For anyone who really knows me, they know that deep down inside my greatest dream is to work in the true artificial life game genera and bring it back from the dead (it died around 2003) and into a new golden era. Yet I have always known it would be a long time coming and I would need to work my way up in the ranks before I ever got the okay to start such a project...Why you ask? Read on...
Why did "true" (Good Dog) AI die?
Well, it's the black box syndrome. Whenever you have AI that partiality creates itself, the designers lose control. Game companies consider the concept a QA nightmare and avoid it like the plague... so much so in fact that they have stopped using many of the academic AI tools in places where, really, they should still be using them and wouldn't really be at risk for a QA explosion. Case in point, digital DNA that controls a color tint in ambient organisms or monsters. A few locks to prevent over saturation, a background breeding master based on genetic drift or even a little selection if it's monsters, and you instantly have something that makes every play through of the game a little bit special.
What replaced Good Dog AI?
Good Dad AI replaced Good Dog AI. I'm borrowing this analogy from Brian Schwab, a very talented AI programmer I met at GDC this year. Good Dad AI is like a dad wrestling with his 5 year old. Sure, he could easily crush the kid, but he doesn't. He also can't just let the kid win because the kid will call him on it. A good dad must lose with style, in an interesting way that makes the kid feel like they have accomplished something. That is what the goal of Good Dad AI is, and it does this primarily through heuristics (rule sets). It's a completely valid form of AI, is a better choice in most game scenarios than something such as a neural network and is no less "real" a form of AI. However, it can't really grow and develop or feel or think. It is what it is, a rule set, a behavior tree, a flow of logic and nothing more.
Why did Good Dad AI Replace Good Dog AI?
In order to be sure the game offers the experience the designers intended, by default the AI must be a good dad. If the AI builds itself there is no guarantee it will be a good dad, it could be the worst dad ever! That is why heuristic approaches have taken over, even if they become dry, and predictable with enough encounters.
Why are heuristics still real AI?
Because they are, in fact, very real in actual biology. Take the orb weaver spider and it's classic web form. That web is built and maintained using a logic set, and each species has it's own mathematical formula for producing it. What controls this rule set is the spider's DNA, and mutations coupled with selective forces can change that formula by changing the DNA. This is the piece that has been removed in Good Dad AI, and for good reason. Why? Because the player could train the spiders to be idiots and leave gaping holes in their webs by selectively killing them off... the control of the developers is lost and we are back to that black box issue, but I digress. The truth is, both heuristics and adaptive AI forms are part of a greater whole. It's not that one is better than the other, but rather, to favor one is to forget half the picture.
So what is with this Good Dad vs Good Dog thing?
They are not so much different forms of AI as they are different goals for the AI. Good Dad is consistent, Good Dog is only as good as it's owner. Naturally, they take different tools to produce, but the sad part comes in when fear prevents the best tools from being made available... plus having real live creatures in a digital environment is cool. =)
What exactly is the "true artificial life" genera?
It's a sub-category of simulation games with a different sort of goal from the more common pet simulation games. Rather than present the player with the illusion of life, these games actually attempt to reproduce it using scientific discoveries in the ways actual life functions. The creatures within these games have very little hard coding and essentially build themselves within a virtual world. These sorts of games are very complex and use digital DNA, neural networks, simulated biochemistry, heuristics and many other forms of AI, all at once. Over the top of this complicated structure is the user interface, which will look very much like a pet sim game, but it won't take long to feel the difference between the two.
In pet games, pets are not born with lethal mutation, but they are in true artificial life games. In pet games, pets do not develop bizarre behavior patterns that make no sense for their dog or cat exterior, but in true AL, that can happen. In pet games, pets do not mutate into a more intelligent form capable of more complex behaviors than previously seen, but in true AL that can happen too. Some people see these differences as flaws in the true AI genera, but they are not flaws, they are what make the genera so amazing. Sure, you can have really bad, weird, buggy looking things happen, but that is part of the fun! This is especially true if the player can access information on the "malfunctioning" creature and find out why it's doing such strange things, or at least try to.
What was the original point of this post again?
The point is, the man who first started the genera back in the 90's, Steve Grand, is bringing it back with a fully 3D AL game built in my engine of choice, Unity! The timing is amazing, and this may be the opportunity I have been working to prepare for for so long. To have the chance to help build that game would be... indescribably awesome.
It's even more amazing that he chose the same engine I have been eying for rebuilding the genera in. I even have the beginnings of code for handling the complex environments that are needed to grow true AL in started.
Granted, I have to first somehow convince him to hire me, but then again... I was made for the job. Even if he doesn't hire me, at least the the dream is moving forward, and perhaps, in time I will get my chance at contributing to the 2020 vision after all.
For people who have money and want to help, he is raising fund here:
http://www.kickstarter.com/projects/1508284443/grandroids-real-artificial-life-on-your-pc/comments
No comments:
Post a Comment