by Brian Greene
Physics protects us, but the price for safety is total separation from our handiwork. And that’s the good news.
The bad news for aspiring universe creators is a more sobering result derived by Guth and his MIT colleague Edward Farhi. Their careful mathematical treatment showed that the sequence depicted in Figure 10.1 requires an additional ingredient. Much as some balloons require that you give a strong initial burst of air, after which they more easily inflate, Guth and Farhi found that the nascent universe in Figure 10.1 needs a strong kick-start to get the inflationary expansion off and running. So strong that there’s only one entity that can provide it: a white hole. A white hole, the opposite of a black hole, is a hypothetical object that spews matter out rather than drawing it in. This requires conditions so extreme that known mathematical methods break down (much as is the case at the center of a black hole); suffice it to say, no one anticipates generating white holes in the laboratory. Ever. Guth and Farhi found a fundamental wrench in the universe-creation works.
A number of research groups have since suggested possible ways of skirting the problem. Guth and Farhi, joined by Jemal Guven, found that by creating the inflationary seed through a quantum tunneling process (similar to what we discussed in the context of the Landscape Multiverse) the white hole singularity can be avoided; but the probability for the quantum tunneling process to happen is so fantastically small that there’s essentially no chance of its happening over timescales that anyone would consider worth contemplating. A group of Japanese physicists, Nobuyuki Sakai, Ken-ichi Nakao, Hideki Ishihara, and Makoto Kobayashi, showed that a magnetic monopole—a hypothetical particle that has either the north pole or the south pole of a standard bar magnet—might catalyze inflationary expansion, also avoiding singularities; but after nearly forty years of intense searching, no one has yet found a single one of these particles.*
As of today, then, the summary is that the door to creating new universes remains open, but only barely. Given the proposals’ heavy reliance on hypothetical elements, future developments may well shut this door permanently. But if they don’t—or, perhaps, if subsequent work makes a stronger case for the possibility of universe creation—would there be motivation to proceed? Why create a universe if there’s no way to see it, or interact with it, or even know for sure that it was created? Andrei Linde, famous not just for his deep cosmological insights but also for his flair for mock drama, has noted that the allure of playing god would simply prove irresistible.
I don’t know that it would. Admittedly, it would be thrilling to have so thoroughly grasped nature’s laws that we could reenact the most pivotal of all events. I suspect, however, that by the time we can seriously consider universe creation—if that time ever comes—our scientific and technical advancements would have made available so many other spectacular undertakings, whose results we could not just imagine but truly experience, that the intangible nature of universe creation would make it much less interesting.
The appeal would surely be stronger were we to learn how to manufacture universes that we could see or even interact with. For “real” universes, in the usual sense of a universe constituted from the standard ingredients of space, time, matter and energy, we don’t yet have any strategy for doing so that’s compatible with the laws of physics as we currently understand them.
But what if we set aside real universes and consider virtual ones?
The Stuff of Thought
A couple of years ago, I had a bout of feverish flu that came with hallucinations far more vivid than any ordinary dream or nightmare. In one that has stayed with me, I’d find myself with a group of people sitting in a sparse hotel room, locked in a hallucination within the hallucination. I was absolutely certain that days and weeks went by—until I was thrust back into the primary hallucination, where I’d learn, shockingly, that hardly any time had passed at all. Each time I felt myself drifting back to the room, I resisted strenuously, since I knew from previous iterations that once there I’d be swallowed whole, unable to recognize the realm as false until I found myself back in the primary hallucination, where I’d again be distraught to learn that what I’d thought real was illusory. Periodically, when the fever subsided, I’d pull out one level further, back to ordinary life, and realize that all those translocations had been taking place within my own swirling mind.
I don’t usually learn much from having a fever. But this experience added immediacy to something which, to that point, I’d largely understood only in the abstract. Our grip on reality is more tenuous than day-today life can lead us to believe. Modify normal brain function just a bit, and the bedrock of reality may suddenly shift; though the outside world remains stable, our perception of it does not. This raises a classic philosophical question. Since all of our experiences are filtered and analyzed by our respective brains, how sure are we that our experiences reflect what’s real? In the framing philosophers like to use: How do you know you’re reading this sentence, and not floating in a vat on a distant planet, with alien scientists stimulating your brain to produce the thoughts and experiences you deem real?
These issues are central to epistemology, a philosophical subfield that asks what constitutes knowledge, how we acquire it, and how sure we are that we have it. Popular culture has brought these scholarly pursuits to a wide audience in films such as The Matrix, The Thirteenth Floor, and Vanilla Sky, tussling with them in entertaining and thought-provoking ways. So, in looser language, the question we’re asking is: How do you know you’re not hooked into the Matrix?
The bottom line is that you can’t know for sure. You engage the world through your senses, which stimulate your brain in ways your neural circuitry has evolved to interpret. If someone artificially stimulates your brain so as to elicit electrical crackles exactly like those produced by eating pizza, reading this sentence, or skydiving, the experience will be indistinguishable from the real thing. Experience is dictated by brain processes, not by what activates those processes.
Going a step further, we can consider dispensing with the sloppiness of biological material altogether. Might all your thoughts and experiences be nothing more than a simulation that leverages software and circuitry sufficiently elaborate to mimic ordinary brain function? Are you convinced of the reality of flesh, blood, and the physical world, when actually your experience is only a crowd of electrical impulses firing through a hyper-advanced supercomputer?
An immediate challenge in considering such scenarios is that they easily set off a spiraling skeptical collapse; we wind up trusting nothing, not even our powers of deductive reasoning. My first response to questions like the ones just posed is to work out how much computer power you’d need to stand a chance of simulating a human brain. But if I am indeed part of such a simulation, why should I believe anything I read in neurobiology texts? The books would be simulations too, written by simulated biologists, whose findings would be dictated by the software running the simulation and thus could easily be irrelevant to the workings of “real” brains. The very notion of a “real” brain might itself be computer-generated artifice. Once you can’t trust your knowledge base, reality quickly sails to sea.
We’ll return to these concerns, but I don’t want them to sink us—at least, not yet. So, for the time being, let’s drop anchor. Imagine that you are real flesh and blood, and so am I, and that everything you and I take to be real, in the everyday sense of the term, is real. With all that assumed, let’s take up the question of computers and brainpower. What, roughly, is the processing speed of the human brain, and how does it compare with the capacity of computers?
Even if we are not stuck in a skeptical morass, this is a difficult question. Brain function is largely an uncharted territory. But just to get a glimpse of the terrain, however foggy, consider some numbers. The human retina, a thin slab of 100 million neurons that’s smaller than a dime and about as thick as a few sheets of paper, is one of the best-studied neuronal clusters. The robotics researcher Hans Moravec has estimated th
at for a computer-based retinal system to be on a par with that of humans, it would need to execute about a billion operations each second. To scale up from the retina’s volume to that of the entire brain requires a factor of roughly 100,000; Moravec suggests that effectively simulating a brain would require a comparable increase in processing power, for a total of about 100 million million (1014) operations per second.4 Independent estimates based on the number of synapses in the brain and their typical firing rates yield processing speeds within a few orders of magnitude of this result, about 1017 operations per second. Although it’s difficult to be more precise, this gives a sense of the numbers that come into play. The computer I’m now using has a speed that’s about a billion operations per second; today’s fastest supercomputers have a peak speed of about 1015 operations per second (a statistic that no doubt will quickly date this book). If we use the faster estimate for brain speed, we find that a hundred million laptops, or a hundred supercomputers, approach the processing power of a human brain.
Such comparisons are likely naïve: the mysteries of the brain are manifold, and speed is only one gross measure of function. But most everyone agrees that one day we will have raw computing capacity equal to, and likely far in excess of, what biology has provided. Futurists contend that such technological leaps will yield a world so far beyond familiar experience that we lack the capacity to imagine what it will be like. Invoking an analogy with phenomena that lie outside the bounds of our most refined physical theories, they call this visionary roadblock a singularity. One broad-brush prognosis holds that the surpassing of brainpower by computers will completely blur the boundary between humans and technology. Some anticipate a world run rampant with thinking and feeling machines, while those of us still based in old-fashioned biology routinely upload our brain content, safely storing knowledge and personalities in silico, complete with backup drives, for unlimited durations.
This vision may well be hyperbolic. There’s little dispute regarding projections of computer power, but the obvious unknown is whether we will ever leverage such power into a radical fusion of mind and machine. It’s a modern-day question with ancient roots; we’ve been thinking about thinking for thousands of years. How is it that the external world generates our internal responses? Is your sensation of color the same as mine? How about your sensations of sound and touch? What exactly is that voice we hear in our heads, the stream of internal chatter we call our conscious selves? Does it derive from purely physical processes? Or does consciousness arise from a layer of reality that transcends the physical? Penetrating thinkers through the ages, Plato and Aristotle, Hobbes and Descartes, Hume and Kant, Kierkegaard and Nietzsche, James and Freud, Wittgenstein and Turing, among countless others, have tried to illuminate (or debunk) processes that animate the mind and create the singular inner life available through introspection.
A great many theories of mind have emerged, differing in ways significant and subtle. We won’t need the finer points, but just to get a feel for where the trails have led, here are a few: dualist theories, of which there are many varieties, maintain that there’s an essential nonphysical component vital to mind. Physicalist theories of mind, of which there are also many varieties, deny this, emphasizing instead that underlying each unique subjective experience is a unique brain state. Functionalist theories go further in this direction, suggesting that what really matters to making a mind are the processes and functions—the circuits, their interconnections, their relationships—and not the particulars of the physical medium within which these processes take place.
Physicalists would largely agree that were you to faithfully replicate my brain by whatever means—molecule by molecule, atom by atom—the end product would indeed think and feel as I do. Functionalists would largely agree that were you to focus on higher-level structures—replicating all my brain connections, preserving all brain processes while changing only the physical substrate through which they occur—the same conclusion would hold. Dualists would largely disagree on both counts.
The possibility of artificial sentience clearly relies on a functionalist viewpoint. A central assumption of this perspective is that conscious thought is not overlaid on a brain but rather is the very sensation generated by a particular kind of information processing. Whether that processing happens within a three-pound biological mass or within the circuits of a computer is irrelevant. The assumption could be wrong. Maybe a bundle of connections needs a substrate of wrinkled wet matter if it’s to gain self-awareness. Maybe you need the actual physical molecules that constitute a brain, not just the processes and connections those molecules facilitate, if conscious thought is to animate the inanimate. Maybe the kinds of information processing that computers carry out will always differ in some essential way from brain functioning, preventing the leap to sentience. Maybe conscious thought is fundamentally nonphysical, as claimed by various traditions, and so lies permanently beyond the reach of technological innovation.
With the rise of ever more sophisticated technologies, the questions have become sharper and the pathway toward answers more tangible. A number of research groups have already taken the initial steps toward simulating a biological brain on a computer. For example, the Blue Brain Project, a joint venture between IBM and the École Polytechnique Fédérale in Lausanne, Switzerland, is dedicated to modeling brain function on IBM’s fastest supercomputer. Blue Gene, as the supercomputer is called, is a more powerful version of Deep Blue, the computer that triumphed in 1997 over the world chess champion Garry Kasparov. Blue Brain’s approach is not all that different from the scenarios I just described. Through painstaking anatomical studies of real brains, researchers are gathering ever more precise insight into the cellular, genetic, and molecular structure of neurons and their interconnections. The project aims to encode such understanding, for now mostly at the cellular level, in digital models simulated by the Blue Gene computer. To date, researchers have drawn on results from tens of thousands of experiments focused on a pinhead-sized section of a rat brain, the neocortical column, to develop a three-dimensional computer simulation of roughly 10,000 neurons communicating through some 10 million interconnections. Comparisons between the response of a real rat’s neocortical column and the computer simulation to the same stimuli show an encouraging fidelity of the synthetic model. This is far from the 100 billion neurons firing away in a typical human head, but the project’s leader, the neuroscientist Henry Markram, anticipates that before 2020 the Blue Brain Project, leveraging processing speeds that are projected to increase by a factor of more than a million, will achieve a full simulated model of the human brain. Blue Brain’s goal is not to produce artificial sentience, but rather to have a new investigative tool for developing treatments for various forms of mental illness; still, Markram has gone out on a limb to speculate that, when completed, Blue Brain may very well have the capacity to speak and to feel.
Regardless of the outcome, such hands-on explorations are pivotal to our theories of mind; I’m quite certain that the issue of which, if any, of the competing perspectives are on target cannot be settled through purely hypothetical speculation. In practice, too, challenges are immediately evident. Suppose a computer one day professes to be sentient—how would we know whether it really is? I can’t even verify such claims of sentience when made by my wife. Nor she with me. That’s a burden arising from consciousness being a private affair. But because our human interactions yield abundant circumstantial evidence supporting the sentience of others, solipsism quickly becomes absurd. Computer interactions may one day reach a similar point. Conversing with computers, consoling and cajoling them, may one day convince us that the simplest explanation for their apparent conscious self-awareness is that they are indeed conscious and self-aware.
Let’s take a functionalist viewpoint, and see where it leads.
Simulated Universes
If we ever create computer-based sentience, some would likely implant the thinking machines in artificial human bodies, creatin
g a mechanical species—robots—that would be integrated into conventional reality. But my interest here is in those who would be drawn by the purity of electrical impulses to program simulated environments populated by simulated beings that would exist within a computer’s hardware; instead of C-3PO or Data, think Sims or Second Life, but with inhabitants who have self-aware and responsive minds. The history of technological innovation suggests that iteration by iteration, the simulations would gain verisimilitude, allowing the physical and experiential characteristics of the artificial worlds to reach convincing levels of nuance and realism. Whoever was running a given simulation would decide whether the simulated beings knew that they existed within a computer; simulated humans who surmised that their world was an elaborate computer program might find themselves taken away by simulated technicians in white coats and confined to simulated locked wards. But probably the vast majority of simulated beings would consider the possibility that they’re in a computer simulation too silly to warrant attention.
You may well be having that very reaction right now. Even if you accept the possibility of artificial sentience, you may be persuaded that the overwhelming complexity of simulating an entire civilization, or just a smaller community, renders such feats beyond computational reach. On this point, it’s worth looking at some more numbers. Our distant descendants will likely fashion ever-larger quantities of matter into vast computing networks. So allow imagination free rein. Think big. Scientists have estimated that a present-day high-speed computer the size of the earth could perform anywhere from 1033 to 1042 operations per second. By comparison, if we assume that our earlier estimate of 1017 operations per second for a human brain is on target, then an average brain performs about 1024 total operations in a single hundred-year life span. Multiply that by the roughly 100 billion people who have ever walked the planet, and the total number of operations performed by every human brain since Lucy (my archaeology friends tell me I should say “Ardi”) is about 1035. Using the conservative estimate of 1033 operations per second, we see that the collective computational capacity of the human species could be achieved with a run of less than two minutes on an earth-sized computer.