Can We Create Fun Games Automatically?

togelius writes "What makes games fun? Some (e.g. Raph Koster) claim that fun is learning — fun games are those which are easy to learn, but hard to master, with a long and smooth learning curve. I think we can create fun game rules automatically through measuring their learnability. In a recent experiment, we do this using evolutionary computation, and create some simple Pacman-like new games completely without human intervention! Perhaps this has a future in game design? The academic paper (PDF) is available as well."

Different "fun" for people

[troll8901]

Good point. Different levels of "fun" and satisfaction.

Someone wrote about putting Age of Empires 2 on showroom PCs, and all the female customers went ga-ga over this game. They would then build mini cities and so on ... all without fighting. He said they wouldn't give a second look at AoE 3, or The Sims 2 ... they just wanted to play AoE 2.

Someone wrote about his entire family playing mostly older games [slashdot.org] (including all Mario games), and mostly avoiding newer, copy-protected games.

It amazes me reading these posts.

Work on Hollywood movies?

[olddotter]

Seeing movies produced by following the "formula", do you want automated games? Do you even want a "formula" for "fun" game design?

Maybe its possible, but this starts to sound like automated art.

Re:Different "fun" for people

[jtogel]

This is a good point. In fact, there is research on identifying different player "stereotypes", and having ways of automatically identifying what stereotype a player belongs to could enable us to automatically create games for particular players. Or just adapt a given game so it suits some player better.

Re:More to the point

[jtogel]

Alright, I'll try that some day. You know, it's not always that easy to get research funding through trying to be original and relevant, so maybe your method is better.

Re:I think the research oversimplifies

[Yvanhoe]

I don't think that this headline defined the problem well. Yes, some parts of fun can be automatically generated. But no, to make a fun game, it has to be interesting to a human, not just to a turing machine. And for that, you really need other humans to make the games, or you don't have the depth required for real "fun".

Why I disagree on the fact that the automatically generatable parts of fun are not enough to make a human-enjoyable game, I don't really have more counter arguments than there are arguments. That would make for an enjoyable Turing test. My only counter-argument is that I know of quite a few games which do not depend on depth to be fun.

game programming would be like photography

[tacitdynamite]

In photography, you set up the boundary conditions, take a TON of pictures, then select the best ones from the ones you have. The best photographers have the best eye for selecting the remarkable ones out of the pack. This would shift game programming from an art like classical sculpture - where you have to plan far, far ahead, and don't get second chances - to an art like photography where it is more about creative curation than creative engineering. Evoluationary development of games wouldn't eliminate the creativity of the process or the product, it would change the creativity of the process and the product.

Re:I think the research oversimplifies

[dontPanik]

Yes, some parts of fun can be automatically generated. But no, to make a fun game, it has to be interesting to a human, not just to a turing machine. And for that, you really need other humans to make the games, or you don't have the depth required for real "fun".

But the things you've outlined, the setting of a game, the feel of a game, and the idea of a game are intentionally not touched by the research. In the games created by the research, very generic names are given to the different objects comprising the game, so that these variables of fun (the setting and the ideas behind the game mechanics) are left out of the equation. With those variables eliminated from the research, the focus is only on the difficulty of the game and the height of the learning curve. That is the "fun" that is being created automatically.
IMO this technology is not ready yet for serious usage, but at this time, the concept protrayed in the study could be useful for those creating games.

Re:So Yankish...

[russotto]

From the Internet bubble to the housing bubble, it's all been "let me have it all without having to work."

No, it's been "let me have it all with some simple work at the beginning, and a smoothly increasing amount of work appropriate to my increasing skills as time goes on".

Although this totally fails to explain Nethack, which is easy to learn but has more of a difficulty cliff than a difficulty ramp...

Defining trickiness

[Twinbee]

Fun for me in games usually means there's always something to do or press. The old 2D games were more like this, but a large reason why I hate modern 3D games, is that there's often lots of sprawling around without really doing much (partially related to the 3D world, but it can be solved with difficulty).

I like the idea of how the article mentions that the algorithm biases towards games which can't just be won randomly. The board game is Go is the ultimate example of this I guess, where there are many *levels* of mastery.

But one has to be careful with this approach. If in a 3D game there's a small opening in (say) a castle wall, and miles around of plain grass, it's pretty easy to solve for a human player, despite the huge searchscape and 'narrow' solution that a computer would find tricky (which would apparently potentially rate as a good 'puzzle').

At the very least, developing models for other human factors such as reaction time, subtlety of graphic elements, and the challenge of pressing certain key combinations, would also be needed before final game automation could be achieved.

Re:game programming would be like photography

[YouWantFriesWithThat]

that model of photography works in the digital age when there is no cost to each shot and thousands can be taken. or if you have a large budget when using film. however, some of the best photographers have done their work when poor, with limited resources to shoot and develop large amounts of pictures.

i have been along for a shoot with truly amazing photographers that use film. one in particular that i knew had 22 keepers on a roll of 24. about 18 of those ended up in a show. an anecdote and a limited sample, to be sure. but i think that the best photographers have a good eye for composition and can see what a subject is going to look like through a lens without ever lifting the camera.

i am not discounting what you say as being valid for some (maybe most, i don't know) photographers. and the point you are raising through your comparison is valid and interesting. i just wanted to clarify that in my experience not all photographers are promiscuous about taking pictures, and photography is not necessarily curation at it's core.

This was attempted in a Shmup

[Sigma 7]

Evolution SHMUP: http://www.kloonigames.com/blog/games/evolution-shmup [kloonigames.com]

This was an experimental game, where the theory was to see how long a player would remain in a given game - as people continued playing, the system adapted gradually in order to maximize the fun value (in this case, the amount of time spent on a single game.)

This experiment has a smaller search space than the article, but isn't generating any "successful" games. This may be caused by the environment(i.e. the evolution scope is too narrow and thus isn't generating a variety of enemies), but the same problem can easily apply to the article in question.