Gran Turismo 4 Used to Train World’s First AI-Generated Multiplayer Tech Demo

  • Thread starter Jordan
  • 16 comments
  • 2,390 views

Jordan

Site Founder
Administrator
24,509
United States
United States
GTP_Jordan
GTP_Jordan
This is the discussion thread for an article on GTPlanet:

Gran Turismo 4 Used to Train World’s First AI-Generated Multiplayer Tech Demo

Gran Turismo 4 has been used to train a new AI research project called Multiverse. It’s the first project from a new AI development company called Enigma Labs, comprising some young programmers (and Gran Turismo fans), to demonstrate a new technique for creating AI-generated worlds for multiplayer games...
 
Gonna be a big nope from me, especially at a time where the Gaming Industry is already treating its longtime Human talents like slaves and leaving them to bite the dust left and right just because corporate isn't making massive profits for their respective quarter.
 
First of all, AI replacing jobs can get in the bin especially in this current climate in the gaming industry.

Secondly, I don't really understand. AI used footage of Gran Turismo B-Spec to create a multiplayer Gran Turismo on the same track with the same cars?
 
This looks interesting. Games take so long to create and are always released as beta’s. Maybe this would speed things up and reduce errors.
 
Secondly, I don't really understand. AI used footage of Gran Turismo B-Spec to create a multiplayer Gran Turismo on the same track with the same cars?
Yes and no.

Aside from... other concerns, the technical breakthrough here is making a multiplayer AI-generated game. There isn't actually a game here: you just go to the interface and type in prompts (I recall the guys said it was WASD) to control your "car" and the AI generates everything you see on the fly as a consequence of your inputs... along with the inputs of the second player.

There's no assets, no game engine, no anything.
 
reduce errors.
I am playing the pessimist card: Never happening.
AI can not reduce errors, because it requires a supervisor that spots every single error that the AI produces and then correct the command because otherwise it would build upon that error and making everything worse in every iteration.
That is more work than doing the work yourself in the first place.
 
I am playing the pessimist card: Never happening.
AI can not reduce errors, because it requires a supervisor that spots every single error that the AI produces and then correct the command because otherwise it would build upon that error and making everything worse in every iteration.
That is more work than doing the work yourself in the first place.
Today sure. Tomorrow, or next year. Not so much. In photography, we used to pay a guy very well, to cut masks around objects during the retouching process. Very slow, meticulous work. Now my iphone guesses what it is, and bangs it out in a matter of seconds. Heck, the nhl is masking players on the fly, during live video, and putting ad’s on the boards behind them. Couple that with long, and often delayed, development times..also the huge budgets..i’d say gaming is pretty ripe for something like this to emerge from.
 
There is the crack, someone needs to make this guess into a 100% hit rate, otherwise you need someone to monitor the error rate and correct it.
Otherwise **** happens.
Heck, the nhl is masking players on the fly, during live video, and putting ad’s on the boards behind them.
That fortunatly doesnt require any quality/precision, only roughly tracking the things to change and then broadly overwrite them.
 
I guess the point was there was a pit of 5 people working for $1-200 and hour for almost 20yrs. Now there is one, and the bulk of stuff gets farmed out to india for $10-15 an hour for that double check. Programming and that sort of thing looks as if its about to discover a similar fate just around the corner.
 
Today sure. Tomorrow, or next year. Not so much. In photography, we used to pay a guy very well, to cut masks around objects during the retouching process. Very slow, meticulous work. Now my iphone guesses what it is, and bangs it out in a matter of seconds. Heck, the nhl is masking players on the fly, during live video, and putting ad’s on the boards behind them. Couple that with long, and often delayed, development times..also the huge budgets..i’d say gaming is pretty ripe for something like this to emerge from.
I suggest we vote with our wallets.

Do you want for tens (or hundreds) of thousands of people to lose their jobs in the arts?

How will they support their lives without universal income?

Don’t support generative ai.
 
Last edited:
Reminds me of that " AI-generated" Doom demo where looking down and then back up would have you in a completely different place. Enemies would change, some would get instantly killed the moment the camera focused on them etc.

These things will get better with time, and I very much dislike it. As an artist, myself, its sad to see people willingly giving up on the more meaningful parts of the joy of creation and handing it to a slop machine. Sure at some point the quality of generated content will improve and many of these models will make sense from a business point of view. Best I can do is urge people to enjoy things created by other fellow humans.

--

From a more "game" point of view, this kind of interaction of simplified game with some sort of "impressive" visual reminds me of when FMV games were all the race in the 4th generation of videogames. Real-life footage and some limited input you can give to the game. Worked in some cases, didn't work in most others. Was a curiosity at best. Granted, that also included having people acting in the most low-budget ways that only niche videogames could do at the time. Night Trap and similar games spring to mind.
 
the irony being lost that this is being trained on gran turismo of all things

not exactly a series known for good AI at all

These things will get better with time, and I very much dislike it. As an artist, myself, its sad to see people willingly giving up on the more meaningful parts of the joy of creation and handing it to a slop machine. Sure at some point the quality of generated content will improve and many of these models will make sense from a business point of view. Best I can do is urge people to enjoy things created by other fellow humans.
You're doing more to understand it than most artists who are screeching to nuke it from orbit because of some morality reason.

In a way, even with all the errors and weirdness, it gives the average individual a bit more chance to express themselves in a way they previously couldn't.

My biggest issue is people selling AI generated content on a commercial level or using the generated results directly (rather than as some sort of guidance or to "fill in the blanks", theres certainly people out there with fun ideas without the general abilities to actually take them somewhere past their own head, and it's unreasonable to expect them to go begging other people to slave away for them if they're not a celebrity...)
Worked in some cases, didn't work in most others. Was a curiosity at best. Granted, that also included having people acting in the most low-budget ways that only niche videogames could do at the time. Night Trap and similar games spring to mind.
FMV games are honestly interesting, as are techniques like pre-rendering 3D graphics (or even motion capture like Mortal Kombat).

(whoever merged these posts thank you)
 
Last edited:
Back