We have been looking at Christoph Adami’s book, The Evolution of Biological Information. We are presently going through Chapter 4, “Experiments in Evolution.”We come now to a section that covers the history of artificial life. Artificial life, or ALife, refers to computer models that simulate worlds that include Darwinian evolution.
After he discusses earlier theoretical models, Adami’s first example of actually running artificial life is a program called VENUS. This model was inspired by the computer game Core War. The goal of Core War is to write a small computer program that can crash other programs while avoiding being crashed itself. VENUS takes this idea and adds mutations so that there are random changes to the programs.
Adami says of the model:
Rather than evolving, the programs were bent on Armageddon. Rasmussen witnessed the programs over-writing each other and effectively destroying each other’s codes. It was digital carnage; only the small snippets of code managed to escape being over-written, but there were effectively rendered too short to self-replicate. Life on VENUS … was doomed.
A Variety of Clever Strategies
But why was life on VENUS doomed? VENUS, as I said, was based on Core War, a game that has been played since 1984 with a wide variety of clever strategies being invented for it. These strategies have been given names like replicators, scanners, bombers, pit-trappers, imps, hydras, quickscanners, bootstrappers, and core clearers. This suggests that similar sophisticated strategies to avoid Armageddon existed for VENUS. But they did not arise by any sort of evolutionary mechanism.
Adami goes on to talk about Thomas Ray’s program, Tierra. Tierra was directly inspired by VENUS. However, he made some crucial changes. Firstly, he instituted protections that prevented one program from overwriting another. Secondly, he made changes to the design of the computer code to make programs less likely to break due to mutations. Thirdly, he seeded the environment with a carefully crafted self-replicating program.
What happened? Tierra sees the evolution of some interesting dynamics like parasites, hyperparasites, and social reproduction. But as I show in my own work, “Tierra: The Character of Adaptation,” those changes involve either removing parts of the original program, making changes to an analogy of regulatory regions, duplicating code, or very small tweaks to the original program. They actually demonstrate a digital version of the sort of changes detailed in Michael Behe’s book Darwin Devolves.
Adami’s assessment is:
However, besides this coevolution, no novelty would emerge within the Tierra world because the world they inhabited was devoid of anything they could take advantage of: it was too simple.
Insufficiency of Darwinian Processes
Adami agrees that evolution in Tierra stalled, and he blames this on the simplicity of the environment. But the evolution we observed in Tierra showed rich dynamics in how different programs could interact. The system is flexible enough that a large variety of other strategies could have been implemented. Furthermore, Ray went on to build a networked version of Tierra that allowed programs to transmit themselves between different computers around the world. But the end result was the same. The issue was not that the environment was too simple; it was that Darwinian processes were insufficient to find useful ways to exploit their environments.
Adami then proceeds to present his own work on Avida as succeeding where VENUS and Tierra failed. He indicates this is because his environment is more complex, providing rewards for performing calculations. But, as he himself indicates:
I quickly learned that just rewarding such calculations with extra energy was not sufficient for these functions to evolve, because the code to trigger the rewards was too rare to simply occur via random mutations.
How Did He Resolve This?
He started rewarding programs for simply reading and writing numbers. The program quickly evolved to read and write numbers, but not to perform the addition he was trying to evolve. He then rewarded programs that read a number and then wrote it out. Then, and only then, was he able to evolve programs that added numbers. He was able to do so only by guiding the evolutionary process step by step to the answer.
Avida does not succeed because it is more complex. Avida succeeds because it abandoned the non-intervention of VENUS and Tierra. Both of those models set up a world and watched it run. They sought to allow the system to develop on its own. They expected evolution to produce the same sort of complexity that is attributed to it in the real world. That did not happen. Avida was not content to let evolution take its course, but carefully guided it so that it succeeded.
All of the simulations thus point to design. Without careful, step-by-step guidance, evolutionary processes are very limited. Once you introduce step-by-step guidance, you are no longer simulating an evolutionary process.