The idea that our universe might be a computer simulation has moved in and out of fashion for two decades. In 2025 it is back, and the reason is partly technological: AI systems can now generate convincing interactive worlds from a sentence of text, which makes the old thought experiment feel less abstract. At the same time, new work in physics argues the hypothesis fails on mathematical grounds. The result is a debate that pulls together philosophy, religion and machine learning.
An old idea with a new edition
Rizwan Virk, a computer scientist and founder of Play Labs at MIT who now teaches at Arizona State University, published a revised and expanded edition of his book The Simulation Hypothesis in 2025. Virk argues that the notion is not a product of science fiction alone. He traces parallel ideas through Hindu and Buddhist teachings about maya, the world as illusion, and through Sufi and other mystical traditions that describe the physical world as a veil over a deeper reality. In his reading, religions have long described the self as something that perceives only a partial version of what exists.
The modern, secular version of the argument comes from the philosopher Nick Bostrom, whose 2003 paper set out the case most people now cite. Bostrom reasoned that if any civilization reaches the point of running vast numbers of realistic simulations of its ancestors, then simulated minds would vastly outnumber original ones, and the odds would favor any given observer being inside a simulation. The argument does not claim we are simulated; it claims that at least one of three possibilities must hold, and that one of them is that we are.
Why AI changed the mood
What has shifted is the plausibility of the first step. In August 2025 Google DeepMind announced Genie 3, a world model that generates interactive, navigable 3D environments in real time from a text prompt, keeping them consistent for several minutes as a user moves through them. Landscapes, water, weather and lighting are produced on the fly rather than drawn by hand. DeepMind positions the system as a training ground for AI agents, but it also demonstrates that generating a persistent virtual world is becoming an engineering problem rather than a fantasy.
Proponents of the hypothesis draw the obvious inference. If a company can already produce explorable environments from a sentence, a civilization a few centuries ahead could in principle produce something far richer, and the question of whether our own world is such a product stops sounding absurd. Critics reply that a model generating a few minutes of plausible video is a long way from simulating conscious observers and the laws of physics, and that the gap between the two is the whole argument.
What scientists say
A few researchers have tried to put numbers on it. David Kipping, an astronomer at Columbia University, applied Bayesian reasoning to Bostrom’s argument in a 2020 paper and concluded that the odds are close to even, slightly favoring the view that we are not simulated. He also noted that if humans ever build such simulations, the balance would tip the other way.
Others reject the premise outright. In October 2025 a team led by Mir Faizal at the University of British Columbia Okanagan published work arguing that the universe cannot be fully simulated by any algorithmic process. Drawing on Gödel’s incompleteness theorems, the authors contend that a complete description of reality requires a form of non-algorithmic understanding that no computation can reproduce. The university described the conclusion as ruling out the simulation hypothesis; other physicists have been more cautious, pointing out that the argument depends on assumptions about the structure of a final theory of physics that are themselves unsettled.
The question that remains
Strip away the technology and the dispute returns to questions philosophers have asked for millennia: whether the world we perceive is the real one, and if not, what lies behind it. Religious traditions answered with notions of a veil, a test or a larger story. Bostrom answered with probability. The UBC team answers with mathematics. Generative AI has not settled any of it, but it has made the simulation hypothesis a live topic in research labs and technology companies rather than only in philosophy seminars, and that shift is itself a measure of how far the tools have come.
