“Generating an incalculable paradox” might sound like a title from some cyberpunk novel, but for University of Louisville computer scientist Roman Yampolskiy, it’s one of several intricately detailed proposals he has listed for cracking into a simulated world. The proposal belongs to a lineage of speculation that stretches from René Descartes’ 17th-century musings about deceptive realities through Nick Bostrom’s 2003 articulation of the simulation argument, which generously approximated the probability of our lives being simulated within an extremely advanced extraterrestrial computer at around 20 percent.
Yampolskiy’s inquiry does not rest on philosophical speculation. Drawing inspiration from video game exploits, he envisions how to put stress on the computational limits of a hypothetical simulation like scheduling coordinated mass meditation followed by intense bursts of activity to trigger collapse or elicit response from the “simulator overlords.” Other philosophers have speculated paradox generation as a way to overwhelm the system, in the form of posing an insoluble problem to a finite computational framework.
But engineering reality of doing so is formidable. The size of computation required to do a high-fidelity simulation of our universe, with each atom of a human body on the order of 10^28 atoms, is gigantic. As critics argued, this would require essentially unlimited computing power, much more than even theoretical concepts like planet-sized processors powered by starlight. In such a view, to highlight the capacity of the simulation can be unrealistic if the system resources are orders of magnitude beyond our achievements.
Bostrom’s first argument relies on a probabilistic approach: technologically advanced civilizations possessing the capability and desire to run “ancestor simulations” in astronomical numbers would mean that simulated beings would vastly dominate true ones in terms of ratio. But his mathematics, later criticized on grounds of derivation errors, assumes as an axiom that if one has the facility, then motivation to perform such simulations follows. Additional analysis has illustrated instances where practically all civilizations do a great deal of simulating but the likelihood of any individual being simulated is small because actual populations are dispersed between non-simulating civilizations.
Physical experimentation in an attempt to detect the simulation has thus far revealed nothing out of the ordinary. Even the Large Hadron Collider, capable of simulating conditions seconds after the Big Bang and probing phenomena like dark matter and quantum entanglement, has yet to show “breaking” a simulated substrate. Alan Barr’s ongoing research at CERN, top quark entangling, probes the foundations of quantum theory itself, with questions of whether space-time is emergent or not. These experiments might, in principle, demonstrate paradoxes indicative of a simulation, but the outcomes of these experiments have consistently accorded with known physics.
From the computational scientist’s point of view, the bounds of validating complicated systems reflect the difficulty of verifying or breaking out of a simulation. The intensive AI safety research has shown that for systems more expressively capable than some threshold, establishing perfect safety is coNP-complete i.e., it might take longer than the universe has existed. Safe configurations in such systems form a measure-zero subset of all possibilities, and therefore are extremely unlikely. If our universe is a simulation of similar complexity properties, “escaping” might entail discovering and capitalizing upon an open or unstable condition that is statistically insignificant.
The philosophical consequences are no simpler. If the simulation hypothesis is untestable, as so many assert, then coming up with an escape plan might be indistinguishable from metaphysical fantasizing. Religious traditions, which generally presume an external world independent of the physical, have given no measurable impact on the supposed simulation. Even awareness of the hypothesis itself appears inert; to be aware one might be simulated does not, in any empirical way, alter the behavior of the simulation.
Yampolskiy acknowledges the risks of existence that accompany attempts to flee. A “baseline reality” may in theory offer vastly more computational capability and direct acquaintance with true physical laws. But the consequences are unclear possibly catastrophic should simulators respond defensively, perhaps by reboots with greater security and erasure of collective memory. This concept aligns with the “Alignment Trap” of AI safety research: the greater the system ability, the greater the requirement for almost perfect safety, but the ability to guarantee it collapses. In each case, the moment of greatest need is at the moment of lowest control.
Whether or not our universe is simulated can never be known. Until it is, the technical hurdles computational, physical, and epistemological are as overwhelming as the philosophical, with humanity, as Yampolskiy concedes, having the choice between seeking the red pill or taking the blue.
