Support NeoGAF

[IDKFA]

This was actually Toots idea, but they didn't want to make a thread, which I think is an interesting idea.

Recently DALL-E 3 has taken GAF and the internet by storm. Type in a few prompts and you can make any image of your imagination (with some limitations, but this pretty amazing)

This tech has come a very long way in a short space of time and this is just for images. How long do you think it'll be before we'll be able to type in a few prompts and you'll have a custom made video game of your dreams?

Imagine you want a FPS set in Jurassic Park, but every dinosaur has the face of Joan Collins? Type in a few lines of prompts and boom, you get that FPS. You can then add to it by adding weapons, story, levels etc etc.

Do you think this tech would ever be possible? If so, when?

Also, if this tech was possible, what would it mean for game developers and publishers? Would they become obsolete?

Edit. This is the best DALL-E could do with my game idea

 

[SlimeGooGoo]

How do you handle collision and physics with AI?

It's one thing to generate textures and 3D models, another to generate response to dynamic events.

At best AI will be a tool to complement the developers skills, not to trivialize the process.

 

[DownLikeBCPowder]

Honestly pretty soon if I had to guess - you can already have the basic chatbots spit out basic game code for text adventures and probably more.

I think it'll be an amazing tool but for now we'll still need the human element to make it work together in more meaningful ways.

 

[GymWolf]

Unless your videogame has Bolo Yeung in it.

 

[tmlDan]

AI always feels soulless, i can't explain it but the best way i can describe it is that AI feels like Mario with Chris Pratts voice.

It's always noticeable, so i'd rather not have games be made solely from AI. Also, don't think it's taken anything by storm, unless you talk about super tiny niche circles (DALL-E) lol

 

[A.Romero]

Within 50 years, is my guess.

It will be possible to interface with VR and AR. It's going to be wild.

 

[jshackles]

I think the first step would be for AI to make convincing non-interactive content aka movies or TV shows, which is already in its infancy (like that weird Seinfeld-like show on Twitch) but a long ways off from being actually engaging entertainment.

But, like anything else with machine learning, humans seem to be pumping out an endless supply of inspiration for AI to make new games. Hundreds of new games are released on Steam every day now. It'd only take someone to train the ML on certain aspects of level design, graphics, etc. and have a reasonable amount of compute behind it to make something useful.

Realistically, I think this could already be done - but only on a simplistic level, like arcade games of the 1970s. It'd be possible for a modern AI to generate (for example) great new Pac-Man boards or interesting level layouts for Atari Combat for the 2600, maybe even interesting new zones or enemy types for something like Pitfall. Moving forward a generation to something like the first Super Mario Bros game, and I think modern AI would fail to make something that wasn't just a randomization of existing assets. Having AI make new power ups, enemy types, or obstacles would probably result in a game that's unplayable or unfun.

For the type of stuff you're talking about, I'd say we're probably 10-15 years away from that, given the current trajectory of the technology. Obviously it could happen a lot sooner than that if there are huge unforeseen breakthroughs or paradigm shifts.

 

[Deerock71]

The REAL question is who's going to have the magic spectacles?

 

[StreetsofBeige]

At some point when it's easy for everyone to make a game (even if it's not AAA), I think a good benefit is that if AI spits out all the code, you'll be able to pass it one or share it.

Thenn someone can take it and adjust it if they want. So lets say someone makes a decent modern warfare shooter. I think it plays great. I take his shared code and tweak it so it's a WWII theme. I know nothing about programming. Maybe it'll spit out a good WWII version.

Who knows.

 

[IDKFA]

Within 50 years, is my guess.

It will be possible to interface with VR and AR. It's going to be wild.

50 years?

With the current rate of AI advancement I'd say 10-15 years.

At some point when it's easy for everyone to make a game (even if it's not AAA), I think a good benefit is that if AI spits out all the code, you'll be able to pass it one or share it.

Thenn someone can take it and adjust it if they want. So lets say someone makes a decent modern warfare shooter. I think it plays great. I take his shared code and tweak it so it's a WWII theme. I know nothing about programming. Maybe it'll spit out a good WWII version.

Who knows.

Yeah, I probably think this is how developers will still have a place with this tech.

For example, the Creative Assembly release a blank Total War maker. With it, you add prompts and make the Total War game of your dream.

The mechanics of Medieval 2, with set in the 17th century (Pike and Shot!), with a complete map of Europe, Aisa, North America and North Africa, full blood and gore etc etc.

After some time you then get the Total War game you've always craved.

 

[A.Romero]

50 years?

With the current rate of AI advancement I'd say 10-15 years.



Yeah, I probably think this is how developers will still have a place with this tech.

For example, the Creative Assembly release a blank Total War maker. With it, you add prompts and make the Total War game of your dream.

The mechanics of Medieval 2, with set in the 17th century (Pike and Shot!), with a complete map of Europe, Aisa, North America and North Africa, full blood and gore etc etc.

After some time you then get the Total War game you've always craved.

Unless quantum computing accelerates and becomes ubiquitous I don't think computing power will become cheap enough to produce stuff like this. Right now even making genAI images can be kind of costly (comparatedly) in the cloud. To have what OP describes we would need GenAI that can generate 3D models (we don't have that right now), cloud based game engines that support it, genAI sound engines, physics genAI engines and even faster internet connections unless it is streamed.

Also it's not only about the tech, most people can't articulate a description of a game they would want fully so the engine would have to fill in a lot of blanks and I believe that will take some time as well.

In any case, it will be pretty soon. 15 years could be a reasonable timeframe for some of that stuff to start happenning.

I like your TW idea, btw.

I'd like a more nuanced system of emergent gameplay: start playing an RPG game and then the game takes you to unexpected outcomes that are generated uniquely for each player.

 

[Holammer]

It's going to come in stages. We're at Tier 1 now, the early days where ChatGPT can assist a coder that knows what he's doing.
I say give it another 5-6 years and it has progressed to the point where it can code semi-complex programs based on input from a smart non-coder. Programs like Photoshop and Blender will have lots of different AI integration. By 2033 it can put together very complex software for a computer-illiterate person.

Along the way it'll learn to interface with other AI generation systems and programs to synthesize code, music and art. Also, AI won't be reinventing the wheel, they will create API and middleware or use existing OSS like everyone else to speed up the process.

 

[IDKFA]

It's going to come in stages. We're at Tier 1 now, the early days where ChatGPT can assist a coder that knows what he's doing.
I say give it another 5-6 years and it has progressed to the point where it can code semi-complex programs based on input from a smart non-coder. Programs like Photoshop and Blender will have lots of different AI integration. By 2033 it can put together very complex software for a computer-illiterate person.

Along the way it'll learn to interface with other AI generation systems and programs to synthesize code, music and art. Also, AI won't be reinventing the wheel, they will create API and middleware or use existing OSS like everyone else to speed up the process.

You raised a good point there about art and music.

A game is more than gameplay and graphics. You also need to factor in art style, music, sound effects, gameplay goals and triggers etc.

If we're talking about creating the dream game with all of that included, we could be some way off.

 

[Johnny Concrete]

At some point you’ll be typing so many prompts to correct the AI that it’d probably be faster to just old fashioned script a game from scratch.
Especially when you are in the last phase of development.

“Make Monster4 hitbox on legs slightly smaller”
“Increase reload time on Gun7 by 0.4s”

 

[ReBurn]

It won't be us creating them. I will be the AI.

 

[IDKFA]

At some point you’ll be typing so many prompts to correct the AI that it’d probably be faster to just old fashioned script a game from scratch.
Especially when you are in the last phase of development.

“Make Monster4 hitbox on legs slightly smaller”
“Increase reload time on Gun7 by 0.4s”

This is where developers will still have a purpose. They'll create the engine, but the player can then customise the game as they see fit.

Almost like modding, but a far easier and user-friendly version. For example, replacing all the models in Titanfall 2 with Warhammer 40K models, then changing up the guns and levels to give it a 40K look. Then you can go further by changing the layout of the levels, adding enemy placements, adding goals etc.

 

[cireza]

We will generate movies first. And that won't take very long.

 

[IDKFA]

We will generate movies first. And that won't take very long.

Agreed. AI is already creating videos. Give it five years and we'll get full blown films. I'll be able to get a real Ghostbusters 3 set in the 1990s. Would be cool if I could also write the script.

 

[Mephisto40]

I wonder what would happen if you told AI to make the greatest video game ever made

 

[Toots]

How do you handle collision and physics with AI?

Is it some kind of calculation ?
Because if so i'm pretty sure AI is better than us at doing it.

At some point you’ll be typing so many prompts to correct the AI that it’d probably be faster to just old fashioned script a game from scratch.
Especially when you are in the last phase of development.

“Make Monster4 hitbox on legs slightly smaller”
“Increase reload time on Gun7 by 0.4s”

For a programmer yes, but it will open a world of possibility for us plebs who don't know how to code. It might also mean that every game ill be a single player experience since every game will be tailor made to fit the desires of the one who "asked" the AI for it.

You raised a good point there about art and music.

A game is more than gameplay and graphics. You also need to factor in art style, music, sound effects, gameplay goals and triggers etc.

If we're talking about creating the dream game with all of that included, we could be some way off.

That is an excellent point, and im wondering what is even a dream game, and is the best game ever something we know we want ? Maybe it's not the game we want, but the one we need, dark knight style. That's what always strike me with art, especially litterature and poetry, when you discover some art that truly resonate with you, it is like you always had it within you, you just did not know it yet. The art piece plucks something from your subconcious and brings it to light. You read a poem, and every word feels familiar, it is something you always wanted to say, without knowing it.
AI can only give us what we know we want. The most creative thing they can do is giving us a super weird version of what we ask, but the human prompt is still there. An artist gives us what we didn't know we needed. I think the real AI revolution, the one that will make us human beings obsolete is when it starts doing stuff for itself, without human interaction. When it starts telling jokes to make itself laugh. We'll never catch up to it then. Maybe it will allocate a fraction of its computing power to create a perfect society for us, maybe it stop interacting with us, maybe it will go skynet and remove the dangerously unpredictable human being variant from the life equation.
Maybe someone will use it to create a computer simulation of a perfect woman in order to practice communicating with girls. However, a freak lightning storm will bring her to life, creating Lisa, a gorgeous genius with the powers of a "magic genie". I mean it's possible, i saw it in a John Hughes movie

 

[SimTourist]

There isn't enough time in life to play even all manmade stuff. An AI like that would lead to Steam being flooded with even more shit.

 

[Toots]

There isn't enough time in life to play even all manmade stuff. An AI like that would lead to Steam being flooded with even more shit.

If we let the merchants have their way for sure.
Let's drive them out of the temple.
I'm only half kidding, we should collectively agree AI is too important for humanity to shackle it with commercial consideration. If human rights exist , AI use is one.

 

[FunkMiller]

There's a very, very big gap between AI being able to structure new content based on existing content, and it being able to create new content itself.

It's a good while off yet before AI can independently create a piece of content or art without any help from humans - if ever.