pieces may be just a way to lure your opponent.

To make plans you would use an AI that is able to pass commands to the other AIs (units).
It needs to not only make plans but also identify plans being made by the player (this is the tricky part).
Also it needs to be able to change plans in the middle of the action in case it knows the current plan is bound to fail.
Some mistakes that people usually do with this type of AI is to change plans too quickly, so in the end it looks like no plan was used at all.

 

[tiagocc0]

Yeah, but the AI that makes the plan should not uses the minimax algorithm.
It should use each and every information available, you will need to store a log (history) of what happened on the last turns, like 10 or 20 turns and also consolidated information like the sum of hp of all your enemies, who is the strongest, who is the weakest.
The more information you are able to squeeze out of your current game state the more tools you will have to predict your next move.

So the prediction algorithm does not look into the future like you described but looks in the present and past to spot common patterns and then issue changes in the plan.

EDIT: It should be fast because it would be using your knowledge to make a decision instead of testing every possible move and then issue a value to it.
Like a medical software where you enter information on what you're feeling, where is the pain and the software tells you the possible 'diseases' you may have and % of each of them.
http://en.wikipedia.org/wiki/Artificial_intelligence#Knowledge_representation

Which means the AI will be as good as the person who feeds information to it.

EDIT2: In the end your AI will be a mix of the above plus:
http://en.wikipedia.org/wiki/Artificial_intelligence#Planning
and
http://en.wikipedia.org/wiki/Artificial_intelligence#Deduction.2C_reasoning.2C_problem_solving

 

[tiagocc0]

Yes, the example that I gave is to make plans, which would be used for an AI that controls an entire team.
Like a group of mercenaries.

If you have monsters like trolls, goblins and orcs and in your universe they are not good at teamwork then it makes sense each one acting individually.

It is hard to implement, yes. It is the best way to go. You can mix it with genetics or neural AI if you want to alleviate some of the expertise required to fully implement it. But in the end the best AI is the one that was made by an specialist and does not require any fancy algorithm at all. Unless you want it to 'learn' as it goes.

It's not about having a challenge all the time, but to have some intelligence in the process, like the AI knows you are trying to flank it and is able to responding accordingly, maybe in a stupid way depending on the intelligence of each individual but it is able to identify your actions.
Instead of always falling into your traps all the time.

 

[Anomalous Underdog]

Tactics [with variations for different enemy types], what is the point?

Does spreading AI behaviors across specialized units somehow make tactics better? Not really. Chess has more kinds of pieces than Go, but this doesn't affect tactical depth. I did not mean to imply that enemy variety would weaken tactical challenge, just that it is irrelevant. I agree that variety and challenge are not in opposition, but they are not related either. You do not need one in order to have the other. So if you think the point of tactics is challenge, then enemy variety should not be relevant.

So, what is the point of tactics?

If you think it is to make the player think, then deep AI seems like a good solution to me.

If you think it is something else, then we should be discussing what exactly.

Ok ok now I get where you're coming from. I agree with you.

I'm not going to go debating about which is deeper: Go or Chess. The other guys can argue with that all they want.

All I'm bringing into the discussion is how to improve a tactics game in theory. I'm not really into about the semantics/definition. And the point of what I'm saying is that both breadth and depth equally make for a better experience for the player in the end.

You *could* make a game where all units are the same (i.e. Go), but well, I'm just not into that.


----


Ok about the actual AI, I'll share a few things about how programmers do it.

One way to do this that I have always preferred, is to vary the ability of the enemy AI to simulate the battle plan.

...

The enemy AI considers the current battle lineup and simulates all the moves it can make within that turn and estimates the outcomes. Then it tries to simulate what the Player will do and compares the result for the given "turn" that includes player and enemy moves.

Anticipating what the player will do, that's typical of Chess AI actually.

But what I want to note here is that letting the AI create a series of steps to do, in the first turn alone, before he even moves any of his units, that's what "planner" types of AI do. "Planner" is an umbrella term and there are many implementations about it.

The first significant one was G.O.A.P. (Goal Oriented Action Planning). This was the AI used in that FPS game F.E.A.R., Empire Total War, and Fallout 3.

Then there's H.T.N. (Hierarchical Task Network), used in Killzone, and those recent Transformer games.

And finally there's Behaviour Trees, the ones used in Halo 2 (and presumably onwards in the series), all the Crysis games, Spore, and probably a bunch of others I don't know.

The problem to note here is that even though the AI can make a plan, it also has to worry about being able to make contingency plans (make a plan B if plan A didn't work). Some implementations struggle with this more than others.

If you guys wanna know the nitty-gritty details on that I made a talk about it before (PDF slides): https://drive.google.com/file/d/0B2Y7jOiWWy4MMDhVeE1GclBaLVU/edit?usp=sharing

Or you could just Google those things I mentioned.

----

If the AI is smart, it chooses the optimal solution i.e. it chooses the answer where it gets the least "damage" and payer gets most. If it is "Dumb" it chooses enemy does most damage (suicidal AI) regardless of damage received. If it is cowardly it chooses the least damage to itself option etc.

This idea of using the optimal solution, in practice, games so far (AFAIK) can only do these reasonably for the short-term. Meaning AI can indeed choose an optimal action to do but it can only reliably choose the best thing to do for this turn only. This is from what I hear, I haven't really tried, to be honest. I guess it's too much strain on a computer to make it think of an optimal solution in the long term always (because the problem I think is taking contingency plans into account: too much permutations).

They call these types of systems "Utility" AI. There's another type of AI system that performs the same, called, decision trees. Not sure if those two are really different in practice. Depends on your particular implementation I guess.

The basic idea here is to make the AI list out all the actions it can do at the moment, as in, everything, charge, retreat, flank, snipe, use his buff ability, whatever. Then assign a "usefulness" score on each action based on "guesses" on the current situation. Then it simply chooses the one with the highest score.

How it actually comes up with that usefulness score, well, it pretty much depends on the formulae you put (which should take the game's rules into account).

Some games also put a "preference" score in addition to the "usefulness" score, to account for various AI personalities, and doing that lets designers tweak the AI personality easily by just changing the preference values.

----

This dichotomy between thinking on your feet (and choosing an immediate gain), VS making investments (losing a little bit at the start with the promise of much better gains later on) and planning for the long term, reminds me of Rommel VS Montgomery in the African theater of WW2.

Rommel was a fast thinker (and did well for the limited supplies given to him), while Montgomery was a methodical planner. Of course, Rommel lost but don't let that make you think that thinking on one's feet is inferior. There were more factors for that battle's results than the personalities of the generals on either side (morale, limitation of ammo & supplies, availability of air support, arguably even political meddling, etc.).

----

You do, of course, have your conventional "finite state machine" that you could use but that pretty much fell out of favor for use in AI (still used in other parts of a game though).

----

For example you can allow the really smart villains in the game to use the past information against you.

If you want this to be automated, I think the keyword you need to Google is "machine learning". The idea is an AI that starts dumb (relatively) and learns from its mistakes automatically each time until it gets really good. I think the pet character in "Black & White" uses that, not sure.

This is a bit related to evolutionary AI, meaning a survival of the fittest kind of AI creation: make a bunch of dumb AI randomly. The ones who perform at least a little good gets to stay, the rest are deleted. Then among the remaining ones, you create new combinations. Repeat the process, the good performers stay, the rest deleted. Until you get to a point where the AI is pretty good.

For the most part, I think it's really hard to "guide" AI like this so it could be fun for the player. It could end up just being too annoyingly smart (i.e. troll) if not done properly.

However I do think it's a good tool for you to fish out overpowered tactics in your game (because the evolved AI will simply exploit any cheesy tactic), to help you tweak the rules. So if you don't have a legion of playtesters to do QA for your game, this will do in a pinch. Check out http://aigamedev.com/open/interview/evolution-in-cityconquest/

 

[Anomalous Underdog]

Oh wait I'd just clarify though. The OP topic is indeed "what is the point of tactics", but I get that the discussion is really about: the real reason why some people think tactics has little point in RPGs, is because they were implemented shoddily in most "tactical" RPGs.

The OP's initial argument was that there's little point for tactics in RPGs and that we might as well go back to RPGs where characters have no movement/positioning.

So that' why I'm offering suggestions for improvement.

 

[tiagocc0]

I find your points pretty valid.

 

[set]

Tactics exist where players have options in an environment to solve a problem.

This has less to do with AI, because good AI is just going to force a player to be reactionary - in JA2 1.13, I tend to put all of my squads in buildings prepared to fight out one looooong ass battle, slowly whittling down the enemy in a well defended location. There aren't that many other options, especially in the early game. In-fact, as much as I like JA2, I don't think it's a good example of a tactical game, because most of the time you creep along until enemies are just in the fringe of your line of sight... and then you kill as many as you can in the ensuing exchange, with you initiating the combat giving you a huge edge.

Er, maybe it's a poor example. But what I mean is.... When it comes to tactics or strategy, it's wiser to treat the game as a binary tree of sorts. You've got your root state on the top, where everything is laid out, and then, you as the designer, ask yourself... how many states should I give to the player? A computer alogorithim could do this procedurally when generating a dungeon, for example. Consider this idea:

ROOT (Entrance to a dungeon) --> ROOM 1 --> ROOM 2 -> ROOM 3 -> EXIT/GOAL

In the exit/goal room is the mcguffin that will save the princess, so naturally the player will navigate there. Common level design, especially for RPGs developed by BioWare, is to build a level like I have already laid out - you have some rooms and they connect linearly to the goal room. Maybe there are some closets to explore along the way, but the paths don't really branch.

A binary tree... is binary. Every node-room can have two children coming off it. A computer can quickly generate and climb down these trees to make assessments.

Let's say ROOM 1 has bandits, ROOM 2 has bandits, and ROOM 3 has more bandits. In a traditional BioWare RPG (or whatever), you kill the three bandits and save the princess. In something modeled after a binary tree...

ROOT -> ROOM 1 -> ROOM 2
ROOM 1 -> ROOM A
ROOM A -> ROOM B
ROOM 2 -> ROOM 3
ROOM B -> ROOM C
ROOM C -> EXIT
ROOM 3 -> EXIT

Of course, in this layout, we've just created two similarly linear paths from ROOT to EXIT, but in a way, we've broken up the level. You now have two distinct choices. Let's say rooms A,B,C are populated by kittens. Now there is a tactic - you can go through the rooms with harmless kittens (the computer weighs each room as a difficulty of 1) or you can go through the rooms with bandits (the computer weighs each room as a difficulty of 5), making each path have a weight of 3 or 15. A computer-controlled algorithim realizes that path 1 is significantly more difficult than path 2, so it populates path 2 with locked doors that require a lockpicking skill of 100, making each path now have a difficulty weight of 15.

Now we've created some form of tactical... or at least strategical thinking. If your party has a rogue, you can breeze through without combat, if your party has a mage with an anti-bandit spell, you can breeze through without needing to make sure you've got enough lockpicks or whatever.

This kind of model is still rather boring, but it can be extended to create dungeons that have more than one ROOT node and more than one exit node - though this no longer resembles a binary tree structure, but something like an edge list or something. But the point is, tactics or level design could all be dictated by a computer algorithm with sufficient planning; it's not impossible for designers to provide many diverse scenarios that are created on the fly with this general strategy. It's easy to program, in theory, though it's not something I've ever seen anyone express as a potential idea of level design. I mean, we've seen procedurally created levels, but I've never seen procedurally created 'solutions'. Arm said algorithim with a vast array of things it can do to switch the game up (locked doors, passphrases for doors hidden elsewhere, various monster encounter groups, friendly NPC or faction character placements, et cetera) and you can begin to (at least I hope) arrive at a game where you're constantly evaluating what's the best path to take for your band of heroes.

The point of tactics I think, is we like navigating these thought-puzzles, finding the best paths to victory, while also being entertained by a narrative and characters and the luck of the draw, and such.

 

[Midair]

Let's make a distinction that has been missing from this discussion, which is between symmetric and asymmetric opponents.

There seems to be agreement that a rival adventuring party should have AI that is as human-like as possible. This type of encounter involves symmetric opponents, meaning they each have similar options and rules.

But if you consider something that behaves much differently than a player would, like a mob of zombies or maybe a trap in the environment, then you have asymmetric opponents.

With symmetric opponents, we can describe the way tactics work in this way: Each side tries to recognize their opponent's tactics and adapt their own. A player will figure out and adapt to an AI, so the AI must be strong enough to recognize and adapt to the player's tactics as well.

Now with asymmetric opponents, the asymmetry might provide an advantage to supplement weak AI. But here I believe you lose the tactical challenge that comes from the back and forth of recognizing and adapting tactics. You could have asymmetry without advantage, as in different races in a well-balanced strategy game. But then you would again need strong AI.

In the example above of only having only one viable tactic, I do not think this is caused by good AI. It is the opposite, the enemy is overpowered to compensate for weak AI.

The generated dungeon you describe is a kind of asymmetric opponent. It would still require strong enemy AI if it is going to offer a tactical challenge. You seem to be saying that tactics would exist just because the player has options. Tactics, to me (if we are talking from a gamist perspective), means having to outsmart an opponent or environment in some way. Just having the option to take one path or another does not present a tactical challenge.

 

[tiagocc0]

In the example above of only having only one viable tactic, I do not think this is caused by good AI. It is the opposite, the enemy is overpowered to compensate for weak AI.

 

[set]

I guess I didn't present the whole picture, but you don't need an active antagonist to have tactics. A tactic is anything which overcomes a problem. A generated dungeon is a "problem" because you have to do something inside it - reach a person, get a thing, kill a thing, etc Choosing the right tactic in a given instance results from weighing your options carefully and choosing the right one. That is tactical play, at least how I see it. You can vary tactics by giving the player an opponent which has its own tactics it tries to employ against you, but enemies don't need to have tactical AI if they are being placed by an algorithm that already ensures there is tactical choice in 'solving' a dungeon.

 

[Midair]

I guess I didn't present the whole picture, but you don't need an active antagonist to have tactics.

Even with something like a trap or the dungeon as a whole, which I called asymmetric opponents, you need an active antagonist. Then we can define tactics as a process where two sides attempt to recognize what the other is thinking and adapt to give themselves the advantage. I cannot think of a definition of tactics in which one side is not an active antagonist.

A tactic is anything which overcomes a problem. A generated dungeon is a "problem" because you have to do something inside it - reach a person, get a thing, kill a thing, etc Choosing the right tactic in a given instance results from weighing your options carefully and choosing the right one.

This is too vague. It does not explain what is tactical about the choice of options. Again, you seem to say that just having different options makes something tactical.

What is unclear about your dungeon algorithm is how it is supposed to create tactical challenges. When the player enters a room, if they don't like what they see do they just retreat and try a different room? Options != tactics.

 

[set]

You're looking at it from a perspective of 'tactics in combat'. I'm looking at it from a perspective of 'tactics in dungoen crawling'. Yes, if you encounter a room you can't cope with, or if there is evidence to suggest (a good dungeon generator would give players clues to know what would be contained in rooms ahead) a certain path is bad, then you retreat. Retreating is in-fact a tactic few RPGs allow you do to anymore, ever since WoW introduced insta-regen monsters and never-ever-stop-chasing-you-AI.

Within fights, there should exist skills under the same kind of schema, where there exist at least two options to deal with fluffy kittens - for instance, you can either use the pet skill or the give-cat-nip-skill to deal with cats. In a shallow constructed game, the pet skill would put you a danger of being bitten if you're petting an aggressive cat, while a passive cat might be roused into a sense of fury from cat-nip, so the tactic is, 'employ the correct skill' - of course, design like this is shallow because there are only two options and it all seems rather arbitrary. But the point is to extend the theory - offer choices which are meaningful, intuitive, where the potential results are able to be grasped from the onset (again, why a good dungeon gives players clues about what they will encounter inside it, so they can prepare and anticipate). If there is no choice, that is to say, no /balanced/ choice, there is no tactic to be had. Giving balanced choices - choices which have equal weight (or weight that varies depending upon the situation, where a choice never has a dominant position too frequently, or where an optimal choice is not easy to immediately grasp) - is all you need to create a tactical game. The absence of meaningful choice /is/ the absence of tactics, but yes, the simple presence of choice is not necessarily enough, but I've been stressing how choices /should/ be weighed to offer /real/ tactics.

In real life combat, you are presented with a situation. You can take cover, or you can charge, for instance. Each action has its own weight of effectiveness that varies depending upon the situation. If you don't have the option to take cover, because there is nothing to take cover with, then there may no be any other tactic available but to charge blindly. A tactical game presents conflicts where there are multiple tactics to employ, where one or more may be the 'correct' tactic to solve the game. Too many games allow you to solve them with the use of any tactic or strategy, or where one tactic or strategy dominates all others to the point where we say a game has 'no tactics'. If a game really offered 'no tactical choice' then the game would not be solvable. We say a game effectively has no tacticalness because there is no effective choice - it can all be reduced to this idea of 'linearity'.

 

[Midair]

Giving balanced choices - choices which have equal weight (or weight that varies depending upon the situation, where a choice never has a dominant position too frequently, or where an optimal choice is not easy to immediately grasp) - is all you need to create a tactical game.

Quake has different weapons and paths through levels. Is Quake a tactical game?

 

[Delterius]

In respects to the AI discussion, a few years back I had a friend who always complained that the computer in games was way too easy. He always asked me to provide a worthy challenger. Time after time I failed to do so, but eventually I got the hang of it. When my friend was finally defeated he started crying and complained that all I did was make a box with a human inside it so the game was stacked against him. Indeed, the only way to escape ultimate boredom seemed to be stacking the game against the human player.

Afterwards I had a talk with him and as it turns out all that had happened was he got too old for that shit. All he wanted was to have fun, but when you know the ins and outs of a single player game there's nowhere else to go. It appears that the fun lies in discovering the game's mechanics and in his case the journey had ended. Fin. Time to move on.

Challenge I think lies in a variety of options and behaviour patterns that utterly surprise the player. Challenge comes from ignorance. And Fun comes from exploration of the world. Trying to combat complete mastery over the rules is like adding a new super dungeon after the Main Quest full of HP and Level bloats.