Two Apple computers connected over an Ethernet cable and played Rock Paper Scissors against each other. The machines exchanged messages over the network, sent their throws, resolved the outcome, and kept a running score. It's a demonstration video, not a research paper, but it raises a genuine question: what would those throws look like over time?
A computer playing against a human player has some well-documented advantages. It doesn't tire. It doesn't tilt after losing a streak. It doesn't start throwing Rock three times in a row because it's frustrated. It doesn't give any of the physical tells that human players display before a throw — the slight muscle tension, the hand position, the shifts in breathing cadence that experienced players read. It simply waits for input, processes it, and responds.
The interesting question is what strategy a machine would run. The naive approach — pure random selection, equal probability across Rock, Paper, and Scissors — is theoretically unexploitable. In a truly random sequence, there is no pattern to read and no bias to target. Humans can't actually produce randomness on demand; we have behavioral defaults, and those defaults become predictable under enough observations. A computer running genuine randomness beats human "randomness" on that axis.
But the research shows that even human players don't play pure random. The conditional response bias documented in the 2014 Zhejiang University study — winners tend to stay with their throw, losers tend to shift — creates exploitable patterns. Against a human opponent, a machine that tracked those conditional probabilities and updated its throw selection accordingly would have a systematic edge.
The demo doesn't do any of this. Two computers talking over Ethernet and playing random throws is a protocol demonstration, not a strategy test. But it's a reasonable seed for thinking about what machine opponents might actually do if you gave them enough compute and enough data. The competitive RPS community has been interested in this question since the first RPS bots started appearing in online tournaments. A machine that genuinely modeled opponent behavior would be something different from a robot that throws randomly.