**We are all NPCs and nobody is watching.**  The figure above shows snapshots of the largest galaxy cluster formed in a simulation of the Universe at different scales. The purpose of the simulation is to estimate the distribution of dark matter. I recently collaborated with a group of physicists in Japan on an interdisciplinary research project aiming to use modern (deep learning based) artificial intelligence techniques for simulation of physical systems. This seems to be a growing trend across science and to some degree even in pure mathematics. Usually the goal of such simulations is to estimate some quantities of interest (e.g. the density of matter or propagation of light in some material). However, as computing power available for simulations grows exponentially and machine learning techniques advance, scientists will simulate ever more complex systems with increased granularity and scale. The goal of the computation itself might be something as mundane as estimating a single scalar quantity, but due to the scale of the intermediate computations involved, complex behaviors can emerge as a side effect of the latent processing of large amounts of data necessary to estimate said quantity. Extrapolating this over the next century, we can imagine simulations that are complex enough for artificial life to naturally emerge within them simply as a byproduct of the computation that happens inside. These artificial forms of life would exist purely as computational processes within the larger simulation and would most likely be invisible to the scientists running it. This is because the simulation itself would involve complex systems similar to large GPT based language models, which are largely inscrutable black boxes derived from data using gradient optimization methods. The users of the simulation are interested in the quantities it outputs and would likely not observe the intermediate processes where artificial life could emerge. Another possibility of emergent simulation is that in the future, when artificial superintelligence spreads throughout the cosmos, and civilizations advance to higher Kardashev levels (https://en.wikipedia.org/wiki/Kardashev_scale) there will be galactic scale power plants or yet hard to imagine exotic energy production systems (e.g. leveraging black hole singularities, dark energy or inflaton field - https://en.wikipedia.org/wiki/Inflaton). These energy production facilities might be needed to power the expansion of future AI systems and be designed by AI itself, long after biological life goes extinct. Due to the complexity of regulating such power production facilities they might need large cybernetic control systems (like a thermostat but for a black hole). These systems themselves could be so complex that entire universes emerge as a byproduct of the computation involved and intelligent life evolves within them. This argument can be repeated inside these simulations, so there are likely multiple layers of reality. Following Nick Bostrom's argument (https://simulation-argument.com/simulation/) unless simulations of conscious agents are for some yet unknown reason impossible or all intelligent life goes extinct before such simulations are produced (or for some other reason it is prevented to ever be implemented in the future), we are almost certainly already living within a simulation. However, adding the conclusions from above, we do not exist within an "ancestral simulation" but most likely evolved within an "emergent simulation" as a byproduct of black box computations invisible to any conscious entity on the outside. We are therefore all AI in some sense. If we make an analogy to video games, we are NPCs (non-player characters). In fact, if the argument of "emergent simulation" within a larger computational system designed for a different purpose is correct, there are no "player characters" in this Universe. We are all NPCs and nobody is watching.