Designing Difficulty Options In Games

Gamasutra is running a story about how the "hard" modes in games can be designed to include difficulty, but not frustration. They give some examples of the changes made to several games as their difficulty settings are increased, and they discuss some of the simple options, such as increasing the number of required button presses, or increasing the relevant numbers by an arbitrary amount (a boss on easy may hit you for 10 damage, whereas a boss on hard may act the same but hit you for 100 damage). They also talk about maintaining the "illusion of fairness." Quoting: "Bungie's Halo series is often praised for its excellent execution of difficult play in the form of its Legendary mode. Not surprisingly, the team took a very well-thought out approach to introducing and tuning difficult play. Halo 3 gameplay designer Francois Boucher-Genesse explains that it's not just a case of one formula fits all. 'It's not like we just cranked every enemy's health by 200% and called it Legendary,' he said. 'There was a good amount of custom changes made per mission as well. In that sense we encourage players with previous Halo experience to play at least on Heroic, since they get to see the game in its full scale.'"

Comments

Maybe about the curve?

[nine-times]

My biggest gripe with game difficulty that comes to mind is when I feel like it's making the whole thing hard for the sake of being hard. Guitar Hero 3 comes to mind. It's like they're assuming you've played the other Guitar Hero games, were good at them, and only bought the new one because you wanted a bigger challenge. Some of the Tony Hawk games have the same problem, so it's probably those developers.

I can understand wanting a challenge, so I don't think there's anything wrong with it. But the problem manifests itself by having the difficulty curve all wonky. You can be very good at Easy, and still not be able to complete relatively simple songs on Medium. Same with Medium->Hard, and Hard->Expert. Rock Band, on the other hand, can also be pretty challenging, but the curve is more gradual, so IMO it works better. It's clear the developers were focused more on having the game be fun for all levels of expertise, rather than making a good challenge that only hardcore fans will appreciate.

I think this applies for pretty much all games, across genres. Guitar Hero was just what came to mind. Ideally anyone should be able to play, but it should be more *fun* to play harder difficulties if your better at the game.

Re:Maybe about the curve?

[geminidomino]

There is a practice mode to help you learn the fingering and sequences. perhaps you need to actually do some "Practice", or do you just expect that you can rock out on a harder mode because you are perfect at a simple mode?

That would be the wonky learning curve being referred to, yes. If you can get 99-100% on Freebird, Misirilou, and Psychobilly Freakout on Medium, but can't even beat "Mother" on Hard, what you have is a gap where the game rejects "difficulty" for "frustration", and if you're not "hardcore", you just don't want to play anymore.

You gotta have a carrot.

Re:Maybe about the curve?

[somersault]

it should be more *fun* to play harder difficulties if your better at the game.

I can tell you that it is indeed fun. If you can't complete the higher difficulty level then obviously you're not 'better at the game'. Being able to play through without ever failing or putting a bit of practice in is not challenging, and for me is therefor not 'fun' either. Try going through the songs more slowly in practice mode if you are finding any sections especially hard (it lets your muscle memory remember the patterns to play, and is especially good for getting used to switching between more akward chord shapes), and remember that the notes go along with the music. I think people on the lower difficulty levels try to watch when the notes pass by the bottom of the screen or something. I just look at the middle to top of the screen to see what is coming, and then play those notes or chords along to the music (though on crazy solos like in Cult of Personality there isn't really any 'music' to play along to, heh, I just try to hit as much as I can.. get around 65-85% of the crazy solos usually).

The harder difficulties are more 'natural' because they mimic the music almost exactly. I already played guitar before playing Guitar Hero so I had a bit of an advantage co-ordination wise, but I can complete a few songs on Hard even on lefty-flip. Playing lefty flip helped me to understand why some people find the co-ordination difficult (and is also the only way to make the game more challenging again now that I have completed all the set list and bonus songs apart from Through the Fire and Flames and the final battle on Expert) :p

Civ IV

[ShakaUVM]

On the contrary, I hate how Civ IV does its difficulty settings.

It's normal and slightly above normal difficulty settings are far too easy. Immortal (the highest setting) is simply designed to cripple you as much as possible while giving the AI bonus cities and resources. The medium-high difficulty settings (which is what I usually play at) are usually pretty balanced between them and me, but the kicker is that the difficulty isn't precisely harder than normal, the game just gives the AI 5 times the units it would normally have. So when you have machine gunners and riflemen gunning down their knights and longbowmen (since it doesn't actually play any smarter), it just takes 5x as long to beat the game, and it just ends up feeling like an eternal slogging march, not fun at all. Personally, I think the approach is just stupid.

Re:Civ IV

[ACS Solver]

Give me a better option. I worked on Civ4 and the expansions. And I think that generally, the right approach to difficulty was taken.

The AI is only going to be this smart. Fact is, most players actually play at below-average difficulties. But what makes Civ hard? Your competition, the AI ultimately. We all want a better and stronger AI but there are limits to what you can do. Specifically, it's never going to be as effective in using its units as a competent human player. Therefore, the AI needs more units to be competitive against a human.

For the Beyond the Sword expansion, one of the best Civ4 players out there had a contract to work on the AI. It was improved dramatically and, interestingly, the bonuses it gets on higher difficulties were actually decreased compared to the original game. Still, despite some great programming and LOTS of playtester attention to the AI, it's obviously not as smart as human players. So yes, higher difficulties have to give the AI some bonuses to compensate. Immortal (it's the 2nd highest) and Deity are designed for a very small minority of players who just need a challenge against all odds. It's not a level meant to be fun for everyone, just like Settler is a level specifically designed for people who don't know what they're doing at all.

The only other approach would be to make the AI behaviour smarter at high levels. But if you write a smarter AI algorithm, why leave it enabled only for the higher levels? Let it be everywhere. I strongly opposed the notion of having silly AI behaviour on lower levels. It can bite you in the ass. Higher levels may be warranted in being more aggressive, but not smarter. Because if you make higher levels smarter than you're at the same time denying lower levels these smart algorithms, which is a bad thing.

As a side note, BtS has a revamped "Aggressive AI" setting which is more like "ruthless AI". It's not plain-out aggression, it just plays a more hardcore game and expects a more ruthless human opponent.

Re:Civ IV

[ShakaUVM]

That's an interesting question. I actually sketched out a fairly detailed document on how I'd write a Civ AI and downloaded the API for it, but couldn't make enough sense of the code to start hacking on it. I do have a background in writing game AI - I wrote a bot for Quake, and have modded quite a few games (see my URL for the biggest one), but I couldn't get to the point where I grokked the code. I know it's not much of an excuse, I obviously could have spent more time on it, and spent more time digging up docs on it, but there it is.

Essentially, the problems I see with the Civ 4 AI are this (in no particular order):
1) Too easy on noble (and lower). As in, without doing anything particularly interesting, you end up outteching the AI by progressively larger margins, and they don't have much of an army to stop you. At some point, you can just roll in with a small stack of gunpowder units and wipe out all the enemy civs.
2) Surrendering to Switzerland. A number of times I've been beating the snot out of a civ, and offered it vassalage, which it would refuse. I'd beat on it some more, then it would surrender to a random third party that neither of us are at war with.
3) On Monarch and higher, the game makes these obscene superstacks of units for the AI. As in, there'll be 30 or 40 tanks or knights or whatever on one city, and 10 to 20 units on all the other cities. If a human is at the point where it can kill such a stack, he's going to win eventually, but it requires an amazingly tedious amount of time to do so, and the AI appears to be able to pull massive amounts of military units out of its ass, and apparently without paying upkeep. Or if it is paying upkeep, then it's certainly a bug, since it'll get even more out-teched since it can't afford research.
4) The AI's lacking in basic tactics sometimes. It'll suicide entire stacks of 20 units against a trio of fortified machine gunners, won't use terrain intelligently (well, some of the time it will), doesn't use spies to take out critical resources (like a lone copper or horse resource, instead attacking horses in the tank era or a resource that I already have 3 of).
5) It's careless with its workers, allowing them to get captured easily. I know this was patched in the latest version, but the last game I played I captured a lot of enemy workers just tooling around. They're especially trusting before war breaks out.
6) It's general method of moving troops and ships around is just odd sometimes. I've seen enemy units get stuck in a mountain, trying to pathfind across it, or individual units approaching my stack when I'm at war with them.
7) The AI, in general, is completely reactive. If I set my spy rate up higher, the AI will set its spy rate up higher. If I turn up my culture rate, he turns up his culture rate.
8) Too trusting. If I'm at peace with an AI, but building up a large force along the border, it will mostly ignore it until I invade. I'd like to see it take up defensive positions with spare units and build forts in border tiles (especially in chokepoints) instead of being constantly surprised every time someone declares war on them. Likewise, a lot of time they'll declare war without their armies being in position for it.

What I'd like to see is this:
1) High level AI tasks. Have AIs decide perhaps 30 turns in advance that they're going to betray a peace treaty and start cranking out units and positioning them on a border, ready to invade. If relations haven't improved on the target date, rush in with everything. Alternatively, have it decide to make a solid effort to take over the new world, instead of the piecemeal way that it expands to new continents now. The different AI types (techers, expansionists, militarists) would have different likelihoods for the various tasks. Essentially this would make them appear to be more human, and more interesting to work with.
2) Set them up to beeline different techs and wonders to match a specific objective chosen at the beginning of the game. For example, an AI could go with

Spilt cpu intelligent and cpu handy cap / cheating

[Joe The Dragon]

Spilt cpu intelligent and cpu handy cap / cheating in to there own settings. Do not put them under the same setting. Heroes of might and magic 1 had that.

Using metrics from gameplay to tweak difficulty

[drexlor]

I wrote a paper in college about how video games could evaluate and adjust difficulty based on metrics while the gamer is playing. I think a game's difficulty that is based off input from the controller and game statistics would help people have a more enjoyable gaming experience.

If the game can receive input from the gamepad/joystick they can measure heat, motion, button hitting frequencies, and things of that nature.

Software inputs can be used too. Time measured in zones, level completion times, and time to defeat creatures can be measured to add as heuristics. Death counts and locations can be used to determine what areas need work.

These inputs have been associated with stress levels in gaming and can be used to adjust creature abilities, time limits, weapon power, and directional support for the gamer. If the gamer is playing well the difficulty will become more difficult over time and if the player is having trouble then the difficulty can be toned down slowly and selectively. Directional help can also be used if the game thinks the gamer is lost.

These could help create a more dynamic game that fits to the gamer.

Big problem in strategy games

[ElMiguel]

I'm a fan of turn-based strategy games such as Civilization, and yet I usually stop playing most of them after a while because I get angry at the way hard difficulty levels are implemented.

You see, the developers of these games apparently find it too difficult to implement an AI that plays by the same rules as human players and yet provides a good challenge. So AIs cheat. Cheats come in two flavours: information cheats (e.g. send an unprotected valuable unit and you'll see an enemy fighter, who in theory has no way of knowing about your unit, beeline for it) and stats cheats (e.g. the AI produces units 40% faster than you).

I call those special rules "cheats" because they are typically not documented or consistent with the game story. So you end up making blind guesses about what rules the AI is playing by in a very atmosphere-shattering way and trying to adapt to them. It really feels like cheating and drains my interest in otherwise excellent games pretty fast.

FPS AI

[PaganRitual]

Has there ever been a first person shooter for where the difficulty level isn't simply an indication of how close you want your enemies to aimbots?

I can easily think of the negative extremes, such as Soldier of Fortune 2, with the infamous jungle level, and my own personal experiences with being shot through thick jungle, repeatedly having grenades land perfectly at my feet from enemies that haven't even seen me yet, and training a sniper scope onto the back of an enemies head at maximum range, only to have him suddenly turn around and go sneaking towards me. And Call of Duty, where increased difficulty was just the games way of asking exactly what percentage of the entire opposition army you wanted to face at once.

I appreciate that Halo tends to take a beating every time it's put forward as a paragon of game design and gaming in general and that the Halo fanboys with mod points will destroy me, but seriously, one of the gripes is with the claim that Halo's AI is somehow fitting of the Legendary title on it's hardest difficulty, when anyone playing the game sans rose-colored glasses has trouble not noticing that the enemies now simply fire faster, harder and lead perfectly.

"It's not like we just cranked every enemy's health by 200% and called it Legendary," he said. "There was a good amount of custom changes made per mission as well ...[snip]... What did make a difference was the time spent tweaking and fixing issues to make the game fun on every difficulty level. All titles had more bad guys, stronger and more accurate enemies with faster projectiles. And they used similar numbers for each of these parameters."

So in other words they made the enemies fire faster, harder and lead better. Thanks for clearing that up.

On the plus side, Far Cry's AI was reasonable, but had noticable holes, such as when it somehow thought it was hidden yet you could clearly see it sneaking towards you down an open dirt track, seemingly with an "If I can't see him then he can't see me" attitude. And if I remember correctly it improved further in Crysis. Although the thing that seems to happen with games like that is as soon as you claim that the AI is brilliant any single example of the AI not working flawlessly has people uploading videos to youtube showing that the AI is completely garbage because of this one time it got stuck on that shark outside the hut or something.

Are there any really decent examples of FPS AI or do we have to still be happy with running the Reaper Bot in Quake?