Creatures Caves | Community

An article written by Andrew Carroll about how the biology of Creatures models real biology.

Andrew Carroll has been very busy lately with exciting projects such as a new world with genetics in its plants so we were surprised and very grateful when he has the time to write us this excellent article on Creatures, his views and its links to Biology. Andrew being a biologist he knows his stuff on both sides of the border, IT and Biology!

When Miff first asked me to write an article about the genetics and such in Creatures, I thought to myself, "Good God, now I have to come up with a funny introduction that gives a brief description of how I wrote the article." After carefully studying key points of various articles written by Lis, Steve, and Matt I decided it would be best to just ramble until I said something meaningful. In any case, now I need to transition to Creatures. Creatures is the kind of game that invokes two feelings in a biologist: fascination, and panic. At a first glance, Creatures looks quite a bit like the Petz games. The resemblance ends at that first glimpse. The technology behind Creatures is nothing short of revolutionary. It's hard to appreciate the enormity of what is going on behind this game, so it might be best to go over what exactly Creatures is based on- real life. I am siting here typing this right now, a huge mass of organic material creating an intelligent (hopefully) thought. It might not amaze you; you see it every day. Instead, look at it from the perspective of a single strand of DNA, namely yours. Wrapped tightly into chromosomes, this long spiral staircase creates you. Inside this microscopic chain is all the information it takes to form you. It would take a computer trillions of times bigger than this to store all that information. The wonder that this tiny strand can create an intelligent person is practically unfathomable. If you'd ask any biologist who knew what DNA could do and ask them if they thought that DNA could create such an organism, I'd bet the vast majority would say no, if it hadn't already happened. Life tears down barriers. Give natural selection some time and it will find the most efficient means of coming up with a solution. This is the feature which Creatures tries to replicate.

Now let's take a look at computers. Suppose you want to write a cob, and you know exactly what you want it to do. So you bang out some CAOS and your cob does X, Y, and Z. You have just use a top-down approach to solve your problem. A top-down approach requires no creativity whatsoever to complete, because the idea has already been developed, and the means to complete it are readily available. But, suppose you wrote your cob for a world with high levels of organic nutrients and most of the community has a world with low levels of organic nutrients. Worse still, there are those with medium levels, and lower and upper medium levels. So now your cob needs to do A-Z. You have 3 options. Take out your cob's use of organic nutrients, specifically program your cob to deal with each varying level of organic nutrients, or make your cob adapt to the world itself. Clearly option one is not ideal. So, you can make 255 additional slightly different copies of your cob for each varying world, or get your cob to adapt. So you decide to get your cob to adapt by modeling evolution, giving it a very simple genetic code and letting natural selection take care of the rest. Viola, you've just used a bottom-up method to solve your problem. It takes up a lot less space, a lot less time, and also solves the problem of different regions of the world having different nutrient levels.

What you've just done is to a lesser extent exactly what Cyberlife has done with the norns, ettins, and grendels of the Creatures series. They don't know how to make a thinking machine, so they use the same kind of thinking to select one out. Sounds good in theory, but how have they done?

Maybe a look at the what CL (Cyberlife) is trying to do in general. CL has taken all the information they know about how real living systems work, and attempted to model it in a computer system. This unique approach Let's start our specific analysis with the genetics. This is very well done; obviously, it is simply impossible given current system resources to model the long process of transcription into proteins. To get around this, CL ingeniously goes to the heart of the matter. A gene represents a single trait, so CL made each gene code for a single trait as directly as possible. This wins points with me in terms of providing the most functionality for the processor power. In the future maybe CL will model transcription better, but the gain from this would be negligible.

Next we'll go onto the brain. This is a mixed bag. While Cyberlife has done a good job in generating decisions and perceptions, but memories are quite limited. To minimize requirements, Cyberlife has stripped their simulated creatures of true memory. First off, they stripped down an experience to the bare minimum. An action is merely linked to an emotion or object; in effect, your norns have no memory. This is by far the greatest limitation for a creature. Memory is a requisite for self-consciousness, and though intelligence can be generated without it, self-consciousness is improbable at best. How can this be fixed? I don't know, but a way around definitely needs to be found. Perhaps giving a creature access to the raw data of a situation and letting it interpret it would be best, even if excruciatingly difficult. Another big failing I see with the norn brain is in its inability to multi-task. A creature can make one decision to reduce a drive, but cannot formulate a chain of actions to solve a complex problem. Norns et al really only have one thing on their minds, their most pressing drive. Multi-tasking is essential to the survival of all moderately complex life forms. Humans, for example, can type at a computer while thinking about a chess strategy while their unconscious analyzes their last date. Multi-tasking leads to preparation for a future event, the basis for imagination and consciousness. I don't blame CL for not including it, if there's anything difficult to provide, it's imagination. Since there's basically no way of knowing how to create such a system, I would say that the best thing to do would be to add the ability for "parallel processing." Freudian psychologists have always proposed that the brain has multiple layers, each with its own intelligence. Many biologists (at least the ones I know) reject this notion, but I think there is at least a shred of truth in it. To get complex behavior, a brain needs to constantly exploring more than one option. Your lowest level might tell you to satisfy your hunger by grabbing an apple and eating it. Another glob of brain might be considering how much you should spend on food if you want to get CA. Another might be contemplating how to advance in your job, and another might be weighing the consequences of each decision that pops out of each glob. Each idea created by each layer creates a collection of thoughts, perhaps giving the illusion of a single, very capable brain when in fact different subsets of that brain have each been making their small contribution. I don't think each glob is permanent, just that at times neighboring brain cells work without collaborating entirely with all of the brain. Instead of a master painter creating a canvas, a horde of painters each paint a tiny corner, and in the end a great piece of artwork results. So maybe the solution is to give each norn a layered brain, and let each layer have a different function; to formulate the best way to avoid grendels as well as the best way to alleviate the most pressing drive. Cannies and Sub-Terras are a step in the right direction in terms of this.

Now we'll delve into how CL has simulated emotions. The answer is simple: dang good. CL has modeled emotions almost perfectly to what we know about them in real life. The real question is do the creatures actually *feel* the emotions, do they actually seek to avoid discomfort or is the neurological reactions in their brains automatic. That's one for the philosophers, but I would say that when given a true memory, creatures will have some form of self-awareness.

If you've read this far, you're a brave soul. I realize really long pieces are a pain to actually read and not just skim through for the good parts. In any case, I think we'll go into chemical reactions, concentrations, etc. This is somewhat disappointing, for though the models themselves are decent, the flexibility they provide is not. Only allowing up to two agents to react to provide up to two products is not the best way to solve this. It might make sense if you wanted to demonstrate how to balance in half-steps a reaction between an acid and a base, but for a more complete artificial life program it is not ideal. In order to get a good simulation of reactions multiple half-step reactions have to occur. This not only slows down a complex reaction that needs to occur quickly, it also hurts the stability of the genome. Because there are a greater number of reactions required to complete a complex reaction, there is a greater chance a mutation will happen to one of those half steps and throw the entire chain off. What you have done is ruined 5 or 6 genes for one mutation. It would be better to expand the reaction capacity so that when a mutation did occur, it would leave an intact gene sequence that would mutate together toward something, or be easily dropped from the gene pool entirely. In addition, fewer genes with more information mean CL's organ "cost" system will be far more accurate. The same complaint holds true for emitters and receptors, it would definitely be better if more than one input or output was allowed. This is a minor CL blunder, which can be gotten around, but it can have a serious impact on the functionality of third party engineered species, and it impairs the speed of evolution in long running creatures worlds. Ah, the executable. The most limiting factor on the evolution of intelligence of creatures, especially in C2. The executable governs far more than it should, setting limits on the number of lobes might help game stability in C2, but it slows brain evolution to a crawl and limits third party developers. Also particularly vexing is the way it has a universal effect on plants. Even if a plant does not depend on nutrients, the executable has a tendency to re-class it as detritus if the organic or inorganic nutrients in a room hit 0. This is extremely bothersome for those who want to go beyond CL's climate model and use something different. If there's any one feature that needs reduction, it's the executables over-reaching authority on the attributes of creatures (especially C2). Just as a tree will not grow in the shadow of a concrete skyscraper, the blossoming of the AL in creatures will be limited by the executable. Giving such authority to a specifically programmed device is a step in the wrong direction. CL obviously put a great deal of time into the executable, it now needs to spend even more time finding a way to get those features into creatures expansions without making them set in stone.

A prime example of a powerful potential neutered by the executable is the C2 climate system. This system could have been an enormously flexible tool for third party developers to expand on, adding their own features for a continually improved game. Instead, the executable's approach to management makes it necessary to create an entirely new system to incorporate such potential, and even then the executable still likes to kill of plants because its system is not used. The climate system itself, though, is fairly well done. Extensive enough to cover most of what cobs need to know, even if it might be insufficient for some. The main problem is that the system is not used enough. Plants are pretty good at utilizing organic and inorganic nutrients, but the real strength of a simulated climate would be to use all the attributes collaboratively to get genetic objects to function best. Overall, the weather is a decent system, even if you can have a conversation about what is wrong with it. Of course, I'm referring to C2; C1 barely had a weather system.

A simulated environment can only be as good as the creatures who inhabit it, and like the climate system they utilize, the plants and animals of creatures are a mixed bag as well. Creatures one creatures couldn't be expected to do too much other than their interactions with norns, grendels, etc; maybe a pollination here and there, but nothing too complex, and they did their duty very well. But the inclusion of a climate system in C2 upped the ante. Now creatures could not only contribute to the emergence of the AL in the world, they could also be used to form complex systems to enhance the adaptability of the game and make a better world. CL did not do much of this. Most of my carrots and tomatoes die with an hour, my goldfish overpopulate the seas, snails and worms don't really breed. There is limited use of room attributes, but it is limited. There is only one object, a cactus, with genetics; and the food web is more of a thread. The potential was there to set up an intricate evolving world, and I think CL dropped the ball on that.

Creatures 1 is a fabulous game, and Creatures 2 with third party cobs is utterly unbeatable (before C3 that is) and the biology of the creatures is well done, but it isn't done enough. If the goal of artificial life is to create artificial life, there needs to be more simulation and less preprogramming. If, however, the goal is to sell an entertainment product, this is just fine as it is. I think, however, that a great deal of the interest generated by Creatures stems from the genuine rudimentary intelligence there seems to be, and it would be a mistake to turn away from simulation and science and towards entertainment. Creatures isn't Petz; and the continued simulation of biological principles will make sure that we will one day discuss philosophy with a computer, not have a small game with squirt guns.

Originally published on HomeCreatures.