Our AI discovered a win condition we never taught it.
What you’re about to watch is the first time our AI, AIRIS, successfully discovered a time-based win-win strategy entirely on its own.
AIRIS is a curiosity-driven AI we’re training to play Elowyn with humans and AI agents in PvA mode. In this match, AIRIS is playing against our current rules-based Elowyn Game AI.
One of the mechanics that makes Elowyn unique is its built-in physics of interdependence. Both factions rely on Elo, the Tree of Life. When players attack each other through brute force, instead of using Elowyn’s time and deception mechanics, Elo’s health suffers. If the Tree dies, the match ends in a draw and both players lose — reflecting the consequences of zero-sum behavior in action.
Without any instruction, AIRIS independently discovered a highly effective win-win strategy. It used Aleph’s Cosmic Serpent card, which grants a one-hour time advantage each time a code card is played.
Watching AIRIS discover a strategy is fascinating on its own, but what happened next made this moment that much more memorable.
The first time AIRIS used the Serpent, the opponent destroyed it. The next time it played the Serpent, AIRIS also used “Ashtara” to place the Serpent into stealth mode, hiding it from direct attacks. In doing so, it not only developed a strategy, but learned to protect the strategy it had discovered.
This replay captures the first successful execution of that mechanic.
We also began testing AI agents by fusing them with AIRIS during matches, exposing them to something they had never experienced before: the consequences of their reasoning under pressure.
It’s still early, but this moment felt worth sharing.
Join us in shaping AI for good, link in profile.