Thursday, April 30, 2020

What is artificial life?

Thomas S. Ray

As I said in an earlier post, artificial life is a branch of computer engineering that builds programs that emulate the behavior of living beings: artificial living beings, or colonies of living beings, such as anthills or hives. Since I have worked in this field, I’ll tell here a little about artificial life.
In 1991, Thomas S. Ray built a program he called Tierra, where a series of artificial organisms evolved and competed for the available resources in the computer. These resources were essentially the computer memory, which was limited, and execution time. The objective of each individual was to copy itself into a piece of available memory. When copied, however, errors (mutations) could be introduced, so that the organisms in question were able to evolve.
The execution took place in a virtual machine equipped with a simple machine language, with 32 different instructions. The individuals were programs made of instructions written in the machine language. Some basic instructions were relatively complex, such as asking the operating system to allocate a certain space. Although very simple, the original program was able to copy itself (with mutations) in the allocated space. The execution of individuals is carried out in parallel, i.e. all are executed together, at the same time.
Since memory is limited and all individuals are trying to get it, it was necessary to include a death routine, which eliminates certain individuals, leaving free the space they occupied. It is clear that individuals with fewer instructions (who occupy less space) have an advantage over bulkier individuals, as they need less space to reproduce. The experiment started with a single individual, programmed by hand. After many generations amazing things happened, such as the following:

  • Appearance of smaller individuals (therefore more efficient) than the original individuals directly programmed by Ray.
  • Emergence of parasites, which use the program of another individual to copy themselves in the available space.
  • Emergence of super-parasites, which parasitize parasites.
  • Appearance of commensal species or symbionts, who collaborate to reproduce, by sharing their instructions.
A few years later, I decided to build a simplified version of Tierra to personally study its operation. My virtual machine, instead of 32 instructions, had only 13. The result was spectacular and surprising. Not only did parasites appear: a new type of normal individual protected against parasites also emerged shortly after, for when a parasite tried to reproduce using the instructions of its host, it just copied the instructions of the host. This provided an advantage to the host, who instead of reproducing once, did it two or more times in each generation. Therefore, after a certain number of generations, this species became dominant, to the point that the parasites disappeared, but then the dominant species ceased to be dominant, as it had lost its advantage, and the number of normal individuals sensitive to the parasites grew again, followed by the reappearance of parasites.
Another interesting detail is that one of the types of normal individuals that arose emulated sexual reproduction in some way, since it was capable of performing genetic recombination, as the new individual inherited part of his father's instructions, while the remainder came from another individual.
The relationship between parasite populations and normal individuals resulted in curves similar to the solution of the Volterra equations, which apply to predator-prey ecological systems. In the attached figure, the horizontal axis is time, and the number of individuals of each type in successive generations is shown; the black curve is the population of normal individuals, susceptible to parasites; the blue curve represents the population of parasites; the red curve corresponds to individuals protected against the action of parasites. Note how these individuals become dominant, but when the parasites disappear, the normal individuals grow again and the parasites reappear.
What is the use of this? Experimenting with biological evolution in this way is simpler and faster than using real living beings. Biological evolution is very slow. A new species of a multicellular being takes, as a rule, about one million years to appear. It is evident that we cannot wait so long to conduct an experiment. In artificial life, however, time is not a problem, for current computers are very fast and the time between successive generations can be very small. It is possible to perform experiments that in real life are beyond our reach.
Can the results obtained in the experiments of artificial life be directly extrapolated to biological evolution? It is doubtful. In the case of Tierra, evolution eventually stops, or becomes a repetition of previous situations, without increasing complexity. But when phenomena such as those explained in the previous paragraphs are observed, it is difficult not to think that perhaps all this means something...
In another post I’ll tell you about my experiments with virtual ant colonies.
The same post in Spanish
Thematic Thread on Synthetic and Artificial LifePrevious Next
Manuel Alfonseca

No comments:

Post a Comment