Soup Opera

Toby Howard

This article first appeared in Personal Computer World magazine, August 1996.

OVER THREE THOUSAND MILLION years ago, the first life on Earth emerged from a primeval soup of non-living molecules energised by light and lightning. How these molecules evolved the ability to make copies of themselves, the key to life, remains unknown. Now, a new PC-based simulation system called Amoeba offers scientists the chance for hands-on research. For the first time, self-replicating digital life-forms 'living' inside the computer have spontaneously emerged from binary chaos.

The brainchild of Dr Andrew Pargellis at AT&T's Bell Laboratories in New Jersey, Amoeba models the primeval soup as a chunk of computer memory. Each memory address may be occupied by a single digital organism, or 'cell'. A cell is itself a computer, albeit an extremely simple one: it comprises four storage registers, a program counter, and a sequence of instructions which forms its genetic code. A cell 'lives' when it executes its instructions. Philosophically-inclined readers might object to such a naive interpretation of 'living', but for the simplest organism it seems a plausible view.

Amoeba draws its inspiration from the 'Tierra' simulator developed by artificial life researcher Thomas Ray, but with one crucial difference: in Tierra, life-forms must be manually introduced into the soup; in Amoeba, the soup is seeded with random instructions, and life-forms spontaneously appear, as if by magic.

One reason for Amoeba's fecundity is that its life-forms have a simple structure. The genetic code of each cell is a sequence of up to 30 computer instructions, chosen from a repertoire of 16 possible instructions. These are based on standard operations such as 'jump to address', 'copy contents of memory location A into memory location B', and 'compare registers and skip next instruction'. Pergellis has carefully designed his instruction set such that it is possible to create a self-replicating cell with only 5 instructions (see illustration).

At the start of a simulation run, 70% of the soup is seeded with cells containing between 1 and 25 randomly chosen instructions. The remainder of the soup is empty. Amoeba then visits each cell and executes its instructions. At this stage, most cells will contain meaningless sequences of instructions, and their execution will have no effect on the soup. A few cells, however, will by chance contain instructions that access other memory locations, copying instructions there or exhibiting viral behaviour by invading another cell and executing its code. Whenever all 1000 locations in the soup contain code, Amoeba invokes a 'reaper' routine, which kills off 30% of the cells, and introduces a few new random cells into the soup. A new generation then begins.

Pargellis found that on average, a self-replicating sequence of instructions would spontaneously arise after about 400 generations. Initially, such 'self-reps' were often unnecessarily complex, and contained portions of code which performed no useful task, or were skipped over. This is directly analagous to the 'introns' and 'exons' found in real biological genes, where useful genetic information is interspersed with patches of nonsense.

Because cells compete for resources of processor time and memory space, Amoeba models evolution and natural selection. Whenever a cell reproduces, there is a 10% probability that one of its instructions will mutate into another instruction. Most mutations spoil the instruction sequence and inhibit further self-replication. In 3% of cases, however, Pargellis observed that mutation allowed self-replication to continue, and led to cells which shed redundant code, lived longer and reproduced more efficiently.

Compared to the richness of nature, Pargellis' system might appear highly unrealistic. But the design concepts are sound. As Pargellis says, "I chose a system such that I wouldn't need a substantial fraction of 4 billion years to see the results I did".

As well as providing insights into our biological prehistory, might Amoeba also offer a glimpse of the future? Cautiously optimistic, Pargellis envisages chips with simple instruction sets, whose programs dynamically mutate in response to environmental stimuli.

Although developing software by self-replication, evolution and natural selection sounds like science fiction, Amoeba has already demonstrated the remarkable fact that in the digital world, order can indeed arise out of chaos. Ironic, perhaps, when our experience with today's software is often quite the reverse.

Toby Howard teaches at the University of Manchester.