Sim-to-Real Gap

A machine-learning term Bratton borrows and declines to read as a gap. Embodied AIs like driverless cars learn in toy worlds, reductive simulations where they can bump into walls and run many generations in an instant before entering physical reality. For a General Theory of Simulations holds that for the AI the boundary between a world built of data and a world perceived as data is already blurred, so the crossing is a hybridization of virtual and physical inputs, not a border; Speculative Philosophy of Planetary Computation names the program's interest as the mixing of thing and shadow; the Agentworld brief extends this into real-to-sim-to-real relays many orders deep. Vivarium, by contrast, treats Sim2Real as an engineering problem, simulations mismodelling physics or missing the real's indeterminacies, and Maughan's Infinity Mirror supplies the failure case: a dog missing from Walmart's drone simulation. Relates to Digital Twin and Reflexive vs. Recursive Simulation.