Support NeoGAF

[IbizaPocholo]

Next-gen could help push “AI and physics” forward, says Arkane

PS5 and Xbox Series X are on the horizon - what could that mean for games that build their own simulated worlds?

www.vg247.com

PS5 and Xbox Series X are on the horizon – what could that mean for games that build their own simulated worlds?

That’s a question we asked Arkane Lyon when we got a chance to speak with the studio for its 20th anniversary.

“Visual fidelity is always nice, but, you know, well. If you ask me, I think Dishonored 2 and Outsider look splendid, so if I can have that running at 60fps in 4K on my console, I would be already pretty happy,” game director Dinga Bakaba said.

“I think everything that has been mentioned about fast loading is of course exciting. Minimising the downtimes during a playthrough, that’s really nice. If you’re going for something like an open world, streaming would be really nice, if it’s as fast as what we’ve seen.”

Current generation games have a bottleneck when it comes to memory, so developers have to make sacrifices to keep games under that limit. Next-gen will free developers up to use more memory, potentially opening the door for advances in other areas outside of graphical fidelity.

“AI, physics simulation, there are a lot of things you can do when you’re developing a game for those platforms,” Bakaba explained. “There will of course be a third-generation, like with all generations of games, there is a first generation that starts to touch the technology. But when we are well into this generation, it will be pretty exciting for us. AI, physics, all those things that definitely reinforce the interactivity of our games a lot.

“We are excited to spend it, rather than – well, I was going to say, ‘rather than split our hair about how to make it fit the memory,’ but I think even with more memory, we’ll still try to get so much out of it that we will still split our hair. Don’t worry. You [will still] have three bald guys in the room. It will not change,” he joked. “But, yes. I think that’s really exciting. More agents in the world, more simulation, and for Sebastien [Mitton, art director] and his teams, also the ability to do a lot of things we weren’t able to. And that’s already starting now, in a way.”

 

[hyperbertha]

Doubt it. FEAR has the best AI in games to date so if nothing has improved since then I think its a developer issue. Too lazy/stupid to code?

Why can't they use machine learning to improve AI? I've always thought machine learning would be a great way to improve AI in games. Can anyone say why this isn't happening yet?

 

[NickFire]

Doubt it. FEAR has the best AI in games to date so if nothing has improved since then I think its a developer issue. Too lazy/stupid to code?

Why can't they use machine learning to improve AI? I've always thought machine learning would be a great way to improve AI in games. Can anyone say why this isn't happening yet?

I don't think its something you should doubt. My rudimentary understanding is the SSD let's them refresh ram more quickly and avoid filling it with less useful assets all the time. So they can optimize the code sent to the CPU better than ever, meaning more code for things like physics. I might be wrong, but if I'm not it seems all but certain we get better ai and physics from any dev who wants to make it.

 

[LordKasual]

Doubt it. FEAR has the best AI in games to date so if nothing has improved since then I think its a developer issue. Too lazy/stupid to code?

Why can't they use machine learning to improve AI? I've always thought machine learning would be a great way to improve AI in games. Can anyone say why this isn't happening yet?

Because making computers "smart enough" to beat players is exceedingly easy and is honestly counterproductive, because humans do not like losing and they certainly don't like being outsmarted.

Just look at fighting games. Seasoned players outsmarting newbies is immediately perceived as "cheating" or "exploiting" to people who are smart enough to play the game, but too inexperienced to realize why they're actually losing. It's only when they understand what's actually happening and how it's their fault, that they begin to blame themselves, and even still sometimes that isn't enough.


An AI that genuinely outsmarts players would simply be written off as "cheap" or "fake difficulty" to the vast majority of players. They wouldn't understand why they're losing (because they're being outsmarted) and thus there would be no difference to them.

A group of AI Soldiers that collaborate and carry out tactics at the level of actual humans would simply NOT be fun to play against for most people. It'd likely feel no different than getting stomped out in PvP.


And deeper than that......since we aren't talking about humans outsmarting humans, there will always be an uncertainty about whether or not the AI is playing at an above-human level, or needs to be "toned down", or whatever.

The solution is making them capable enough to be difficult/believable but dumb enough to still be outsmarted / exploited by players. And you do not need "good AI" to do this.

 

[hyperbertha]

Because making computers "smart enough" to beat players is exceedingly easy and is honestly counterproductive, because humans do not like losing and they certainly don't like being outsmarted.

Just look at fighting games. Seasoned players outsmarting newbies is immediately perceived as "cheating" or "exploiting" to people who are smart enough to play the game, but too inexperienced to realize why they're actually losing. It's only when they understand what's actually happening and how it's their fault, that they begin to blame themselves, and even still sometimes that isn't enough.

An AI that genuinely outsmarts players would simply be written off as "cheap" or "fake difficulty" to the vast majority of players.


And deeper than that......since we aren't talking about humans outsmarting humans, there will always be an uncertainty about whether or not the AI is playing at an above-human level, or needs to be "toned down", or whatever.

The solution is making them capable enough to be difficult/believable but dumb enough to still be outsmarted by players. And you do not need "good AI" to do this.

That's not what I mean at all. There are genuine ways to improve AI's in ways that let them outsmart players and yet not be frustrating or unbeatable. I'm talking about making them behave exactly like humans rather than making all choices with computer precision as to be unbeatable. Difficulty in games theses days is mainly through numbers/hp increase, but I'd like that to be replaced by smarter AI. If that means having to let go of easier difficulties, I have no issues with it at all.

 

[M1chl]

Finally someone speaking some sense.

 

[LordKasual]

That's not what I mean at all. There are genuine ways to improve AI's in ways that let them outsmart players and yet not be frustrating or unbeatable. I'm talking about making them behave exactly like humans rather than making all choices with computer precision as to be unbeatable. Difficulty in games theses days is mainly through numbers/hp increase, but I'd like that to be replaced by smarter AI. If that means having to let go of easier difficulties, I have no issues with it at all.

What i'm suggesting is that the very act of "being outsmarted", "read", "predicted", ect is frustrating for the majorty of players, casual or otherwise.

In fighting games, it means the AI not letting predictible players hit them and shutting down options before they even get a chance to try them because they know its coming.

In shooters, it means stepping out from behind cover and being sniped, or grenades/bullets/rockets being lobbed in locations that seem predictable for the AI (where you're headed) instead of predictable for the player (their current location).


If AI starts playing like Humans.....then AI is absolutely going to become annoying.

The way humans stomp out players in PvP is a completely different kind of stomping from the way current AI is designed to stomp out players.


Just imagine a fighting a giant boss in Bloodborne/Souls that starts memorizing your evasion habits and then just adjusts accordingly, resulting in an instant kill.

Would it be cool? Absolutely, but it would also completely change the way the game is designed. The only reason those games aren't unfair is because you have the benefit of being able to memorize/outsmart/exploit the static AI programming.

If enemies stop getting cheesed by collision exploits, pattern loops, throwing out unsafe and telegraphed attacks.........it's a completely different game LOL

 

[hyperbertha]

What i'm suggesting is that the very act of "being outsmarted", "read", "predicted", ect is frustrating for the majorty of players, casual or otherwise.

In fighting games, it means the AI not letting predictible players hit them and shutting down options before they even get a chance to try them because they know its coming.

In shooters, it means stepping out from behind cover and being sniped, or grenades/bullets/rockets being lobbed in locations that seem predictable for the AI (where you're headed) instead of predictable for the player (their current location).


If AI starts playing like Humans.....then AI is absolutely going to become annoying.

The way humans stomp out players in PvP is a completely different kind of stomping from the way current AI is designed to stomp out players.


Just imagine a fighting a giant boss in Bloodborne/Souls that starts memorizing your evasion habits and then just adjusts accordingly, resulting in an instant kill.

Would it be cool? Absolutely, but it would also completely change the way the game is designed. The only reason those games aren't unfair is because you have the benefit of being able to memorize/outsmart/exploit the static AI programming.

If enemies stop getting cheesed by collision exploits, pattern loops, throwing out unsafe and telegraphed attacks.........it's a completely different game LOL

I'm not talking about machine learning being implemented in the final game. Just using it to improve the AI to a good level during development and playtesting and then shipping the game with a fine tuned AI that provides a satisfactory experience. People getting 'frustrated' at challenging enemies is as dumb as games journalists calling for ez mode in all games and shouldn't be taken seriously. Just create a beatable, yet perfectly realistic human AI, and npcs that behave like real people and react according to things that you do logically.

 

[JareBear: Remastered]

I don't want to sound like a dick, especially because Arkane is awesome, but isn't their statement kinda obvious?

Eh, regardless, AI seemed to take a step BACK this gen, so any improvements on that front are more than welcome to me. AI has been neglected for far too long now, imo

 

[LordKasual]

I'm not talking about machine learning being implemented in the final game. Just using it to improve the AI to a good level during development and playtesting and then shipping the game with a fine tuned AI that provides a satisfactory experience. People getting 'frustrated' at challenging enemies is as dumb as games journalists calling for ez mode in all games and shouldn't be taken seriously. Just create a beatable, yet perfectly realistic human AI, and npcs that behave like real people and react according to things that you do logically.

Yeah, i mean either approach could technically work. There's no rule that says decent AI needs to make decisions on the level of a skilled human.

But i'm just providing an answer for the reason that it hasn't already happened -- it's an avenue that isn't worth pursuing, at least from the perspective of development costs.

FEAR absolutely pushed the envelope, but then again, Call of Duty sells millions and that AI is dumb as fucking bricks.


Modern games have bypassed that issue entirely. Apex Legends and Overwatch didn't even have AI enemies on release, it's purely just PvP. And they're massively successful.


But don't get me wrong bro, i 100% agree with you. AI in older games like FEAR or Halo: CE were glorious.

 

[ZywyPL]

Those are the two aspects I'm mostly looking forward next-gen, games took a huuuge step backward because of Jaguar, but with 16 Zen2 threads at devs disposal, things should look really good in the upcoming years.

 

[-YFC-]

Thank god, AI and physics have been shitty these past 2 generations. Especially AI. I just recently played the old Unreal, and jesus christ the AI in that game is insane. And this is a game from 1998. How is it possible that rarely any modern game has come close in terms of AI to Unreal, Unreal Tournament and FEAR?
So if this info is true, then yeah, thank god for that. It's about time these aspects improve in videogames.