The Long Hunt for a Competitive Game of Life

Conway's Game of Life is famously a zero-player game: you set the board, hit go, and watch. That's the whole appeal — no moves, no opponent, just consequences unfolding from an initial state. Which makes a running side-project of Life's fan community a little perverse: for over fifty years, people have tried to turn it into something you can actually play against another person.

They haven't had much luck. Here's the trail — Martin Gardner's original verdict, a Birkbeck research paper that took the challenge seriously, and the handful of implementations that eventually shipped, none of which used the harder, more interesting fix that paper proposed.

Gardner's verdict, 1983

Martin Gardner introduced Life to the world in his October 1970 Scientific American column. Thirteen years later, in his 1983 collection Wheels, Life, and Other Mathematical Amusements (W.H. Freeman), he devoted three chapters to what people had done with Life since — including attempts to make it competitive. His verdict was one blunt sentence:

"Attempts have also been made to invent competitive games based on 'Life', for two or more players, but so far without memorable results."

As far as the public record shows, that stood as the accurate assessment for the next two decades.

p2life: the paper that took the challenge seriously, 2002

In 2002, Mark Levene and George Roussos, at Birkbeck College, University of London, published "A Two-Player Game of Life", introducing a variant they called p2life. They open by quoting Gardner's put-down directly and treat it as an open problem worth solving properly.

Their first move is explaining why the obvious fix falls short. A two-player variant already existed — usually called Immigration or "black and white" Life — where cells are white or black, and a birth assigns the new cell to whichever color holds the majority of its three parent cells. It's an elegant one-line rule change. But it only touches birth. Survival still plays out exactly as in the original single-player game, blind to who owns what. Two colors can be on the board, but only one game is actually being played.

p2life goes further: both birth and survival become contested. A cell survives based on the balance of same-color versus opposite-color neighbors; exact three-three ties at birth get resolved by a coin flip. It's a small rule change with a real consequence. The paper shows a symmetric seven-cell configuration that annihilates into empty space in standard Life — but under p2life's competitive survival rule, resolves into a surviving black block. Another configuration that decays into six blinkers in ordinary Life instead resolves into a clean win for black. Adding competition to survival, not just birth, is what turns a spectator pattern into a contested one.

They also measured how close the games run. Across 400 simulated matches on a 100×100 grid, over 69% ended with the loser still holding more than a third of the final population — mostly close games, not blowouts, which is a good sign for something meant to be played rather than merely simulated.

Then, in their closing paragraph, they name the actual next step:

"A problem that we are now investigating is how to convert p2life into a 'real' game, i.e. where players are allowed to make moves between generations, which change the configuration of the tokens, and devise strategies to overpower or live side-by-side with their opponent."

As far as we've been able to find, that follow-up was never published. The interactive version of p2life — the one where you actually play, instead of just watching two colors fight it out from a random starting grid — stayed an open problem in the academic literature.

What actually got built

Nobody picked up p2life's harder rules. But over the following decade, a handful of independent projects built multiplayer Life anyway — using the simpler Immigration ownership rule Levene and Roussos had already identified as insufficient on its own.

lifecompetes.com, live since 2015 and still running today (verified live while writing this), is the clearest example. It's a persistent, real-time free-for-all: everyone shares one board, a birth is assigned to whoever holds two of its three parent cells, and a leaderboard ranks players by peak cells controlled. What turns it into a game rather than a spectacle is a simple drip economy: you place cells from a limited pool, and anyone holding fewer than twelve banked cells gets a new one every six generations. No survival contest, no p2life sophistication — just majority-birth ownership plus scarcity, and it's been enough to keep a hobbyist project alive and populated for over a decade.

Life Arena tried a different shape: a turn-based duel with a cell budget per side, a pattern library to place from (Glider, Pulsar, Gosper Glider Gun among them), and both flat and spherical arenas. The single-player, same-device hot-seat version worked. But "Online PVP Coming Soon" — text we found sitting directly in the site's own bundled code — never became real; no functioning online multiplayer ever shipped, and as of this writing the site itself is offline entirely.

Multi-Life takes concurrency furthest: an ever-evolving shared grid that anyone connected can edit live, no turns at all — the project describes itself as "chaotic multiplayer," which is fair. It's still running, but it has no scoring or win condition, which makes it a different genre wearing the same rules rather than a real answer to Gardner's challenge.

The pattern

Line these up and the shape is obvious. Every project that actually shipped converged on the same two ingredients: Immigration's majority-color ownership at birth, plus a limited, regenerating resource for placing cells. Nobody, in the twenty-plus years since p2life proposed it, has shipped competitive survival — the harder half of Levene and Roussos's design, and the specific mechanic that turned their annihilating configurations into real wins instead of empty boards.

That's either a sign the harder version isn't worth building, or a sign that the easy rule set is good enough to ship while realtime netcode — where Life Arena's ambitions quietly died — is the actual bottleneck. Nobody has published the answer.

There's a plausible design reason interactive p2life is harder than it looks, beyond netcode: standard Life already runs the whole board forward one generation at a time, everywhere at once. The moment you let a player make a move between generations, you have to decide how that move interacts with a board that's simultaneously evolving on its own — whether moves happen before the tick, after it, or get folded into the birth/survival rule itself. Levene and Roussos flagged that exact question as future work and, as far as the record shows, never answered it in print. Every shipped game sidestepped the problem by keeping the contested part (ownership) and dropping the autonomous part (competitive survival) — which is simpler to reason about, but also why none of them feel like they're using the full Life rule set, just borrowing its aesthetic.

It's the kind of question you only resolve by building the thing, which is exactly the design space we've been digging into while thinking about what to build next on top of our own pattern library and simulator.

If you want to poke at the raw material yourself in the meantime, our MCP server can evolve any pattern forward exact generations and check a predicted outcome against the real simulation — including the kind of two-color collision patterns p2life's paper describes by hand. Fifty-six years after Gardner's original column, and forty-plus since he wrote off competitive Life as a dead end, the "real" version he doubted still hasn't been built by anyone. We're curious enough about that gap to keep pulling on it.