A full game made from AI without an engine. Tim Sweeney is not amused.
You're viewing a single comment thread. View all comments, or full comment thread.
Comments (43)
sorted by:
AI can not build video games. Not really.
I'm one of those people that see video games as, at least partially, art. The concept of a game is first to construct a rewarding gameplay loop, followed by the art of actually building the aesthetics of the world that contains your loop.
AI generation to make a "game", rendering it from scratch, as you play it, is not even really a game. There's not even really a good way of building out a world. One of the unique features of a game is that it allows people to have both similar and different experiences at different times, as the result of interacting with the core gameplay loop. A "full game" where the AI builds things as a human interacts with it, will inevitably fail because no AI can match the ingenuity of a human, and the AI isn't constraining itself within a core gameplay loop.
The best that AI can actually do is what one guy already has done. I can't remember the name of the game, but a single person is building a heavily AI built Space Sim game like Elite Dangerous or Star Citizen. Lot's of AI generated assets and content around the game he actually built. All it's doing is reduce his personal workload, and make the game cheaper. The core game is still developed by a person.
This is one of the reasons I'm saying that AI is currently a bubble. Most of what is claimed to be AI, literally isn't anything beyond basic algorithm work, or is being badly misapplied to do things that don't make sense. Just like with the Dot Com Bubble, investors are dumping shitloads of money into gimmicks and scams.
Interesting. Not sure I agree with that but it does sound like the opinion a lot of the gamers I see on forums. I don't think some of the old masters of game design were thinking about "gameplay loops" in the past. It's more the work of relatively modern post-mortems trying to make the art of fun into rigorous theory. My only rule for a game is interactivity. Walking simulators may be poor examples of entertainment, but they are still games. Maybe AI could only make shitty games.
Them not using the specific term for it doesn't mean it wasn't part of their development process.
Mario 64 had a gameplay loop of "enter painting level, get star, get kicked out of level, repeat until unlocking next area." The loop was clearly noted because its one of the major changes Odyssey would do away with later after many iterations of playing with it. Heck, handhelds as an entire system were built with something resembling a "gameplay loop" in mind because most of them were designed to be played in short chunks instead of longer sit down sessions.
A gameplay loop is just the most distilled down bones of what the game wants you to do with the mechanics it presents you.
The modern issue is that the "loop" is now designed less in the interest of fun or soul, and more to create addiction via skinner boxes and other nonsense. Cart before the horse design, which is why it has such a negative connotation. An AI can create a loop the same as anyone, but because its designing it in that cart before horse manner it'll usually be quite shitty.
It's a problem of "desconstructing" something in an attempt to define it in simple terms, which end up being entirely alien to the original intent. I'm fairly certain this is the reason that AI feeding upon it's own output produces progressively worse products. It may very well be that, by the constraints of programming, such loops do emerge within a game's design. Yet if you try and focus on such things alone you end up with progressively worse results.
That's inevitable when things go from "a bunch of dudes from various backgrounds come together to create a passion project" to a literal industry with college classes, billion dollar corporations with management giving powerpoint slides, and an entire legion of "game dev" video series on Youtube.
Things that everyone just understood now need a name so they can be compartmentalized and taught separate from the whole so they people who lack the extreme passion and talent of the old masters can still understand and create from it by eventually assembling those pieces.
Its not necessarily a good thing, but that's just a problem with our education system in general. Like, for those of us who were just capable of math from the outset the systems upon systems that teachers use to teach people who cannot do math the ability to do basic shit look completely alien and retarded. But its probably the only way they can grasp it, regardless of how slower and soulless it is.
I disagree. Gameplay loops come from programming in what it is a very simple and early game called: "The Game of Life" by John Conway.
The whole point of The Game of Life, is you are given an arena, a set of basic rules, and you may choose a starting position. The results will then vary and generate different experiences with your input. This is the interactivity of which you speak. The gameplay loop is how you interact with the program, it is the rule set.
The reason The Game of Life is useful as an example here is because it is a "zero-player" game, with no objectives, and yet is fulfills the basic tenets of what a game must have to be a 'game' or specifically what we think of as a "video game". You're not even competing against anything, you are simply interacting with the rule-set. The core gameplay loop is the rule-set, and the passage of time.
Games do not even necessarily need to be fun, but they have to cause some kind of human emotion. In this case, it's curiosity. And the level of detail within this simple rule-set that humans can input is shockingly large.
Even the earliest game that is known to exist has a similar function. The Royal Game of Ur it has 2 players competing against each other to get to the end of the board. It uses dice to add random chance for movement, and betting to add additional risk and reward. Despite this game emerging from the the earliest human civilization to ever exist as far as we are aware, once you have the rules, it is actually a quite entertaining and competitive game that combines both chance and strategy. Again, the basic rule set is the environment you are operating in, and the loop is each each iteration of positioning against your opponent.
You could create an AI that could follow some basic mathematical equation for what a good game should entail, but the entertainment value in the aesthetic and variations still need to come from humans.