When AMC set out to build Guess It!, the pitch was simple: charades, but playable with anyone, anywhere, anytime. "Anytime" turned out to be the hard word — it meant building an opponent who never sleeps.
The referee problem
A guessing duel needs a referee. When a player asks "Is it big?" about a hedgehog, someone has to rule. Humans handle ambiguity with shrugs and laughter; software has to commit. Our AI referee answers yes/no questions about every card consistently and fairly — which meant teaching it not just facts about each card, but judgment: what a reasonable person would say. Players notice instantly when a referee feels unfair; consistency mattered more than cleverness.
Making the AI a worthy opponent
The same engine plays offense. The AI opponent builds question trees the way strong humans do — each question chosen to split the remaining cards — but we deliberately tuned it below perfect play. An optimal guesser is a miserable opponent; a disciplined but beatable one (players are finding the cracks) makes you feel sharp when you win.
1v1, the hard mode of multiplayer
Live 1v1 duels meant matchmaking, reconnection handling, and a shared round-state two phones must agree on across flaky networks. It's the feature most party games skip, and the one our players use most.
What's next
More decks, more languages, and smarter daily challenges. Try the result — free on Google Play — and tell us where the referee got it wrong. We read everything.