discussion How long before AI is able to make a full Renpy VN/games like the ones we play today?

[ShiaLabeouf]

AI is progressing pretty fast, but I'm not sure if it's consistent enough to actually put a game together in a competent and coherent manner. It's not just about making a game that's functional, it's also about making a game that makes sense.

If the pace of AI continues, then I don't doubt at some point it'll be able to make a pretty convincing game, but what of updates or bug fixes? Will it be possible to have an AI read comments and then change the game based on that feedback? That's pretty complicated; at some point along this development cycle, I'd imagine a human would have to step in.

 

[LUVGAMZ23]

I have no idea. But let's pretend i.e. NOT REAL:
In 15 years, anything creative, ANYTHING, computer stuff, paintings, books, food (using robots), etc... can be done with AI + Robots BETTER than humans can do.

Would your reaction be:
1-Oh. I guess I was wrong. I've learned something today.
2-OMG. Machines are better than humans. We have to make this illegal / regulated because...

NOT trying to pick a fight, but I've always liked trying to challenge others or myself.

Speaking personally: I would be very disappointed if in 10-20 years, AI creativity wasn't at least on par with humans, if not better. I'm a misanthrope i.e. not a fan of humans... The more humans are out of the equation, the better... for me.

 

[magicace]

At the rate AI is developing, I'd say 3 years or less. Developers are already integrating AI into games.

 

[Eastin0504]

A full blown VN? we're far, far away and it's possible it'll never exist. But AI tools that will assist in the creation of VNs will get better and better as well as people's ability to use them

 

[zedascebolas]

Probably 4 or 5 years. For now we are just seeing ai cgs, but I hope AI gets so much better before we get full blown stories written by AI

 

[andy0665]

in 5 -10 year it will get another level

 

[offblest]

For developmental purposes, it would assist the developer greatly in writing text or composing images. But I think it would take a decade or more before we see AI make a game completely by itself.

 

[Ribaso]

50-100 years, and for simple reason...Artificial "intelligence" when it comes to how humans operate in real life is rather stupid.
No amount of iterations can spawn a extraneous notion, or what a human might call an innovation, there is a reason for the "uncanny valley" after all,
It's your brain respecting whatever you're seeing but finding it biologically unfit...Maybe a breakthrough in quantum computing would be finally the tip of the iceberg.

 

[jnathan]

Yeah, it's not coming as quickly as people think, if ever. AI is not operating at the same rate as Moore's Law. It doesn't double in power every two years. We had computers that were having conversations with humans in the 50s and 60s, when the Turing Test was created. So why didn't we hear a lot about AI until the last handful of years? Well, because big business seriously bought into AI very recently thinking money would get us there faster. It hasn't. AI has short bursts of surprising "growth", and then plateaus quickly based on current computing power and how well a human can teach a computer to recognize patterns. That's all AI does right now, pattern recognition. The current way of expressing patterns, LLMs, also has its limits.

A VN combines a number of disciplines. To make a VN, AI would have to be able to do all of them. And right now it does none of them very well, especially at scale.

For context, I regularly run local LLMs on my PC to generate images. (Primarily using ComfyUI, but occasionally other LLMs.) My RTX 3060 non-TI with 12GB of VRAM takes about two minutes to generate a 1920x1080 image based on text prompt that I provide. These text prompts are carefully created by hand after many hours of experience. And I can easily go through hundreds of generated images in a day and only collect a small handful that were okay. It'd take much longer to generate several images of a very similar looking character wearing the same clothes in a different pose with a different camera angle. To say nothing of a consistent environment, with different angles of the same location, matching light conditions, matching color grading, matching detail, etc. Add more characters to the shot and it loses control. AI can't even get fingers right, let alone multiple subjects. And even if it could, I can't image trying to get stills to transform into video that also has to have the same character wearing the same clothes in the same environment. And this is just images and videos.

I've also spent many hours generating music (mostly hundreds of songs on Suno), and it's pretty hit or miss too. And that's with a human who has learned how to write prompts to make it more successful. It takes even longer if I want to expand the length of a song and add details. The bigger it gets, the more it falls apart.

The same thing happens when using AI to write code, or tell a story. AI is barely okay at small tasks, and completely useless at large ones. And the amount of human work and computing power it would take to get one discipline to scale is well beyond what is currently available and what will be available for decades.

I'm personally of the opinion that the limit of AI is far below the limit of human potential, because humans train AI, and humans don't understand themselves, let alone the patterns a computer would need recognize to match human output. Uncanny Valley is very real, and I don't personally believe AI can cross that valley. AI is at the limit of our own imagination, and the limit of circuits.

 

[frafildur]

Sooo.... I'm a Singularitarian who has been following the field very closely since early 2023. No comp-sci/engineering experience, but genuine professional experience in ethics of new & emerging tech. First, some responses...

In podcasts that I have already listened to on the subject (interviews basically, in portuguese), one from an expert in the field and another from a respected neuroscientist, they said that the hype for AI is exaggerated. The technology and AI expert said that we are very far from a real AI, and an entire game is something very complex that could only be done by a real AI in my opinion.
...
Sorry bad english, I hope most of my comment to be readable for you.

I don't think the hype is exaggerated. The task I ask of skeptics is this: try ChatGPT 3.5, then try Claude Sonnet 3.5. These models were released less than two years from each other. Although it doesn't -cement- the nature of what's happening as exponential increase, I believe it strongly suggests it. Somewhere between 4-18 months from now, once GPT5, Claude 4, Gemini & Grok (ugh) 2 are out, we will have a much clearer picture whether or not the gains are exponential. Interesting times...........

5 years. This stuff is exponentially increasing, and it'll take us by surprise. Just hope some of the low-level (task based) engines remain controllable by people, as in we input prompt/seed and it takes it from there.

Exponential increase, I'd think more like... 2 years. Solitary 1-shot LLM material is already vastly inferior to multiple model systems. I'm trying to say something like: the systems the public has access to right now is already vastly outdated. With the next tech leap, likely within the next 6 months, we should get a much better picture of whether or not this is exponential. If it is exponential, it seems like we're around Day 26-28 of the 30 day lilypad-doubling thought experiment.

It's already generating things like this lol. You're finally awake!!!!
c'mon I can't w8 GTA 7 that's made by AI alone, ain't no human involved, I hate humans they do stupid things, demand water/food/green paper, meh.

Embrace the singularity. Accept the ruling of the robot overlords. Superintelligence, by definition, knows perfectly what is best for you.

So, like, you know, I'm absolutely ready to admit I'm not fully commited to believing that we are in an exponential cycle just yet. However, the evidence I follow suggests it's more likely than not that we are... in which case: AGI sometime in 2025 followed by a tempered (due to physical constraints) intelligence explosion resulting in ASI by 2027.

In which case, to answer the OP's question: mid-2025 latest.

 

[person985]

A good while, the problem right now is that AI lack consistency. Without human intervention an AI will just spiral into nonsense and insanity without a hint of coherence. There is also the fact that AI aren't very good at actually creative endeavors if left to its own machinations as everything an AI make has to be derivative of something else because of how it collects information.

 

[Trunk Monkey]

Using Dall-E3 via ChatGPT4, you absolutely can write a prompt that generates a sequence of consistent images that tell a story. It's been done. handling branching storylines would be problematic, but I don't think it'll take too long before a LLM can handle it useing something similar to Dall-E to create the images.

 

[stevetheegg]

Soon, I hope! Small developers don't have the resources to keep up production at a good pace. The better AI gets, the faster creators will be able to focus on storytelling.

 

[blackstars91]

soon hopefully but it will probably need some human interraction for being cohérent

 

[User_40809]

Considering how quickly AI has evolved, I would say it won't take long. Maybe about 5 years.

 

[DarEx947]

i say 5 years before its attempted. and 10 before it becomes a readily available thing.

 

[Loliholic]

Since most games share the same 5-7 plots, i wouldnt be surprised if they were already made by some scrappy AI without the slightest hint of creativity...

 

[User_111574]

At this stage and in the short term, nothing will get done without human involvement. As for the longer term, we can only guess. For me, the closest date is 2040. And beyond that... To paraphrase the words of Hoja Nasreddin, during this time one of us will die - either the Shah, or the donkey, or me.

 

[Archer]

Like a few other posters, I'm an enthusiast of AI, but not the best on the technical side of things.

Will AI be able to generate its own games? That's a cool dream, basically infinite erotica that you could have to meet your demands.
Most likely no.
Potentially, yes, but it would be really bad or repetitive

My experience with AI and talking with others shows that things like AI storytelling really starts to break down the longer the story is. Mistakes become common, characters get switched around, story lines get randomly dropped.

If it happens, I'd expect it to be very repetitive. This is an experience I've had with a lot of AIs. I ask it to tell a story, generate an idea, etc. and I'm amazed with the result. Then I give it a couple more goes and I quickly see either similar things happen or I try to force more randomness and it just loses coherency.

So could AI speed up development of games?

I don't know enough about the art side to comment on that.

Storylines? Maybe in like a sandbox style game you could have the AI write a bunch of generic dialogue to flesh out the world. That would be helpful and reduce the workload, but I think humans will still need to be thinking of and doing the majority of the writing on stories for some time to come.

I guess a question is could an AI generate a bunch of the story based off prompts and the writer go through and edit it? Something like a prompt 'write a dialogue between the main male character and his crush next door that hints at sexual tension between the two' and then the game designer edits that dialogue. Maybe, but I doubt that would be quicker, and editing is even more boring than writing. But if we were going to see something speed up game development (outside the art side), my guess would be something like that.

 

[MystrageX]

I dont know long, but I cant wait to see what they can do when it comes. well that's until SkyNet happens of course.