Emulators and AI

This stuff is still firmly in the lab. As far as your want of some kind of cabinet that fires up, figures out how to play games and then maybe even does pretty good at them, we are far, far, far off. I'm talking decades at least.
It's not decades off when it's being done on current hardware using ML or DL techniques and a bit of python. You really just need a scripting language, a way to extract frames from a video (ffmpeg, etc), a way to tag frames for data, then using an object detection model like yolo or others out there to create a dataset. Then it's a matter of using the training dataset.

AI has been big in the news lately, but it's good only at certain narrow things. It would take more of a generalized intelligence to be able to recognize the content in real time from the screen, figure out what it's looking at and send back game inputs quick enough to matter.
This misrepresentation of the term "AI" only further confuses people in the conversation and elsewhere. LLMs are general purpose sure, however ML/DL models are usually quite specialized at running specific tasks. What's shown the news today is AI in the broad sense sure, but they're just LLMs, most of which can run locally on consumer hardware (a la llama 4).

An example, OpenAI's Five is an algorithm (trained using ML/DL methods) that plays Dota 2 at a competitive level and came out in 2017.

I've built a few detection systems using openly available object detection models and it's not entirely complicated or definitely not decades away, it's rather who want's to spent the time to make something described in the thread.
 
It's not decades off when it's being done on current hardware using ML or DL techniques and a bit of python. You really just need a scripting language, a way to extract frames from a video (ffmpeg, etc), a way to tag frames for data, then using an object detection model like yolo or others out there to create a dataset. Then it's a matter of using the training dataset.


This misrepresentation of the term "AI" only further confuses people in the conversation and elsewhere. LLMs are general purpose sure, however ML/DL models are usually quite specialized at running specific tasks. What's shown the news today is AI in the broad sense sure, but they're just LLMs, most of which can run locally on consumer hardware (a la llama 4).

An example, OpenAI's Five is an algorithm (trained using ML/DL methods) that plays Dota 2 at a competitive level and came out in 2017.

I've built a few detection systems using openly available object detection models and it's not entirely complicated or definitely not decades away, it's rather who want's to spent the time to make something described in the thread.
I think the difficulty is really upped because he doesn't just have a game or two that he wants it supported on -- he wants to toss a whole ROM collection at it and watch it go. Sooner or later someone'll do it, probably just not today. ::biggrin
 
I would ask how is it easier* than just looking at any of the live streams like this one from Kawaii Games?

(*) original hardware would take so much hardware, and a computer a ton of setup... And that's if AIs really existed (see prev comments about LLMs).
 
I would ask how is it easier* than just looking at any of the live streams like this one from Kawaii Games?

(*) original hardware would take so much hardware, and a computer a ton of setup... And that's if AIs really existed (see prev comments about LLMs).

If I wanted to just watch some videos I would simply do that, don't you think?

I'm asking how to automate a particular task on my computer, and you people keep answering me with some "why don't you just watch a video of someone doing that task on his computer?", or some "why don't you just play the game instead?"
 
If I wanted to just watch some videos I would simply do that, don't you think?

I'm asking how to automate a particular task on my computer, and you people keep answering me with some "why don't you just watch a video of someone doing that task on his computer?", or some "why don't you just play the game instead?"

It's because the thing you ask for is not really feasible currently.
 
I'm sure it's doable, especially with MAME and 2D fighting games such as street fighter and king of fighter series.

This is 6 years old
Look, it was you who said it was complex to take your browser and look for the video. And that's what my answer tried to explain, that this is more difficult.

I mean, even if you know python, that could take quite some setup... One has to add the time to train the model for each game...

I just checked it and the python code seems pretty straightforward. You could give it a try.

Maybe it's easier than looking for a video (thus, proving my point moot).
 
Look, it was you who said it was complex to take your browser and look for the video. And that's what my answer tried to explain, that this is more difficult.

I mean, even if you know python, that could take quite some setup... One has to add the time to train the model for each game...

I just checked it and the python code seems pretty straightforward. You could give it a try.

Maybe it's easier than looking for a video (thus, proving my point moot).
that is not complex code just the abuse of using branch algorithm, spamming ↓→ lp and ↓→ ↓→lp
 
So dude just want ai to play games so he can monetize effortlessly dozens of ai slop letsplays at once
 
ai can be helpfull (bots) for grinding money/loots on mmorpg
 
I'm sure it's doable, especially with MAME and 2D fighting games such as street fighter and king of fighter series.

This is 6 years old

That's a cool video, but as you can see it's not that fun to watch, the AI spam the most effective moves without really reacting to the game. I think using recorded input files from good players with Retroarch's replay system would better fit your project.
 
I'm sure it's doable, especially with MAME and 2D fighting games such as street fighter and king of fighter series.

This is 6 years old
The overlap between people claiming a certain thing is easy while also not ever trying or knowing how to do that same thing is pretty large.

If you really think creating an "AI" that can make interesting gameplay demo/videos/sessions from arbitrary games picked out of the 50,000+(?) released over the last 40(?) years, I say go for it. You'll certainly be rich if you get this done, as an AI that can take on a task like that can certainly accomplish a lot more. I think you'll quickly find that even if you restrict this task to JUST MAME and JUST fighting games it's still a massive thing to take on.

If you think what it takes for a person to play a fighting game, you'll start to appreciate it -- it involves looking at a screen in real time and trying to decide how to win and the current state of the game and then extremely rapidly sending game inputs back. Most people aren't even good at all fighting games, never mind a program that can take this on in real time...
 
I think most arcade games have rythm-based, reproducible sections and optimal paths that wouldn't even require an AI to complete. Some people are able to complete games blindfolded. The difficulty of those games is when you *dont* know the patterns and rely on reactions and reflexes to learn and complete the games.
 
If you really think creating an "AI" that can make interesting gameplay demo/videos/sessions from arbitrary games picked out of the 50,000+(?)...
 

Users who are viewing this thread

Connect with us

Support this Site

RGT relies on you to stay afloat. Help covering the site costs and get some pretty Level 7 perks too.

Featured Video

Latest Threads

An article about retro gaming from 1996

So, I was reading some old gaming mags and came across an article about retro games and how they...
Read more

Metal Gear : Rogue (PC)

in development, btw you can already try it, i had some fun testing it...
Read more

Mid-Year 2025 GOTY Discussion

We've gotten past the halfway point of the year about a week and a half ago. (time sure does fly...
Read more

Weird South Park Reference

I was just rewatching the World of Warcraft episode of South Park which came out in 2006...
Read more

What fighting games do i get? (DreamCast Edition)

So a while back, i posted a thread for people recommending fighting games on the PS1, and now...
Read more

Online statistics

Members online
97
Guests online
170
Total visitors
267

Forum statistics

Threads
10,288
Messages
255,192
Members
820,031
Latest member
Hurtcarnival

Advertisers

Back
Top