Hi all

After reading This article about the proposed future if open world games it occurred to me that there were a few issues with it from a technical standpoint.

The article talks about games that create themselves. Not just missions, or the world map, but even the art assets etc entirely created by AI and Google image trawling...

What is everyone else's opinion on this?

My opinion on this is that AI isn't there yet. The AI can't cook up a story, and make it believable, short of building it from a set of human constructed sections that will still require some very clever writing.

I think most people do play games for plot and wouldn't want a game where there is no story. If there is no story, people will get lost in the game and get bored.

With current AI the game would struggle to create even a simple side quest that didn't devolve into a simple fetch quest or assassination.

So, what do you think?...

Assumption that most people play games for the plot is just false. Just look Minecraft for example.

Assumption that most people play games for the plot is just false. Just look Minecraft for example.

You are right of course, however minecraft doesn't make its own enemy types, it's own graphics and textures or try and make missions at all... I think that's more what I'm getting at...

There was a guy who explained the issue very well. I'm going to try to repeat it what he said:

The problem with procedural content is not that it's unlimited. Take a kaleidoscope for example. There are billions of possible combinations, each one unique. However after 5 minutes of playing with it, we get bored. It's not because we've seen all combinations. We now know how it works, and its limits within the unlimited permutation options it offers. It's because we've grasped the concept.

The AI is still the hurdle. Sure you can visually process terrain/props visualization but what of the active content of the game?

For the behavior of the inhabitants will you need AI that can watch and decypher movies to see how things move /act/interact ? And you will still need alot of manhours to vet what it comes up with as far as logic for the auto-generation.

Look at the state of game AI, which is still quite limited because of the huge cost and effort it involves (and the processing power that is needed to do it in real time).

I agree with the OP. I think the writer is mistaken. I think complete procedural generation would be possible, but in the far-flung future. Not near future. Then- it's kind of the realm of sci-fi rather than reality-based theory.

I think the author's mistake is assuming that since now procedural terrains, planets, and universes are being made, therefore similar algorithms can be applied to other areas, which just isn't true. All of that stuff is based on noise-generators with lots and lots of tweaking, which in the wide-scope of tech, are fairly simplistic.

In theory, it might be possible to create a narrow AI to scan photos from the internet to decide what a "rock" or "human" or topic looks like and then generate something, but it would take time and research on the same level as Honda's Asimo to get useable assets.

There are a lot of other procedural technologies that already exist - animation, texturing, procedural composite assets, etc, but the the tech gap between now and full-procedural games is fairly massive.

What happens when the AI stumbles across a pic of Dwayne "The Rock" Johnson? lol Do we get characters that look like rocks, or rocks that look like Dwayne?

I won't live to see an error free version come to fruition.

Not that I have huge experience in the field, but here it goes.

In my opinion for a creation of a game that seems believable by humans, it must be created by humans (Through some medium) or by something that has a remarkable similar thought pattern and same level of understanding of basic human emotions, lives, interests, etc... Now, in my ears this all hints a massive combination of various Neural Networks, as it is by far the closest we've come to simulating the behavior of a normal human brain (I'm not saying that it's anywhere near! And the neural networked model has various of issues!). Let's assume that one day that we come to be able to recreate the neural architecture of a "brain", and we actually manage to archive the same level of functionality. The issue would in my opinion be the amount of data needed to train this neural network in order for it to understand "us". How would this being learn? Through videos, images, text? To experience what we experience, the being must partake in what we do. And after 20 (Don't take this value seriously, just to make a point) years of solid learning, it could feel, laugh, understand, emphasize, etc... But, hold on, now you've basically created another human? What now? Copy these beings by 100, 200, 300? Do you have the right? You see where I'm going...

Not that our current computational models aren't fast, but (in my opinion) people always misjudge the sheer size of the typical neural network in an adult brain, it's massive!

People can make good games because they have full access to approximately how a persons brain works (their own). For an AI to even in theory be able to make comparable games, we must first digitally encode essentially an entire person.

Maybe google can datamine that out of searches, but that would get us very mainstream games...

It should be possible to make a story writing AI. People have been writing stories for a LONG time, so Im sure there are enough systems and approaches to writing stories that a computer could write one (if you just collect enough of those in one place). Of course it would be limited to a subset of all possible stories, but people seem to be fine watching tv series that follow the same pattern over and over again...