how would you feel if AI generated levels reached human-level quality?

[jazzmaster9]

I would judge it the same way i would any other map. If its fun to play and navigate then I'll like it and will replay it in then future  it its not then i wont like it and will move on to the next mapset.

 

There was an idea for a chalice dungeon type Game mode for Abysm 2 where AI generated maps would have been a godsend.

Edited September 10, 2022 by jazzmaster9

[Solmyr]

22 minutes ago, Captain POLAND said:

What if they started randomly including Terry WADs in every batch of WADs they designed?

Then people who dislikes Terry WADs would start opening up PWADs and Megawads in Slade looking for suspicious looking map layouts, and read ACS, ZSCRIPT and DECORATE entries to make sure there are no "terry-like" actors and scripts. Other folks would just stick to AI mapping content that only targets boom, mbf 21and limit removing formats.

[dmnwizard]

I would like to play AI-made wads. In my opinion, this could be a new category of Doom content, curation of  AI-made wads. And I also appreciate the idea of an AI creating for me wads based on what I like to see on a map, what to face in terms of challenges, antagonists, protagonists, plots, etc.

[Kyka]

12 hours ago, Deadwing said:

TBH, I wouldn't be able to distinguish Oblige-generated maps with "humans" ones, unless I played a lot of them to find the patterns. Some of my older maps were even described as oblige-like (not in a negative way, also, afair).

 

Funnily though, I do enjoy more playing (mostly) one-man megawads due to how the mapper handles these patterns and find new creative solutions to them, creating a large sense of "narrative", if that even makes sense 

 

 

Totally agree. I tend to lean towards single mapper megawads, or also megawads where there was a shared vision among several mappers. (Eviternity comes to mind.)

 

I love my dooming with supersized narrative woven through it.

[QuaketallicA]

Probably the same way I feel about a.i.-generated MUSIC that feels like human-level quality. 

 

At first very impressed, mind-blown.

 

Then terrified by the idea that my genuine emotions (expressed through music) or my genuine creativity in level design looks no different or no better than something an a.i. came up with just by mimicking traits of other designers (that's what neural networks do I believe is keep guessing randomly and self-correcting till they get closer and closer to what a huge data set of stuff done by humans says is right, eventually finding a sweet spot of maximum closeness). Like playing a game of "hot and cold" with mathematical functions and algorithms.

 

On the one hand, you have to appreciate beautiful art. You can't deny its beauty. On the other hand though, I equally detest it for devaluing and competing with real art done by real humans. Especially when money is concerned, because it just means even if the art (levels, music, etc.) is only half-decent, but way cheaper to license access to software than pay an artist, then guess what most businesses are going to do. That's why I think most movies today are so reliant on CGI. And I'm not saying there isn't a lot of high-quality CGI out there...it's just, if I wanted to see CGI cutscenes, I may as well just play a video game. I prefer real costumes and models; practical effects and on-location shooting. Things grounded in real places just feels more real. Anyway I digress. 

 

Good for those who can't make high-quality stuff themselves and want to pay a small price for it (or even free if it goes open source). Bad for the people who used to get paid to do it. Same old story with technology replacing workers, I suppose. Nothing new there.

[QuaketallicA]

19 hours ago, Sena said:

 

At risk of derailing the thread, this comment seems to suggest that the only reason people indulge in creativity is because they want to make the best work to have ever been created, when that's really not true. Some people might like to strive for that, but humanity has been around long enough that if this were the case, there really wouldn't be any point in doing anything already - why should I learn how to paint when I know I'll never be as talented as da Vinci?

 

I was just feeling that way earlier today about playing guitar. I've been playing over 2 years now, and I like to think I'm pretty good. I have a number of songs/riffs I've written that I'm proud of; I can play the Metallica songs that inspired me to play in the first place (and seemed impossible-tier when I first heard them) with relative ease. I'm even getting decent at learning Eruption. But then I go and watch Paul Gilbert or Yngwie or another shredder online who can cover something one of those guys made, and it's just like I'm a nobody all over again. But I remind myself that 1) the biggest factor in how good you are is how experienced you are--how long you've been doing it which isn't something I can control, except to keep going and 2) it also doesn't matter as much how technical you are. So many of the best songs and most iconic bands are at a way lower level of technicality than the super-shredders...and guess what, those guys are more famous and I personally like them more anyway. So just do what feels good. But then at the same time there's always a part of me that wants to be the best and is always working to get there little by little, so seeing me blown away by the "competition" is a bit of an ego hit, but it's only realistic.

 

"No matter how good you are at something, there's always about a million other people who are better at it than you."

-Homer Simpson

[dmnwizard]

2 hours ago, QuaketallicA said:

 

I was just feeling that way earlier today about playing guitar. I've been playing over 2 years now, and I like to think I'm pretty good. I have a number of songs/riffs I've written that I'm proud of; I can play the Metallica songs that inspired me to play in the first place (and seemed impossible-tier when I first heard them) with relative ease. I'm even getting decent at learning Eruption. But then I go and watch Paul Gilbert or Yngwie or another shredder online who can cover something one of those guys made, and it's just like I'm a nobody all over again. But I remind myself that 1) the biggest factor in how good you are is how experienced you are--how long you've been doing it which isn't something I can control, except to keep going and 2) it also doesn't matter as much how technical you are. So many of the best songs and most iconic bands are at a way lower level of technicality than the super-shredders...and guess what, those guys are more famous and I personally like them more anyway. So just do what feels good. But then at the same time there's always a part of me that wants to be the best and is always working to get there little by little, so seeing me blown away by the "competition" is a bit of an ego hit, but it's only realistic.

 

"No matter how good you are at something, there's always about a million other people who are better at it than you."

-Homer Simpson

In my opinion, this would not be a competition against machines, because they would not be competing with humans in the task of making maps (except in a hypothetical scenario in which a competition was opened and human participants are in such a competition with the same goal of creating a map that meets the competition's parameters and then sent for the referees or the community, for voting). This would therefore be a new way of art in Doom. And I also believe that it would be much more fun if the AI that created a wad explained, in the description of its work, what it would like to provide to the player; for example, The AI explain why it made a curve with a certain shape, or why it chose certain colors, songs, textures, to provide the player with different challenges and sensations, feelings, throughout their gameplay. In other words, that the AI explained, in the description of the wad, what it took into account when deciding the concept of beautiful, what would provide aesthetic pleasure in it work.

[erzboesewicht]

Just played some OBSIDIAN maps to see how far we are away still from this scenario, and I think we're still lightyears away.

Visuals were nice but in all maps seemed chaotic (almost Hazmat-Hazama style but more boring/samey), a lot of health and ammo was placed always in the first room (Nirvana-inspired?), weapon placement was nuts (in one map I got the shotgun last and the BFG as one of the first weapons, and 50% of the surface of that map was only there to achieve a single goal - that shotgun :) - the exit was readily accesible before. And no interesting optional fight either ...)

 

Biggest weakness however seemed monster placement. They were roaming around completely randomly, there were some homogenous groups but their placement didn't make any "deeper" sense.

 

And that's also where I think the problem is - would an AI be able to really produce creative monster placement? In theory it's all math, so a statistic AI like this famous image generator (I forgot the name) should be able to, but wouldn't then all maps seem like a collection of earlier monster placement ideas?

 

I think one could for sure create also a kind of "creativity module", exploring new ideas, maybe collecting data from the world outside of Doom to get inspiration. For example, they could explore the data and structure of other games (from board games like chess, up to other video games) and sports, create new games and sports from scratch and then "Doomify" ideas out of it, etc. But would it be as focused, as good, as a human-made map with a really original idea - to give an example, like skillsaw's "Mancubian Candidate"?

 

(I for myself are quite interested in the AI topic, and I would in fact love such a scenario where humans and machines compete from equal to equal. I think humans would still prevail in some way, though, as they can always have all the AI help and then add something "human-esque" to it. It wouldn't be like in correspondence chess.)

 

Edit: To those familiar with Obsidian/Oblige etc.: Which was the best procedurally-generated map you played? Are there already some "masterpieces" and "classics"?

Edited September 11, 2022 by erzboesewicht

[dasho]

I would say single maps are arguably our biggest weakness if that's what you tried, especially WRT the weapons it gives out. If you run GZDoom, I would select "ZDoom Family" as the engine and perhaps give one of the procedurally generated bosses a shot as well as the Epic Resource Pack. Monster placement is a bit of a deeper subject, as gunfire will wake up quite a bit of the map, even with a few mitigating factors in place. Appreciate the feedback, though.

Edited September 11, 2022 by dasho

[erzboesewicht]

Thanks dasho for that comment! I think Obsidian is pretty good for a procedural generator, above all if the main aim is maybe to create visuals as a kind of rapid-prototyping tool, and I think for such a generator to really become better (and come close to "human" creativity) in monster placement may be a big challenge ... Yeah, I chose single maps indeed, as I wanted just to try out the software and I'm a pistol starter anyway, and the ZDoom family in fact was the only one that worked (Boom/prboom+ didn't, and I didn't try vanilla). Will try to play around a bit with the settings when I've time, I mainly used standard settings but with higher monster density.

 

What may be an intermediate step to a full AI would be maybe a module which can try to simulate the player's experience, an independent software function which acts like a bot and "plays" the levels a number of times using typical human reaction patterns, gathering stats about how often it dies etc..

 

If this function becomes good it could also help not only the software but also players to find wads they may like. I could then let the bot play a lot of available mapsets and milk their stats for map elements I may like (in terms of gameplay, for example slaughter, non-slaughter hard or easy, forgiving/requiring precision, RNG-heavy, puzzle-combat etc.). I wonder if this in the end would be good or bad. I wouldn't have to play wads I won't like, but maybe I would miss some "excentric" wads with some "likeable" element I can't analyze statistically ...

[Murdoch]

I struggle to think of anything that interests me less than AI generated levels. It's simply never going to be as good, heartfelt or creative as those of an actual person. I might play one or two out of curiosity, then never give them another thought.

 

[Chezza]

It would be an interesting little social experiment to tell a group of testers a human made an A.I. map and visa versa to see if there is a natural prejudice on either result.

 

I'd imagine people would look for repetitive, consistent quality and symmetry to assume it's not a human creation. 

[Pechudin]

4 hours ago, Chezza said:

It would be an interesting little social experiment to tell a group of testers a human made an A.I. map and visa versa to see if there is a natural prejudice on either result.

 

I'd imagine people would look for repetitive, consistent quality and symmetry to assume it's not a human creation. 

 

DooM Turing test - if an AI can make a map that is indistinguishable from a human-made map, it is sentient.

[Captain POLAND]

What if there was a way to make an AI download and analyze every single WAD in the idgames archive and make a map based on all of that data? I wonder what it would be like.

 

 

[Pechudin]

7 hours ago, Captain POLAND said:

What if there was a way to make an AI download and analyze every single WAD in the idgames archive and make a map based on all of that data? I wonder what it would be like.

 

 

Probably uncompletable, unless you also told it how to join parts together which would be harder.

AI is not magic, it's still just algorithms that need to be programmed by humans.