Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Cruxador]

It could be.  It depends on how it works.  Random could mean an evolutionary system where enemies develop abilities randomly and propagate the ones that get results, or it could just be each unit has a chance to develop an adaptation to your tactics, so it's not automatically becoming immune to everything you do.

He consistently referred to it as though it's a per-unit thing. I thought he was pretty clear on the system: If a unit is successful, keep it, if it demonstrates low performance (he didn't say how they quantify this) then that basic unit chassis gets a random mutation – it swaps one or more body parts out with a replacement. Although that body part is chosen at random, it presumably has some kind of weighting and possibly limitations based on the flow of the game, but even with no weighting a system like this would allow natural selection to over time create enemies more effective against your strategy.

It'll be a tough balance.  The important part is making it so you have to vary your tactics within a match, rather than making it an all-or-nothing thing where you have to change your whole setup each time.

I don't agree with the notion that all meaningful gameplay should be at the tactical level, at the expense of the strategic. The tactical gameplay doesn't seem that dull to me that you need to mix things up within the (somewhat arbitrary, with regards to gameplay loop) bounds that are the beginning and end of a single skirmish.

Mist sentinels are huge immobile mushroom things, I have no idea what they do

Spread mist, from the sound of it. I don't know what that means since it wan't in the previous backer build and I've not had time to peruse gameplay footage since the new one dropped, but it was described as a "literal fog of war", so I guess it's like smoke grenades and probably other enemies will have the potential to utilize it for special abilities if they roll the right body parts. Edit: Saw someone say that the mist not only occludes your vision but also, everything that it touches is visible to the enemy regardless of line of sight.

[Cthulhu]

That's not what I mean.  I mean I want to see something where you have to build your squad for different threats in the tactical game, not one where you're switching from one one-trick layout to another as the enemies become immune to whatever your last setup was.  Which it seems it won't be.

[Reelya]

The AMA definitely implied it'll be changing only one enemy unit at a time, and that there would be a clear break with the old unit type, which would be retired before you see the next one. Since units are phased out one at a time, there will be continuity to the system.

If you always use flamethrowers, then one by one, the old enemies will be replaced by ones which handle the flamethrowers better, but they won't all magically gain flame-resistance at once.  So, if no enemy is good against flamethrowers then at first they'll be very effective, but gradually less-so.

Also, because the game looks at performance-data then it won't necessarily have hard-coded ideas in it like "if player uses flame weapons, add flame resistance", because adding flame resistance isn't necessarily the best use of limited resources for dealing with flamethrower attacks. snipers would be more cost-effective, or leaping melee units. The enemy that's the most-useless against your flamethrowers will be phased out first, meaning that you'll still be able to use the flamethrowers against some enemies, but you'll have to be planning for new enemies to appear, at the same time.

As for metrics, I'd imagine that each unit class would have a different mission role, and thus have different metrics. You definitely don't want something naive like keeping all units that survived a long time. Cheap meat-shields and/or tanks should be there to draw fire, so even if all your "zombie" units died during a battle, that might be a great success, because they drew fire away from your heavy-dps wizard-type creature. So, for shield/tank units, the percentage of player fire that they drew during the battle should be one of the metrics.

For the metrics, hand-crafting them seems like it would be pretty iffy. If I was doing it, I'd set the AI to battling other AIs, collecting data about each unit's performance (this would include all actions taken by the unit, and also against the unit), then feeding that into a *randomized* set of metrics. Different AIs then optimize their armies based on their random metrics, and the dumbest AIs get weeded out and the best AIs get mutated. Thus, you can evolve AIs which are better at working out how to evolve their armies.

[Retropunch]

As for metrics, I'd imagine that each unit class would have a different mission role, and thus have different metrics. You definitely don't want something naive like keeping all units that survived a long time. Cheap meat-shields and/or tanks should be there to draw fire, so even if all your "zombie" units died during a battle, that might be a great success, because they drew fire away from your heavy-dps wizard-type creature. So, for shield/tank units, the percentage of player fire that they drew during the battle should be one of the metrics.

For the metrics, hand-crafting them seems like it would be pretty iffy. If I was doing it, I'd set the AI to battling other AIs, collecting data about each unit's performance (this would include all actions taken by the unit, and also against the unit), then feeding that into a *randomized* set of metrics. Different AIs then optimize their armies based on their random metrics, and the dumbest AIs get weeded out and the best AIs get mutated. Thus, you can evolve AIs which are better at working out how to evolve their armies.

I'd love it to be that performance based, but I'm sceptical about how they'd pull it off. It's less that they don't have the skill (I'm sure they do) it's more that I believe it'd be extremely tricky to balance.

I had a friend that worked for a big FPS games company back in the the mid-00s, they helped developed some incredibly 'realistic' AI for an FPS - it could work as a team, adapt to your strategies and flank almost perfectly. The issue was that it was super, super frustrating for players and so they had to scrap all the hard work they did and make an incredibly cut down version of it. It didn't matter that it was cool, it got super annoying to be facing someone that could adapt to your strategies well - part of what we like in games is winning (or at least finding out a working strategy as in DF) and if it was constantly destroyed by an AI it becomes no fun.

All that to say, I imagine that it'd be a better option for them to just go 'if player has x> snipers: give some enemies long range cloaks' - you'll probably end up with the same result as a paired down survival-of-the-fittest based AI without the hassle of building it.

[umiman]

As for metrics, I'd imagine that each unit class would have a different mission role, and thus have different metrics. You definitely don't want something naive like keeping all units that survived a long time. Cheap meat-shields and/or tanks should be there to draw fire, so even if all your "zombie" units died during a battle, that might be a great success, because they drew fire away from your heavy-dps wizard-type creature. So, for shield/tank units, the percentage of player fire that they drew during the battle should be one of the metrics.

For the metrics, hand-crafting them seems like it would be pretty iffy. If I was doing it, I'd set the AI to battling other AIs, collecting data about each unit's performance (this would include all actions taken by the unit, and also against the unit), then feeding that into a *randomized* set of metrics. Different AIs then optimize their armies based on their random metrics, and the dumbest AIs get weeded out and the best AIs get mutated. Thus, you can evolve AIs which are better at working out how to evolve their armies.

I'd love it to be that performance based, but I'm sceptical about how they'd pull it off. It's less that they don't have the skill (I'm sure they do) it's more that I believe it'd be extremely tricky to balance.

I had a friend that worked for a big FPS games company back in the the mid-00s, they helped developed some incredibly 'realistic' AI for an FPS - it could work as a team, adapt to your strategies and flank almost perfectly. The issue was that it was super, super frustrating for players and so they had to scrap all the hard work they did and make an incredibly cut down version of it. It didn't matter that it was cool, it got super annoying to be facing someone that could adapt to your strategies well - part of what we like in games is winning (or at least finding out a working strategy as in DF) and if it was constantly destroyed by an AI it becomes no fun.

All that to say, I imagine that it'd be a better option for them to just go 'if player has x> snipers: give some enemies long range cloaks' - you'll probably end up with the same result as a paired down survival-of-the-fittest based AI without the hassle of building it.

Yeah, I definitely agree with this.

It's what bugs me a lot when people complain about the AI in strategy games. I understand it's easy to riff on them being stupid, but they as the player seem very unaware that they themselves are super dumb.

All it takes is one single MP match to humble them into the ground or make them ragequit forever.

They complain that they want better AI but I'd bet infinity money that these same dumbasses wouldn't want to play a game with that same better AI because they are garbage.

I mean, look at those Dota 2 bots. Given time no human can even remotely hope to challenge that. And even if it was nerfed to 1/5th the power most humans couldn't handle it.

To program an AI that can balance between the idiots and the geniuses as some tier of god-level programming.

Incidentally, in my personal taste I actually disable these AI learning features in games. I did it in Metal Gear Solid 5 for example, where the AI slowly ramps up their stuff based on what you do. I just turned it off because it's super annoying when every enemy has night vision and body armor and heightened senses and decoys don't work, etc. etc.

[Kagus]

Y'all want some fun AI? Find yourselves a copy of Battlefield 1942 and play a match against bots. Before Battlefield 2, Dice hadn't quite learned the intricacies of hamstringing an AI's accuracy... As such, every AI sniper is quite literally an aimbot and will completely dominate the entire map until another AI sniper shows up at counter-snipes him.

The response was apparently strong enough that for Battlefield 2, the AI sniper was nerfed to the point where they couldn't hit the broad side of a barn if the barn had a target painted on it and was black.


As for RTS games though, AIs aren't actually that good at playing them in a general sense. They obviously cannot be beaten in micro, but strategy and maneuvering are not really their forte. Similarly for the DOTA 2 bots, while they have completely unmatched reaction times and preternatural awareness of health levels and stun timers, they're also hysterically predictable and uncreative.

What makes these AIs stand out is how well the general gameplay structure happens to support their strengths, or how much weight it instead puts on the aspects that they're not as good at. There's definitely a lot of "toning down", but there are also some intelligent behaviors that are remarkably difficult to implement.

[Cruxador]

Yeah, I definitely agree with this.

It's what bugs me a lot when people complain about the AI in strategy games. I understand it's easy to riff on them being stupid, but they as the player seem very unaware that they themselves are super dumb.

All it takes is one single MP match to humble them into the ground or make them ragequit forever.

They complain that they want better AI but I'd bet infinity money that these same dumbasses wouldn't want to play a game with that same better AI because they are garbage.

I mean, look at those Dota 2 bots. Given time no human can even remotely hope to challenge that. And even if it was nerfed to 1/5th the power most humans couldn't handle it.

To program an AI that can balance between the idiots and the geniuses as some tier of god-level programming.

Incidentally, in my personal taste I actually disable these AI learning features in games. I did it in Metal Gear Solid 5 for example, where the AI slowly ramps up their stuff based on what you do. I just turned it off because it's super annoying when every enemy has night vision and body armor and heightened senses and decoys don't work, etc. etc.

I don't think the idea that people who want better AI are dumb necessarily follows from this particular anecdote. Especially when you bring in comparisons of multiplayer and single player, you're making fallacious assumptions. There's no particular reason to suppose that people who think it should be smarter only play single player, first of all, and even if someone is bad at multiplayer (which I doubt has any strong causal relationship with opinions on AI) that doesn't invalidate their opinions on AI whether in a singleplayer or multiplayer context. Besides, most games have a lot of space in between the status quo and an actually good AI.

[umiman]

Yeah, I definitely agree with this.

It's what bugs me a lot when people complain about the AI in strategy games. I understand it's easy to riff on them being stupid, but they as the player seem very unaware that they themselves are super dumb.

All it takes is one single MP match to humble them into the ground or make them ragequit forever.

They complain that they want better AI but I'd bet infinity money that these same dumbasses wouldn't want to play a game with that same better AI because they are garbage.

I mean, look at those Dota 2 bots. Given time no human can even remotely hope to challenge that. And even if it was nerfed to 1/5th the power most humans couldn't handle it.

To program an AI that can balance between the idiots and the geniuses as some tier of god-level programming.

Incidentally, in my personal taste I actually disable these AI learning features in games. I did it in Metal Gear Solid 5 for example, where the AI slowly ramps up their stuff based on what you do. I just turned it off because it's super annoying when every enemy has night vision and body armor and heightened senses and decoys don't work, etc. etc.

I don't think the idea that people who want better AI are dumb necessarily follows from this particular anecdote. Especially when you bring in comparisons of multiplayer and single player, you're making fallacious assumptions. There's no particular reason to suppose that people who think it should be smarter only play single player, first of all, and even if someone is bad at multiplayer (which I doubt has any strong causal relationship with opinions on AI) that doesn't invalidate their opinions on AI whether in a singleplayer or multiplayer context. Besides, most games have a lot of space in between the status quo and an actually good AI.

I can also see why there might be confusion.

We're talking about adaptive "realistic" AI opponents here. AI that predict and change according to your actions as a player. The context of the conversation thus far has been how that usually ends up being bad despite it appearing good.

We're not talking about non-functioning AI like Stellaris or Civ 6 because yeah, that definitely needs to be improved. I'm not sure they even qualify as AI in their current state.

[Cruxador]

Ah, yeah. I thought you were making a point about AI in general, more inspired by the discussion of adaptive AI than contained entirely within it. Civ VI did come specifically to mind as the Civilization series as a case where some people think they're good while people who play online multiplayer would consider them scrubs. I think any good AI will adapt to the player to some extent, but that becomes a matter of semantics.

[Kagus]

That is a remarkably different post from its original draft.

[Retropunch]

Ah, yeah. I thought you were making a point about AI in general, more inspired by the discussion of adaptive AI than contained entirely within it. Civ VI did come specifically to mind as the Civilization series as a case where some people think they're good while people who play online multiplayer would consider them scrubs. I think any good AI will adapt to the player to some extent, but that becomes a matter of semantics.

I think there's an important thing to pull out here - a lot of the times (especially for things like 4x games) the AI in single player isn't geared towards being challenging, it's more geared towards letting the player experience more stuff. So it'll be different by design from multiplayer, where you're aiming to win instead of let the other player experience lots of stuff and provide a bit of a challenge.

One of the things which has always bugged me is how limited interaction is with AI in 4x (and similar) games - most of the time it's just a set of options that pushes a modifier up or down (give gift, make alliance, go to war). I always think they could do so, so much more with diplomacy and getting the AI to act like a cohesive person rather than just 'if x do y'

[Egan_BW]

IMO, in a 4X game the opposing AI should not act like a player, but like a nation.

The AI in phoenix point should not act like a player trying to win the match, it should act like aliens, trying to... do whatever it is aliens want to accomplish. Eat tasty humans and not get wiped out? Sure, that sounds good.

[Retropunch]

IMO, in a 4X game the opposing AI should not act like a player, but like a nation.

The AI in phoenix point should not act like a player trying to win the match, it should act like aliens, trying to... do whatever it is aliens want to accomplish. Eat tasty humans and not get wiped out? Sure, that sounds good.

Yeah, person was the wrong word, I sort of meant that I wish AI had more of a cohesive 'drive' behind it - so often they seem to lack any sort of personality.

[EnigmaticHat]

Personally I think making the AI play a symmetrical game to the player is a mistake.  Most of the things humans find fun (most things in general really) AI is either really good at, or really bad at.  For example positional play; there are ways to fake it, there are ways to brute force it.  But for an arbitrary play area with a limited "budget" of processor time, there's no known way for video game AI opponents to match a human player.  Likewise anything involving input precision (fighting games, aiming in shooters) is going to feel janky because the AI doesn't actually have buttons they need to press.  Simple shit like how you can sidestrafe in Overwatch to throw off people's aim is almost impossible to convincingly simulate.

Compare to say Left 4 Dead, where reviewers have mentioned that the special infected AI is "smart"* even tho Valve still hasn't invented an AI that can navigate a level on its own.  The special infected can be smart because they aren't playing the same game as the player, the survivor AI will always seem stupid and aimbotty because its trying to imitate a human.

*although I'm not sure experienced players would agree with them

[Kagus]

I remember the days of playing the L4D2 demo version with a few of the other forumites.

I remember using the console to spawn in approximately 20-30 survivor bots, all of them Coach.

I remember the mayhem.