Artificial Life: The Field Life Helped Found
In September 1987, about 160 scientists gathered at Los Alamos National Laboratory for a workshop that had no proper name yet. There were computer scientists and biologists and physicists and anthropologists, and they shared a frustration that was hard to articulate: each of their fields contained pieces of something larger, some theory of how life-like complexity arises from non-living substrates, but no single field contained the whole thing. Christopher Langton, who had organized the meeting, proposed a name for what they were all circling around: artificial life.
The workshop — formally titled the Interdisciplinary Workshop on the Synthesis and Simulation of Living Systems — was co-sponsored by the Santa Fe Institute and Apple Computer. The proceedings Langton edited from it became the founding text of a new discipline. And running through almost every paper, as precedent, as inspiration, as canonical example, was Conway’s Game of Life.
Why Life Was the Right Starting Point
The connection between Conway’s Life and artificial life is not superficial. Life did not merely inspire ALife researchers the way a pretty painting might inspire a sculptor. It demonstrated something specific and important: that the behavioral signatures of biology — self-organization, persistence, pattern formation, apparent purposeful motion — could emerge from a system with no biology in it whatsoever. No chemistry, no carbon, no metabolism, no evolution. Just a grid, two states, and four rules.
This was a proof of concept for the central ALife thesis. Before Life, the argument that “life-like behavior could arise from non-biological substrates” was philosophical speculation. After Life, it was an empirical fact. You could run it on a PDP-10 and watch it happen.
Langton made this connection explicit. His definition of artificial life centered on the idea of studying “life as it could be” rather than “life as it is” — exploring the full space of possible life-like processes, not just the ones that happened to arise on Earth. Life was the simplest known member of that space. ALife was the systematic exploration of the rest of it.
Langton’s Self-Reproducing Loops (1984)
Before organizing the 1987 conference, Langton had already contributed the first major ALife result that built directly on Life’s precedent.
Von Neumann’s self-reproducing automaton, designed in the 1940s, required 29 states per cell and a configuration containing millions of cells — a theoretical proof of possibility, but barely a working model. Conway’s Life had reduced the complexity of interesting CA behavior from 29 states to 2. Langton applied the same reductionist impulse to self-replication specifically.
In 1984, published in Physica D, Langton designed an 8-state cellular automaton containing a self-reproducing structure of just 86 cells: Langton’s loops. The loops store their construction instructions in a dynamic circular tape — information coded into the pattern of states flowing around the loop — and use those instructions to build copies of themselves in neighboring space. They cannot compute universally, as von Neumann’s constructor could. But that was the point. Langton showed that universal construction was sufficient for self-replication but not necessary. You could replicate without being able to build anything.
Langton’s loops were the first demonstration that von Neumann’s complexity was accidental, not essential. Self-replication, like the complex behavior of Life, was achievable with much simpler rules than anyone had thought.
Thomas Ray and Tierra (1991)
Thomas Ray came to artificial life from an unexpected direction. From 1974 to 1989 he was a tropical biologist, studying the foraging behavior of vines in the rainforests of Costa Rica. What frustrated him, he later explained, was that as an ecologist he could only observe the products of evolution — he could not watch evolution happen. The evolutionary processes that had produced the biodiversity he was studying had taken millions of years and left no record of their intermediate steps.
Tierra, which Ray unveiled in 1991, was his solution. It was not a cellular automaton — Tierra operated at the level of machine code rather than grid cells — but it was deeply connected to the ALife tradition that Life had established. Ray created a digital “soup” of self-replicating assembly-language programs competing for memory and CPU time in a virtual computer. The “ancestor” program, 80 bytes long, could copy itself into a fresh block of memory by executing its own code. When Ray added random bit-flips as mutations and allowed programs to compete for the limited memory resource, evolution began.
What happened next surprised even Ray. Parasites appeared: shorter programs, as small as 45 bytes, that could not replicate themselves but had evolved to hijack the replication machinery of the ancestor. Then hyper-parasites evolved that could resist the parasites. Then cheats. The system, which had begun with a single hand-coded organism, bootstrapped an entire ecology in hours of CPU time. Punctuated equilibrium — long periods of stasis punctuated by rapid change — emerged naturally, without being programmed.
The connection to Life was philosophical but important. Both systems showed that ecological complexity — competition, parasitism, co-evolution — could emerge from simple substrate rules without anyone designing it in. Life showed it with two-state cells; Tierra showed it with machine code.
Karl Sims and Evolved Virtual Creatures (1994)
If Tierra demonstrated that ecology could evolve in a digital medium, Karl Sims demonstrated that morphology could. His paper “Evolving Virtual Creatures,” presented at SIGGRAPH 1994, is one of the most visually dramatic results in ALife history.
Sims created a virtual 3D physics environment and populated it with evolving creatures represented as directed graphs: nodes described body segments, connections described joints and actuators, and a separate graph described the neural control circuitry. A genetic algorithm varied these structures across populations of several hundred creatures, selecting for specific behaviors: swimming, walking, jumping, competing for a green cube.
The results were astonishing. Creatures evolved that swam with sinusoidal undulations, walked with asymmetric gaits, and competed for the cube using strategies that included what looked uncomfortably like wrestling. Some locomotion strategies that evolved would, Sims noted, “be difficult to invent or build by design.” They were not designed — they were discovered by the evolutionary process itself.
Sims’ work sat at the intersection of ALife and robotics, and it extended the Life paradigm in a specific way. Life showed emergence in a flat 2D grid; Sims showed it in 3D physical space, with continuous-valued physics rather than discrete cell states. The principle was the same: give evolution simple materials and rules, and it finds solutions you didn’t anticipate.
Christoph Adami and Avida (1993)
Christoph Adami, working at Caltech in 1993 with Charles Ofria and C. Titus Brown, created Avida — a digital evolution platform that has since become the primary research tool for studying evolutionary biology in silicon.
Avida differs from Tierra in a crucial respect: it was designed not just to observe evolution but to use digital organisms as instruments for testing specific hypotheses about evolutionary biology. Avida organisms are self-replicating programs that evolve in a population with mutations, and they can be given “logic tasks” — the ability to perform Boolean operations — that are rewarded with extra CPU time (the system’s proxy for fitness). By tracking which tasks evolve and in what order, researchers can study questions that are impossible to address with biological organisms: How does irreducible complexity arise? What is the evolutionary path from simple to complex features?
The landmark result came in 2003, when Adami and Ofria, in a paper published in Nature, demonstrated the evolution of a mathematical equals operation from simpler bitwise components. Critics of evolution had argued that complex features requiring multiple simultaneous mutations could not evolve gradually. Avida provided direct experimental evidence that they could.
Adami has described his approach as treating evolution as a form of information processing — a perspective that connects directly to the information-theoretic aspects of Life and to Claude Shannon’s framework. Life generates structured information from unstructured initial conditions; Avida generates biological information (adaptation) from random mutation. The substrate differs; the principle is the same.
The Philosophical Question: Is Any of This Alive?
The question has followed ALife from the beginning, and it has not been resolved.
The weak ALife position holds that simulations of life are useful models — they help us understand biological processes by analogy — but are not themselves alive in any meaningful sense. A simulation of a hurricane is not a hurricane. A simulation of evolution is not evolution.
The strong ALife position, associated with Langton and with philosophers like Mark Bedau, holds that the behavioral properties of life are not substrate-dependent. If a system exhibits self-organization, reproduction, adaptation, and metabolism, it is alive, in the same sense that a computer program running the right algorithm is computing — regardless of whether it runs on silicon or carbon. On this view, Tierra’s parasites are genuinely parasites; Avida’s organisms are genuinely evolving; Life’s gliders are genuinely moving.
The debate is not merely semantic. It has consequences for how we think about the hard problem of consciousness, about the definition of death, about the moral status of digital entities. If Life’s self-replicating patterns (first demonstrated by Dave Greene and Paul Chapman in 2010) are alive in any sense, then Life is not just a mathematical curiosity — it is a habitat.
Conway himself was amused by the question. He pointed out that mathematicians had been debating the ontological status of mathematical objects for centuries without resolution, and that ALife was just a special case of that older argument. Whether or not the glider is “alive,” he observed, it certainly does something interesting.
The Field Today
ALife has matured into a diverse research community with its own journal (Artificial Life, published by MIT Press since 1993), its own conference series (ALife I through the present), and productive connections to evolutionary biology, robotics, machine learning, and philosophy of mind.
The canonical text of the field’s founding generation is still the proceedings of Langton’s 1987 Los Alamos workshop. But the most active current research is happening at the intersection of ALife and deep learning: neural cellular automata, differentiable physics, and transformer-based generative models that exhibit emergent behavior are all extensions of the project that Langton named and Life exemplified.
The question Langton asked in 1987 — what is essential to life, stripped of its earthly accidents? — remains open. But the tools for exploring it are vastly more powerful than anything available at Los Alamos. And the example that motivated the question, Conway’s grid of cells and rules, still runs on every machine.