Artificial Life – some preliminary findings
I fnished my artificial life simulator over the weekend, and I have been running simulations on it. There were a few bugs initially that I quickly fixed. My initial simulations gave me some inexplicable “population explosion”. For example, in one cycle I would have 31 creatures and in the next cycle I would suddenly have around 9,000. It didn’t make very much sense until I analyzed the log files.
If any of you remember, I mentioned earlier that my “creatures” are nothing more than assembly programs. Each of these instructions (and parameters) translate into a bit pattern. Changes in the bit patterns lead to changes in the behaviour of the creature.
One of the instructions is a “reproduce instruction”. This prompts the creature to create a copy of itself, with possible mutations. The instruction has a parameter that tells the creature how many copies to create. During the course of the simulation, certain creatures evolved with the reproduce instruction as their first instruction.
My supervisor program (God program) runs one instruction of every creature during one cycle (sort of like an pre-emptive multitasking operating system). If a creature happens to reproduce during the current cycle, then the supervisor program will execute the child’s instructions as well in that current cycle. This led to an interesting situation. When the mutant that had the reproduce instruction as the first instruction reproduced, its child (if it didn’t have a mutation that changed the instruction) also had the reproduce instruction as the first instruction. So the supervisor program would run the child, and the child would reproduce with a similar offspring. This went on until a mutant that did not have a “reproduce instruction” as the first instruction, evolved. As you can see, this is what led to the population explosion. I corrected this problem by adding “sexual maturity” and “reproductive energy threshold” parameters to the creatures. Basically, a creature can reproduce only if it has been in existence for a certain number of cycles, and if its current energy is above its reproductive energy threshold. This got rid of the population explosion problem.
My simulations gave me some other interesting results. One of the instructions that I have is a MOV instruction. This instruction moves a creature from one cell in the two-dimensional array to another cell. There is a variant of this instruction, known as the MOVA instruction. In the MOV instruction, if the creature tries to move to an occupied cell, it will retreat to its original position. However, in the MOVA instruction the creature will try to kill (and subsequently eat) the occupant if the destination cell is occupied. I had created a creature that simply ate, reproduced, and moved to different cells. I noted that after many generations, this creature had evolved into an aggressive one, that would eat, reproduce, and move aggressively (MOVA) into other cells. That was rather interesting!
I also saw natural selection at work. When I corrected my population explosion problem, I made it so that the reproductive threshold of a creature was higher than its initial (starting) energy. Over the course of many cycles, I saw that the creatures evolved such that the initial energy would be very high, and that their reproductive threshold would be very low. In additon, their reproductive age was lowered. I thought that my creatures were having it too easy, so I modified my code to make reproduction a costly instruction. After running my simulation, I noticed the opposite! Now the creatures evolved such that the reproductive energy threshold would be much higher than their initial energy. It would seem that the creatures wanted to build up enough energy to where they would be able to reproduce successfuly.
There is also one very interesting behaviour that I noticed. I provide branch instructions in my instruction set. A creature can branch to any part of its code. In addition, I also have a Decrement and Branch instruction that is useful for iterative loops. One of my creatures evolved a clever strategy. It would decrement and branch into the middle of the branch instruction. When I analyzed the code, I noticed in doing so, the creature was performing a MOVA instruction in fewer bytes than writing an explicit MOVA instruction. It also had the added benefit of performing the MOVA instruction in an iterative fashion so that it could travel around the two-dimensional array. There was a reason for this kind of behaviour though. A creature consumes energy when it performs an instruction, and the energy consumption is proportional to its code size, so it is amazing, but not surprising that such behaviour could arise.
There is still more work that I need to do. I am thinking of giving my creatures some “memory” in the form of stack and RAM. They will have some Store, Load, Push and Pop instructions whereby they can write to and read from memory. Also, I am thinking of adding some more “registers”. Right now, the creatures have two count registers for performing iterative loops. My reason for adding more is this – I want the creatures to be able to act on the data in memory. For example, I can provide instructions that allow the creature to check the status of a particular cell, like if it is occupied, how much food it has, and so on. The creature can load data into the registers, and then use these registers as arguments for the cell checking instructions. This way I am hoping that my creatures will be able to evolve some rudimentary form of “intelligence”, or “intelligent” behavior. This will also increase my instruction set. Right now, if a mutation results in an invalid instruction, the creature dies when it tries to execute it. Actually, that reminds me – I had creatures that evolved with invalid instructions, but they had also evolved branch instructions that would jump over the invalid section of code!
Most of these complex behaviours aren’t that surprising when you consider natural selection… it is only err… natural that such behaviour should arise!
Well I guess that’s all I have for now. I might make a separate page for this project where I can describe this project in more detail. I’ll post more updates here after I run more simulations.