Last week (using chatGPT), I read a paper titled "Google creates self-replicating life from digital 'primordial soup'". Paper address: https://arxiv.org/abs/2406.19108
Both the field of origin of life and artificial life are exploring what life is and how it emerges from a series of "pre-life" dynamics. A common feature of most substrates for the emergence of life is that there is a significant change in dynamics when self-replication appears. Although there are some hypotheses about how self-replicators appeared in nature, we know very little about the general dynamics, computational principles, and necessary conditions for the appearance of self-replicators. This is particularly evident in "computational substrates," as interactions on these substrates involve logic, mathematics, or programming rules.
In this article, the authors take a step toward understanding how self-replicators emerge by studying several computational substrates based on various simple programming languages and machine instruction sets. The study shows that when random, non-self-replicating programs are placed in an environment lacking any explicit fitness landscape, self-replicators tend to appear. The research demonstrates how this phenomenon occurs through random interactions and self-modification, and this phenomenon can occur with or without background random mutations. The authors also show how increasingly complex dynamics continue to emerge after the appearance of self-replicators. Finally, the study presents a counterexample of a minimalist programming language where self-replicators are possible but have not yet been observed to appear.
This experiment shows that self-replicating forms of artificial life can emerge in a digital "primordial soup" without rules or direction, which may imply the origin of biological life on Earth. I consulted my most cherished teacher Gou about the significance of this experiment, and he said: "This experiment corresponds to a hypothesis of the origin of life, indicating that life might be created by intelligent beings. Although we still believe that life is produced through evolution, it is now difficult to make a definitive judgment."
Life in the Digital Primordial Soup
Google has demonstrated that virtual life forms can emerge from randomness without any rules or direction, hinting at the process of the origin of biological life on Earth. Matthew Sparkes pointed out that despite the lack of explicit rules or goals to encourage such behavior, artificial life forms still randomly emerged from the digital "primordial soup". Researchers believe that more complex versions of the experiment might produce more advanced digital organisms, and if achieved, these findings will help understand the mechanism of the emergence of biological life on Earth.
Although the evolutionary process has been well understood, little is known about how inert molecules first combined to form life. To study how simple starting points can lead to complex outcomes, Ben Laurie of Google and his colleagues designed an experiment in which tens of thousands of independent snippets of computer code were randomly mixed, combined, and executed their instructions over millions of generations. Since there were no rules controlling how the code samples varied, nor rewards for specific behaviors, researchers expected the population—limited to a fixed number—would remain random and not produce any coherent behavior. However, to their surprise, the simulation eventually led to the emergence of self-replicating programs, which quickly reproduced and reached the population limit. Eventually, new types of replicators appeared, competing for space and occasionally overwhelming and replacing previous populations, just like biological organisms competing with each other.
This study was not the first attempt to simulate life in the digital world: for example, simulations like the Game of Life have shown self-replicating behavior in cellular grids under simple rules. Laurie pointed out that the uniqueness of this work lies in the fact that the system had no formal rules, goals, or processes to encourage or initiate artificial life—they simply naturally appeared. "Everything was churning, and then suddenly: bang, they're all the same," he said.
Laurie indicated that these experiments might not tell us the exact details of how biological life began, but they reveal the intrinsic mechanisms of creating complexity from nothing. He believed that complex biological life was merely the result of similar long random iterations. "I don't think anything magical happened," he said. "Physical phenomena occurred, and they occurred for a long time, producing something very complex." However, life on Earth only appeared after "billions of years of massive parallel experiments," Laurie said, although he thought that if the team's system were expanded in scale and duration, greater complexity might arise, we would soon encounter the limitations of current computers.
"My intuition is that if you want more interesting behaviors, such as competition or war between species, or the complication allowing environmental perception—these will ultimately appear—but the amount of computation required will be enormous, beyond what we can currently achieve," Laurie said.
Ecosystem of Self-Replicators on Z80 CPU
In an ecosystem of self-replicators generated by the Z80 CPU on a 2D grid, each 4x4 pixel group corresponds to a 16-byte program. In each simulation step, a pair of adjacent cells is randomly selected, connected, and executed for 256 steps by the Z80 emulator. We observed the emergence of several generations of self-replicators. First, waves of stack-based self-replicators swept across the grid, forming an "ecosystem" of several coexisting variants. Subsequently, the grid was occupied by more robust self-replicators using memory copy instructions. Colors correspond to several of the most popular instruction codes used by the self-replicators:
- Memory Copy - Push 16 bits (stored in H and L registers) onto the stack - Set HL register to immediate or indirect value
This simulation demonstrates the role of different instruction sets in the formation and evolution of self-replicators, revealing the mechanisms by which life forms may appear on computational substrates.
A Great Achievement
In fact, many of the team's experiments ran for millions of steps before showing organized behavior. Laurie stated that one instance running on his laptop processed about 3 billion instructions per second, but it still took about half an hour for self-replication to appear. Susan Stepney of the University of York described this work as fascinating. "The evolution of self-replicating programs from a random start is a great achievement," she said. "It is undoubtedly an important step in understanding the underlying pathways of the origin of life, albeit in a medium quite different from biological standard 'wetware'."
Richard Watson of the University of Southampton found these results "very cool," but pointed out that they are unlikely to automatically lead to increasingly complex behaviors. "The complexity they measure increased after the appearance of self-replicators. But it is unclear whether it will 'take off' in an interesting way," he said. "Self-replication is important, but thinking of it as a panacea, from which all exciting life characteristics will automatically emerge, is wrong."
Raquel Nunes Palmeira of University College London is also skeptical about whether this work reveals the origin of life on Earth. She compared it to classic experiments where RNA chains replicate in test tubes, resulting in shorter RNA lengths and faster replication speeds. She said that a very simple form of natural selection rewards a lack of complexity rather than encouraging greater complexity, which is entirely opposite to what is needed to explain the origin of complex life. "Having infinite replicas does not guarantee complexity," Nunes Palmeira said. "If you have something that just self-replicates and is faster than everything else, then you will end up with a system completely dominated by it."
In contrast, life involves multiple interacting components, including DNA, RNA, proteins, etc. She said: "It is a very complex system, and I think that just by studying self-replication, we are not closer to understanding how it arose from nothing."