More and more games seem to suck on thier own, but can be great with mods. You have entire platforms like roblox where all the games are more or less mods. How long until the platform itself is community created and managed and the viability of games created by companies dissappears?
As is, theres no sign things will go that way quickly. The increase in larger mods is more a product of increased funding and increased (legal) support from publishers. Things like Roblox’s microtransactions make it very profitable, even if a lot of time or money is spent in development. For more general game development, most of it hasn’t changed in about a decade, and I don’t see AI affecting much. AI can’t reliably create good results in any field, nonetheless combining them. Basically, making any large project just costs too much to give away for free.
Yeah, it could still be a good while away. AI can help a lot in some places. Not so much in others. Like if you had modules and plugins that can work like legos to make a very simple game. AI can help get your initial game wired up. For the work of making it unique or interesting, AI can’t help as much. Though it could quickly spin up lots of graphics to choose from and such so that a person with no graphics skills could make their game have its own look. The other place it can help is in running tests. Like for new hardware that an engine or what not needs to support. It can even help add tests to some extent, but you still need a skilled person to look over what it did.
My understanding is that there are a lot of boring mundane tasks needed for maintaining the framework and such. The kind of thing that turns off opensource contributors. So maybe some of that can be offloaded and help get more people involved for free on a product that they can then use for free.
This is basically how modern development tools (I.E. the Unity Editor) work - they let you import all the resources then provide a framework for connecting it all together.
That said, this process of connecting everything is also one of the parts AI is actually worst at. As AI doesn’t understand context or logic, it can’t fit things together in a complex or meaningful way, nonetheless a unique way. Its for the same reason AI is bad at large/complex programming tasks (like game development). AI can make passable (albiet not great) individual art assets, but when you need to fit them together in logical ways, things start to fall apart. The same problem applies to testing. Tests where an agent effectively hits random buttons aren’t very useful, since they’re too inefficient. You need logical, structured and/directed testing, which AI can’t meaningfully do.
Basically, for easy, boilerplate stuff, its going to be largely done by the engine, or assets you import. Anything else is too complex or too important for AI.
I know the engines do that. But I mean even more. Like it could expand to the point that I could have a fully functional game just by adding some plugins. Not a great game. But something that was technically playable. I don’t think you can do that today. And as far as AI and testing. I am talking more mundane. Like say you already have good tests. But you just need to run them on additional hardware simulations or something. That is an edit it can do. Tweaking for that hardware and rerunning is also a pain. You could let it do that, then review the changed. That kind of stuff often requires digging through documentation and what not that can be very time consuming all for just one new peice of hardware. The AI can do that and iterate until things work. The dev just has to validate that the AI didn’t cheat to get a pass. For writing new test. Of course you can’t just tell it to make up tests. You have to dirdct it a lot. But once you tell it what you want it to test, it can write that, and interate on getting the test to work. It would be great if it was better at it than it is now. But I think over the next years it will get better. But I expect you still need to be the brains that tell it what to test. But also, if someone records a playthrough of a bug, you can probably tell it what the bug in the playthrough was, give it the playthrough and tell it to write a test. I’m not talking about expecting AI to do the real thinking. I am talking about it acting like a very junior dev that you can give well described tasks to, and like a reasearch assistant that can dig through large amounts of documention to get the information you need. Both of these things dev probably don’t generally enjoy doing.