They trained a tiny patch of neurons to respond to low-voltage electric impulses. The cells don’t know they’re playing Doom. They don’t have any kind of social context or even video feedback.

Imagine if I stuck you in a sensory deprivation chamber, handed you an NES controller, and asked you to hit the buttons. Then, periodically, I said “Yes” or “No” based on the buttons you pressed. And when I pulled you out of the tube at the end of an hour, I told you “the yes and no messages were intended to encourage you to correctly navigate Mario through the first level of the original game.” What if, instead of Mario, I’d been telling you how to play Street Fighter?

It doesn’t matter if its Doom. They likely picked Doom because the I/O is so rudimentary that you can install the game on practically anything. The cellular matter has no idea what it’s doing beyond the “Yes/No” signaling.