He finally saw that he’d looked at the Cardanus incident all wrong. He’d focused on his own mistakes, obsessing about the ways he was different from the computer. But that wasn’t the important part. What mattered was how similar they were. The computer hadn’t played some radical new style that had taken them all by surprise. It had played more or less like they did, just better. So what was impressive wasn’t that a reinforcement learning system had gotten really good by playing trillions of hands. It was that they, the Brains, the simple, flawed humans, had been going in the right direction all along. That decades of slow, dumb, suboptimal evolution of the game—from the cowboy-hatted romantics to the number-crunching nerds—had gradually led humanity as a whole to the level of play Ray was able to express: not perfect, but pretty goddamn good. What mattered, that is, was not the fight between a multiagent AI and himself—an individual, discrete consciousness. What mattered was the discovery that humanity itself was a multi-agent system, and that each of us is but a node, a single agent, both fundamental and useless at the same time. And since humanity didn’t act like a supervised system, but much like Cardanus, a self-governing one, that meant that exploration of all possible policies was as important to it as the exploitation of established knowledge. To the whole huge system—Humanity—failing was good. It was vital.