"Humans partnered up with bots to come up with new levels in Super Mario, a popular Nintendo platform game."
Super Mario is a game? Never heard of it. Is it a popular game on the Nintendo platform or a popular platform game by Nintendo?
AI can get pretty good at creating content like images and videos, so researchers are trying to get them to design game levels. Machines are okay working on their own and can regurgitate the same material seen in the numerous training examples fed by its human creators. It's fine if what you're after is more of the same thing …
I resent the fact that the researchers seem to consider that making a game level is just placing blocks and sprites like you sprinkle powdered sugar on a cake.
Making a game level requires at least two things : having a goal in mind, and planning out how you intend to reach that goal. Statistical analysis machines fail on both of these counts.
A good game level is made knowing the game mechanics, to be able to create areas that are of a specific challenge. Well placed, such areas can be simultaneously very desireable and very difficult to hold or go through. It is that combination of hard and yet unable to resist that crafts the best levels.
No computer will be able to attain that level of craft just by looking at existing maps because no computer can understand either the word "difficulty" or the word "desire".
But hey, they got grant money, so it's all good.
Procedural generation cannot hope to create "interesting" content, with or without "AI" (as we currently know it). And I'm talking in a broad sense here, about game levels / environments in general, not just platform games. Yes, if all you care about is exercising the mechanics of the game then more or less any collection of "platforms" will do and will give you something to jump around on. But the interesting and memorable levels exploit some specific aspect / synergy of the way the game's mechanics fit together that goes beyond blindly rewarding ability to press "jump" at the exactly right millisecond.
More generally, complex game environments beyond a mere set of platform or pipe tiles become interesting when copied directly from (or modelled after) some part of the real world because everything there is full of history and purpose - everything is where it is because somebody built or put it there at some point, and that is the very essence of what makes that environment believable and interesting. Even if you're just trekking through a landscape full of rubble, all of it is where it is because it used to be part of a building or structure in the past, then it took a hit from something at some point or just crumbled on its own, and all that brick, concrete and twisted metal came to rest where you see it today - that is still purpose, that is still history.
Even when such environments are wholly artificial and man-made, the ones that manage to stay "interesting" incorporate all that same, albeit fake history and purpose - because the designer went to considerable lengths of inventing a back-story, imagining _why_ their scene looks the way it does based on who is expected to have shaped it according to the lore. Slowly discovering that history - or just immersing oneself in its atmosphere - is what makes a well-designed environment attractive and a generic procedural one lifeless and boring.
Yet all of that is very obviously missing from environments designed by a machine throwing dice and following a set of rules. It's the difference between "that portrait hangs where it does because the lonely occupant of this room liked to gaze at it from his bed" and "human bedrooms tend to include an average of 1.69 wall-affixed pictures". And there's no way in hell any kind of machine, AI-assisted or not, is going to be able to create that kind of environment until they themselves experience all relevant aspects of human existence...
At least GOFAI symbol-processing based machines, possibly with genetic algorithms to get some creativity in had a fighting chance to build something of moderate interest in a formulaic way, but CNNs?
Next up: outysourcy subcontinent coders who deliver ingenious, quality-assured code that totally blows away what could be developed in-house.
Biting the hand that feeds IT © 1998–2021