by Brian Greene
And that’s with today’s technology. Quantum computing—harnessing all the distinct possibilities represented in a quantum probability wave so as to do many different calculations simultaneously—has the capacity to increase processing speeds by spectacular factors. Although we are still very far from mastering this application of quantum mechanics, researchers have estimated that a quantum computer no bigger than a laptop has the potential to perform the equivalent of all human thought since the dawn of our species in a tiny fraction of a second.
To simulate not just individual minds but also their interactions among themselves and with an evolving environment, the computational load would grow orders of magnitude larger. But a sophisticated simulation could cut computational corners with minimal impact on quality. Simulated humans on a simulated earth won’t be bothered if the computer simulates only things lying within the cosmic horizon. We can’t see beyond that range, so the computer can safely ignore it. More boldly, the simulation might simulate stars beyond the sun only during simulated nights, and then only when the simulated local weather resulted in clear skies. When no one’s looking, the computer’s celestial simulator routines could take a break from working out the appropriate stimulus to provide each and every person who could look skyward. A sufficiently well-structured program would keep track of the mental states and intentions of its simulated inhabitants, and so would anticipate, and appropriately respond to, any impending stargazing. The same goes for simulating cells, molecules, and atoms. For the most part, they’d be necessary only for simulated specialists of one scientific persuasion or another, and then only when such specialists were in the act of studying these exotic realms. A computationally cheaper replica of familiar reality that adjusts the simulation’s degree of detail on an as-needed basis would be adequate.
Such simulated worlds would forcefully realize Wheeler’s vision of information’s primacy. Generate circuits that carry the right information and you’ve generated parallel realities that are as real to their inhabitants as this one is to us. These simulations constitute our eighth variety of multiverse, which I’ll call the Simulated Multiverse.
Are You Living in a Simulation?
The idea that universes might be simulated on computers has a long history, dating as far back as suggestions made in the 1960s by the computer pioneer Konrad Zuse and the digital guru Edward Fredkin. I worked at IBM during five summers spanning college and graduate school; my boss, the late John Cocke, himself a revered computer specialist, spoke frequently of Fredkin’s view that the universe was nothing but a giant computer chugging along, executing something akin to cosmic Fortran. The idea struck me as taking the digital paradigm to a ridiculous extreme. Through the years, I hardly gave it a thought—until I encountered, much more recently, a simple but curious conclusion by the Oxford philosopher Nick Bostrom.
To appreciate Bostrom’s point (one that Moravec had also hinted at), begin with a straightforward comparison: the difficulty of creating a real universe versus the difficulty of creating a simulated universe. To create a real one, as we’ve discussed, presents enormous obstacles. And if we succeeded, the resulting universe would be beyond our ability to see, which invites the question of what motivated us to create it in the first place.
The creation of a simulated universe is a wholly different enterprise. The march toward increasingly powerful computers, running ever more sophisticated programs, is inexorable. Even with today’s rudimentary technology, the fascination of creating simulated environments is strong; with more capability it’s hard to imagine anything but more intense interest. The question is not whether our descendants will create simulated computer worlds. We’re already doing it. The unknown is how realistic the worlds will become. Should there be an inherent obstacle to generating artificial sentience, all bets are off. But Bostrom, assuming that realistic simulations prove possible, makes a simple observation.
Our descendants are bound to create an immense number of simulated universes, filled with a great many self-aware, conscious inhabitants. If someone can come home at night, kick back, and fire up the create-a-universe software, it’s easy to envision that they’ll not only do so, but do so often. Think about what this scenario might entail. One future day, a cosmic census that takes account of all sentient beings might find that the number of flesh-and-blood humans pales in comparison with those made of chips and bytes, or their future equivalents. And, Bostrom reasons, if the ratio of simulated humans to real humans were colossal, then brute statistics suggests that we are not in a real universe. The odds would overwhelmingly favor the conclusion that you and I and everyone else are living within a simulation, perhaps one created by future historians with a fascination for what life was like back on twenty-first-century earth.
You may object that we have now run headlong into the skeptical quicksand we planned at the outset to avoid. Once we conclude that there’s a high likelihood that we’re living in a computer simulation, how do we trust anything, including the very reasoning that led to the conclusion? Well, our confidence in a great many things might diminish. Will the sun rise tomorrow? Maybe, as long as whoever is running the simulation doesn’t pull the plug. Are all our memories trustworthy? They seem so, but whoever is at the keyboard may have a penchant for adjusting them from time to time.
Nevertheless, Bostrom notes, the conclusion that we’re in a simulation does not fully sever our grasp on the true underlying reality. Even if we believe that we’re in a simulation, we can still identify one feature that the underlying reality definitely possesses: it allows for realistic computer simulations. After all, according to our belief, we’re in one. The unbridled skepticism generated by the suspicion that we’re simulated aligns with that very knowledge and so fails to undermine it. While it was useful when we began to weigh anchor and declare the reality of all that seems real, it wasn’t necessary. Logic alone can’t ensure that we’re not in a computer simulation.
The only way to dodge the conclusion that we’re likely living in a simulation is to leverage intrinsic weaknesses in the reasoning. Maybe sentience can’t be simulated, full stop. Or maybe, as Bostrom also suggests, civilizations en route to the technological mastery necessary to create sentient simulations will inevitably turn that technology inward and destroy themselves. Or maybe when our distant descendants gain the capacity to create simulated universes they choose not to do so, perhaps for moral reasons or simply because other currently inconceivable pursuits prove so much more interesting that, much as we noted with universe creation, universe simulation falls by the wayside.
These are among numerous loopholes, but whether they’re large enough for the proverbial truck to drive through, who knows?* If not, you might want to spice up your life a bit, make your mark. Whoever is running the simulation is bound to get tired of wallflowers. Being a cynosure would seem a likely path toward longevity.5
Seeing Beyond a Simulation
If you were living in a simulation, could you figure that out? The answer depends in no small part on who is running your simulation—call him or her the Simulator—and the manner in which your simulation was programmed. The Simulator, for instance, might choose to let you in on the secret. One day while taking a shower you might hear a gentle “dingding,” and when you’d cleared the shampoo from your eyes you’d see a floating window in which your smiling Simulator would appear and introduce herself. Or maybe this revelation would happen on a worldwide scale, with giant windows and a booming voice surrounding the planet, announcing that there is in fact an All Powerful Programmer up in the heavens. But even if your Simulator shied away from exhibitionism, less obvious clues might turn up.
Simulations allowing for sentient beings would certainly have reached a minimum fidelity threshold, but as they do with designer clothes and cheap knockoffs, quality and consistency would likely vary. For example, one approach to programming simulations—call it the “emergent strategy”—would draw on the accumulated mass of human knowledge, judiciously invoki
ng relevant perspectives as dictated by context. Collisions between protons in particle accelerators would be simulated using quantum field theory. The trajectory of a batted ball would be simulated using Newton’s laws. The reactions of a mother watching her child’s first steps would be simulated by melding insights from biochemistry, physiology, and psychology. The actions of governmental leaders would fold in political theory, history, and economics. Being a patchwork of approaches focused on different aspects of simulated reality, the emergent strategy would need to maintain internal consistency as processes nominally construed to lie in one realm spilled over into another. A psychiatrist needn’t fully grasp the cellular, chemical, molecular, atomic, and subatomic processes underlying brain function—which is a good thing for psychiatry. But in simulating a person, the challenge for the emergent strategy would be to consistently meld coarse and fine levels of information, ensuring for example that emotional and cognitive functions interface sensibly with physiochemical data. This kind of cross-border meshing takes place in all phenomena and has always compelled science to seek deeper, more unified explanations.
Simulators employing emergent strategies would have to iron out mismatches arising from the disparate methods, and they’d need to ensure that the meshing was smooth. This would require fiddles and tweaks which, to an inhabitant, might appear as sudden, baffling changes to the environment with no apparent cause or explanation. And the meshing might fail to be fully effective; the resulting inconsistencies could build over time, perhaps becoming so severe that the world became incoherent, and the simulation crashed.
A possible way to obviate such challenges would be to use a different approach—call it the “ultra-reductionist strategy”—in which the simulation would proceed by a single set of fundamental equations, much as physicists imagine is the case for the real universe. Such simulations would take as input a mathematical theory of matter and the fundamental forces and a choice of “initial conditions” (how things were at the starting point of the simulation); the computer would then evolve everything forward in time, thereby avoiding the meshing issues of the emergent approach. But simulations of this kind would encounter their own computational problems, even beyond the staggering computational burden of simulating “everything,” right down to the behavior of individual particles. If the equations our descendants have in their possession are similar to those we work with today—involving numbers that can vary continuously—then the simulations would necessarily invoke approximations. To exactly follow a number as it varies continuously, we would need to track its value to an infinite number of decimal places (for instance, as such a quantity varies, say, from .9 to 1, it would pass through numbers like .9, .95, .958, .9583, .95831, .958317, and on and on, with an arbitrarily large number of digits required for full accuracy). That’s something a computer with finite resources can’t manage: it will run out of time and memory. So, even if the deepest equations were used, it’s still possible that computer-based calculations would inevitably be approximate, allowing errors to build up over time.*
Of course, by “error” I mean a deviation between what occurs in the simulation and the description inherent in the most refined physical theories the simulator has at his or her disposal. But to those like you who are within the simulation, the mathematical rules driving the computer would be your laws of nature. The issue, then, is not how closely the mathematical laws used by the computer model the external world; we’re imagining that you don’t observe the external world from within the simulation. Rather, the problem for a simulated universe is that when a computer’s necessary approximations permeate otherwise exact mathematical equations, calculations easily lose their stability. Round-off errors, when accumulated over a great many computations, can yield inconsistencies. You and other simulated scientists might witness anomalous results from experiments; cherished laws might start yielding inaccurate predictions; measurements that had long since converged on a single widely confirmed result might start producing different answers. For long stretches, you and your simulated colleagues would think that you’d encountered evidence, much as your forebears had throughout the previous centuries and millennia, that your final theory wasn’t so final after all. Collectively, you’d closely reexamine the theory, perhaps coming up with new ideas, equations, and principles that better described the data. But, assuming the inaccuracies didn’t result in contradictions that crashed the program, at some point you’d hit a wall.
After an exhaustive search through possible explanations, none of which was able to fully explain what was happening, an iconoclastic thinker might suggest a radically different idea. If the continuum laws that physicists had developed over many millennia were input to a powerful digital computer and used to generate a simulated universe, the errors built up from the inherent approximations would yield anomalies of the very kind being observed. “Are you suggesting that we’re in a computer simulation?” you’d ask. “Yes,” your colleague would answer. “Well, that’s nutty,” you’d say. “Really?” she’d reply. “Take a look.” And she’d produce a monitor showing a simulated world, which she had programmed using those very same deep laws of physics, and—catching your breath after the shock of encountering a simulated world at all—you would see that the simulated scientists were indeed puzzling over the very same kind of strange data that troubled you.6
A Simulator who sought more assiduously to conceal herself could, of course, use more aggressive tactics. As inconsistencies started to build, she might reset the program and erase the inhabitants’ memory of the anomalies. So it would seem a stretch to claim that a simulated reality would reveal its true nature through glitches and irregularities. And certainly I’d be hard pressed to argue that inconsistencies, anomalies, unanswered questions, and stalled progress would reflect anything more than our own scientific failings. The sensible interpretation of such evidence would be that we scientists need to work harder and be more creative in seeking explanations. However, there is one serious conclusion that emerges from the fanciful scenario I’ve told. If and when we do generate simulated worlds, with apparently sentient inhabitants, an essential question will arise: Is it reasonable to believe that we occupy a rarefied place in the history of scientific-technological development—that we have become the very first creators of sentient simulations? We may have—but if we’re keen to go with the odds, we must consider alternative explanations that, in the grand scheme of things, don’t require us to be so extraordinary. And there is a ready-made explanation that fits the bill. Once our own work convinces us that sentient simulations are possible, the guiding principle of “garden variety,” discussed in Chapter 7, suggests that there’s not just one such simulation out there but a swarming ocean of simulations, which constitute a Simulated Multiverse. While the simulation we’ve created might be a landmark feat in the limited domain to which we have access, within the context of the entire Simulated Multiverse it’s nothing special, having been achieved a gazillion times over. Once we accept that idea, we’re led to consider that we too may be in a simulation, since that’s the status of the vast majority of sentient beings in a Simulated Multiverse.
Evidence for artificial sentience and for simulated worlds is grounds for rethinking the nature of your own reality.
The Library of Babel
During my first semester in college, I enrolled in an introductory philosophy course taught by the late Robert Nozick. From the very first lecture, it was a wild ride. Nozick was completing his voluminous Philosophical Explanations; he used the course as a dress rehearsal for many of the book’s central arguments. Just about every class shook my grasp on the world, sometimes vigorously. This was an unexpected experience—I’d thought that upending reality would be the purview solely of my physics courses. Yet, there was an essential difference between the two. The physics lectures challenged comfortable views by exposing strange phenomena that arise in wholly unfamiliar realms where things move fast, are extremely heavy, or are fantastically tiny.
The philosophy lectures shook comfortable views by challenging the foundations of everyday experience. How do we know there’s a real world out there? Should we trust our perceptions? What thread binds our molecules and atoms to preserve our personal identity through time?
While I was hanging around after class one day, Nozick asked me what I was interested in, and I brazenly told him that I wanted to work on quantum gravity and unified theories. This was generally a conversation stopper, but for Nozick it presented a chance to educate a young mind by revealing a new perspective. “What drives your interest?” he asked. I told him that I wanted to find eternal truths, to help understand why things are the way they are. Naïve and blustery, for sure. But Nozick listened graciously and then took the idea further. “Let’s say you find the unified theory,” he said. “Would that really provide the answers you’re looking for? Wouldn’t you still be left asking why that particular theory, and not another, was the correct theory of the universe?” He was right, of course, but I replied that in the search for explanations there might come a point when we would just have to accept certain things as given. That was just where Nozick wanted me to go; in writing Philosophical Explanations he had developed an alternative to this view. It’s based on what he called the principle of fecundity and is an attempt to frame explanations without “accepting certain things as given”; without, as Nozick explains it, accepting anything as brute-force truth.