Question 1: What do you consider to be the three most common criticisms of the concept of a technological singularity? And which of these seem to be coming from a motivation to rationally understand this phenomenon?
Common criticisms of the technological singularity are essentially arguments against a continued double exponential growth in computational complexity, in either hardware or software, and the rapid approach of that complexity to human equivalency. I'm not sure what the three most common criticisms are within the general population, but among those futurists, independent scholars, and academics who have discussed these issues with me at http://www.SingularityWatch.com, here are some of the more intriguing counterarguments I've heard:
1) all trend curves must eventually stop, even the growth of computation, because growth in any particular population or technology is always "S curved," meaning that it eventually saturates, after an early exponential phase (the lower half of the "S")
2) human complexity is vastly greater than we anticipate, and it's probably going to take many centuries before we even begin to approach human level complexity within our computer systems, and
3) technology always requires human designers, and yet the argument in #1 still applies: humans have for several years now been overwhelmed by even the limited technological complexity we've created (i.e., the recent rise in software unmanageability, from the human perspective), so the marginal rate of development of total computational complexity is actually slowing, and might even slow down further, for many years to come.
These are fascinating propositions, and I don't think anyone who has used technology in meaningful ways is coming from a position of ignorance when discussing its rate of change. We've all got a deep fund of past experience to draw upon, from a wide range of backgrounds. Certainly those of us who are a bit older have noticed the 'technological quickening' across a number of environmental domains (even as we wonder if this is simply a sign of our approaching old age!). Kids, by contrast have always expected miracles, and yet perhaps never more so than in our current era. Teasing out the true rates and trajectory of technological change, through research, writings, professional associations, and organized conferences, is one of the goals of my own organization.
With regard to the three arguments above, I think #1 can be refuted by a careful study of computational growth as a generalized process. Yes, particular technologies always saturate, but information processing itself seems to continually jump to new substrates. We've got all kinds of laws that suggest we'll hit limits, such as Rock's Law, which projects that at their rate of current cost growth, chip fab plants will be too expensive for the future economy to support around 2020 (about the same time that Moore's Law itself is supposed to peter out). But we've continually seen new paradigms emerge which allow vastly more computational capacity at any given level of resources (matter, energy, space, money) and miniaturization over time. It's as if the universe has a special structure which facilitates exponentially more computation within its local microarchitecture. ("Computational turtles all the way down" ?)
We can always find saturation points in various trends: we stopped the commercial shuttling of humans faster than the Concorde around the mid-1970's, but we can't use that analogy to imply the same will happen in the computer industry. Those of us who've studied transportation futures know we could have pushed on to create Planetran (a vaccum maglev train technology that would eventually allow us to commute between any major city in the world in less than 30 minutes, at 1-2 G's accel/decel) but we didn't see any need for the horrendous economic investment that would have allowed that continued exponentiation. Why? Because the information flow simply jumped substrate—we found that TV/fax/conference calls gave us a 'reasonable first approximation' to being there in the flesh, so our global brain, our emerging species consciousness, concentrated on exponentially increasing the quality and quantity of that information flow instead, simulating physical travel in a computational matrix. So the acceleration never stopped, and the growth rate of these computational sims has never saturated. Those who have studied computer design also know that there are a raft of faster and more complex technologies waiting in the wings when the integrated circuit supposedly hits a 'Moore's Law Limit' circa 2020. More likely now, its looking like the IC's are just going to continue to morph into something more complex, at the microstructural level, long before we approach the transistor's silicon miniaturization barrier. Have you heard of HBT's? Spintronics? Photonics? Quantum cascade lasers? Get ready to learn some new vocabulary, friend.
With regard to #2, I don't think this is the case, either. One of the amazing things is just how finite, bounded, and, simulable the human organism is becoming. We know the maximum number of neurons that could fit in the brainspace. We've got a good handle on their arborization potential, and are closing in on their morphological limitations. We know most of the neurotransmitters. We've got an inventory of the different cell types (about 250, equivalent to the number of countries on the planet). There are perhaps 100,000 relevant neural proteins (and around 300,000 in the entire human proteome). We've synthesized the human genome, expect to have the human proteome sequenced in another five years (!!), and are now developing chip assays to probe both the transcriptome and the proteome in vivo (not all transcription products are expressed as proteins at any particular time, this is part of the complex cellular signal processing and transduction story). If we approximately solve the protein folding problem this decade, we may have a reasonable facsimile of the metabolome (a 3D model of what these all these genes and proteins are doing, in human biochemistry) beginning about ten to fifteen years out. No, the human organism is appearing all too finite, in relation to continually hyperexponentiating computational power. And as Ray Kurzweil (The Age of Spiritual Machines: When Computers Exceed Human Intelligence, 1999) says, even if the Penrose-Hameroff microtubule/quantum computing hypothesis is true (proposing that humans think with quantum devices—highly doubtful), that doesn't change the finity of the situation. Fat fingered 20th century humans have even managed to make crude quantum computational devices. Fancy that. If there's a millionfold more unappreciated complexity at the bottom, we'll still close the remaining gap in less than 20 more years, at present complexity growth rates (and perhaps less than 10, given that present growth rates are themselves accelerating). Humans are becoming a computational end game.
I find #3 to be one of the more clever arguments, but it still doesn't seem to hold water. Wherever you look, technological evolution is becoming exponentially more human independent. Humans catalyze technological development, but we are less and less needed the more sophisticated our tools (the technological substrate) becomes. As an example, several working ASICs have recently been designed, using evolutionary computation methods, entirely without human intervention. And while human-written software has been in saturation mode for years, we're even discovering some ways around that. There are minor rejigs, as in the way our software is starting to standardize itself (object oriented programming), distribute itself (open source), and even surprise us in its continually more intimate links to the hardware (HP's new automatic PICO compiler: Program In, Chip Out). But there are major bypasses as well: by implementing technological self-improvement ever more faithfully at the hardware level, which has always been growing at a double exponential rate (for reasons I speculate on in my book), we seem to be figuring out how to get rapidly assisted by an inherent 'universal complexity potential' that appears built into the physics of small things, and is mostly independent of bottlenecks in human creativity.
Consider the amazing performance gains possible the more you stay at the hardware level: a detailed simulation of a natural neuron can take a few minutes to run in current software. The same process happens biologically in milliseconds. But when you implement the ANN entirely in hardware, it can recapitulate neural processing more than ten million times faster than wetware. It's not difficult to see what we're heading for: reconfigurable hardware that is intrinsically complex, in an associational and developmental sense, and continually adaptive and scalable. If you follow the recent advances in evolvable hardware as I do, you might agree that we're finally getting close enough to smell the breakthroughs.
Question 2: You seem to have a somewhat esoteric take on the concept of the singularity. How and when did you first become acquainted with the topic?
I was around 12 years old, doing my first serious thinking about "the SETI question" (also known as Fermi's Paradox) for a biology class, when it suddenly struck me why we hadn't heard from any putative ubiquitous more advanced neighboring technological civilizations. As I learned later, I'd deduced (more likely, induced) the "transcension scenario," still one that is overlooked, both by many astrobiologists and by most serious thinkers on accelerating change. In transcension, advanced systems simply leave local spacetime once they reach a certain point in their development. Put another way, they are proposed to change to a shape that is nearly indistinguishable to less computationally advanced organisms, the way the complexity of a human is nearly indistinguishable to a bee, or a bacterium.
To me, it suddenly seemed possible that this shape might be a certain class of black hole, and I have more to say about that in my forthcoming book. But this isn't a popular interpretation of the future of very complex systems, and even Kurzweil misses addressing the transcension scenario in his latest 60 page précis, The Singularity is Near. Notwithstanding this minor criticism, his excellent work should be considered required reading for lay futurists, and is available at http://www.kurzweilai.net/articles/art0134.html?printable=1. Investigating his site, http://www.KurzweilAI.net, is also well worth your time.
What came intuitively, even at that age, was the suspicion that the observable universe was itself probably another finite computational substrate that is eventually outgrown, in the same way that all somas (bodies, developmental environments) are eventually discarded by the germline (fundamental universal parameters, seeds of the cellular automata) in subsequent evolutionary cycles. This kind of 'germline tissue recursion through disposable somas' appears to be the way of all complex adaptive systems in universal development. Thus I began to think about the singularity from an early age from within a kind of 'universal developmental reference frame,' putting the technological singularity (accelerating technological change) within a cosmological and computational framework I call the 'developmental singularity,' many years before even Vernor Vinge began using the term. It was a great relief to me to finally bump into his essay ("The Coming Technological Singularity," 1993) in the mid 1990's, through the magic of the world wide web. Until that time, I'd searched in vain for individuals and books discussing these topics directly, and that search continues today, though it is now more systematic, and yields greater fruits every year.
I've since characterized a core set of memes that I think are important to what Damien Broderick (The Spike, 2001) and I call "singularity studies," the systems theory approach to investigating accelerating computational change. I've organized them by an acronym, AMDEC, and as you might imagine, not everyone considers all the AMDEC memes important to the topic of singularity studies. To my knowledge, Vinge, Kurzweil, Broderick, and others don't even discuss the "MD" memes in AMDEC yet, but hope that will change in coming years. For my part, I consider them all equally important to understanding the whole puzzle. Don't hold me to just these five, either: I expect I'll add more letters to this acronym over time, and perhaps change a few, too, as the picture becomes clearer. That's part of the fun of the quest. Here they are:
Accelerating Change/Growth
Metrics
MEST Compression/Free
Energy Rate Density
Developmental Cosmology/Dev.
Biology/Cyclic Tuning of CAS Parameters
Evolutionary Computation/Human-Independent
Technological Evolution
Complexity Studies/Complex
Adaptive Systems Theory
I don't have time to dig further into these memes in this interview, but you can find out more about them on main page of my site and also in the "Advanced Topics" section, where I've currently relegated the more speculative "MD" memes (woo woo science about computational matter, energy, space, and time compression, black holes, and multiverses), until such time as I can explain them better for critical review. Enjoy!
Question 3: Many futurists see a link between a technological singularity, and molecular nanotechnology (MNT). Some argue a technological singularity will lead inevitably to MNT. Others argue that human-initiated MNT will create a singularity. What's your take on this debate?
I'm with the first group. To my mind, Erik Drexler did a tremendously visionary job in expressing the potentials of a full blown nanotechnology in Engines of Creation, in 1986. That was a worldview-shifting book, and such a clever achievement: deconstructing our biological assemblers, and showing us the promise of reappropriating this technology to conscious, rational ends. At the same time, I've always had the perspective that most of the Engines vision would first require the emergence of autonomous AI to implement.
But at the same time I was extremely encouraged to see his work on nanocomputation in Nanosystems, 1992, because I think that is the bridge between the various camps on this issue. We all seem to agree that nanocomputational breakthroughs at the hardware level will be very important. If any of us disagree, it's primarily on how important: I see such continued breakthroughs as critically necessary to create the emergent AI that will lead to a full Drexlerian nano. At the same time, I'm sure we'll have weaker versions continue to explode on the playing field, and see a lot of money be made before our highly programmable, super-fast assemblers arrive. And I'm counting on the good folks at Atomasoft, Foresight, and a few other organizations to help us get there, too. ?
It's also great to see technologically advanced governments in the U.S. and Japan finally seriously funding (over a billion dollars, collectively) long-term, blue-sky nanotechnology research at a substantial level (even if most of it is a materials science implementation that Drexler might not yet consider worthy of the term). I see this new level of investment as another soft sign that we are on track to a singularity relatively soon.
Bottom line, it seems to me that mesocomputation in all its forms (MEMS, microphotonics/fluidics, etc.) will lead us into nanocomputation (a quantum cascade laser, or HEM transistor could each, arguably, already be considered shipping nanocomputational devices—actually, shipping femtocomputational devices, but that's a story for another time), and ever more versatile meso, nano and femto computational devices will be part of the scalable hardware substrate that will likely revolutionize evolutionary computation (evcomp) sometime in the coming century. I think evcomp will move from the minor paradigm position it occupies now, with perhaps 50,000 adherents doing productive and powerful but primarily software implementations, to the inexorable juggernaut of tomorrow, where the majority of computation will include a core of evolvable hardware, tended by humans serving as 'digital gardeners,' with only a dim appreciation of these system's true internal complexity or potential. We're rapidly heading that way now, of course. When was the last time you tried to fix a modern car engine?
Question 4: In Vernor Vinge's internet essay The Coming Technological Singularity (1993), he states that he would be surprised if the singularity occurs before 2005 or after 2030. What is your current estimate as to when the technological singularity will occur?
For my website and book I've accumulated a range of published documents on this topic. It's my interpretation that most of the careful thinkers on the topic, defining the technological singularity as a point of post-human-equivalent "takeoff" of technological complexity, seem to congregate around 2040, in a broad range between 2020 and 2060. That interval includes Vinge, Kurzweil, Moravec, Minsky, and some other good company, and even allows for the contrarians who propose some still unforeseen snags on the way there, such as human brains computing on microtubules, using quantum effects.
Yet it is interesting to note that this range leaves out a number of individuals as well. On the aggressive end, it doesn't include well-meaning folks like Nick Hoggard, who has written a paper and a 60 page précis proposing 2001-2004. It also excludes the 2012 Mayan calendrics contingent, like Terrence McKenna. Eliezer Yudkowsky arguably squeaks in, as he anticipates 2008 to 2020, last I checked. Some of these individuals, though they have added greatly to the debate, are coming from a perspective of either youthful ebullience or a narrow education in information technologies, or both. They also often evince an attitude of technological determinism (TD) that indicates a lack of understanding of the social complexities of our still-human-catalyzed technological evolution. As Langdon Winner has said, using TD to explain the world is like attempting to describe all instances of sexual intercourse using only the concept of rape: it misses the whole realm of human consensuality, limits, balances, and negotiated paths, even if we are all heading toward an ultimately inevitable emergence. Proponents of TD also seem to currently underestimate the complexity of the network conditions that are going to have to be in place to precipitate truly autonomous technological change. A number of them also anticipate a "hard takeoff" scenario, the lone romantic savior of the world toiling away in the lab making the AI, that I don't think fits at all with past evidence of complex emergences. This said, I don't want to paint all those who are committed to "making the singularity happen" with a broad brush of criticism. Some of these individuals, such as Brian Atkins of the 'Singularity Institute for Artificial Intelligence', are moving beyond a simplistic activism into investigation of the deeper forces and social context of the choices available to us. Thanks, Brian.
On the conservative end, a 2020-2060 range also excludes some sober scholarly work by the likes of Richard Coren, who's written an interesting book, The Evolutionary Trajectory, predicting a 'universal singularity' in 2140 (though its possible to interpret his data as allowing a technological singularity circa 2040). It also currently excludes Robin Hansen, whose latest prediction, based on economic trends, is circa 2150. Neither author has written on the amazing advances in evolutionary computation in recent decades, and I wonder if they have a sense of just how human-independent computer evolution is starting to become. It has also not escaped my notice that both individuals make their projections from within the social structure of academia. A degree of professional conservatism, placing the disrupting event safely beyond current expected lifespans, may be in play.
Certainly my best current
projected range of 2020-2060 is voodoo like anyone else's, but I'm satisfied
that I've done a good literature search on the topic, and perhaps a deeper
polling of the collective intelligence on this issue than I've seen elsewhere
to date. To me, estimates much earlier than 2020 are unjustified in their
optimism, and likewise, estimates after 2060 seem oblivious to the full
scope and power of the AMDEC processes in the universe.
Question 5: Bill Joy, and less forcefully, a few other individuals in the technological sphere have recently argued that we shouldn't try to develop intelligent machines. At the same time, some physicists (Penrose) and philosophers (Searle) argue that we can't develop genuine intelligence using conventional computer technology. What is your response to these critics?
I've discussed Penrose earlier, and will pass on Searle at present, if you don't mind. With regard to Joy, I've recently written a few essays (both "Complex Immunity" and "Evolutionary Computation in the Universe" are relevant here) to be published on my site and elsewhere in a few months in response to Bill's Genetics/Nanotech/Robotics (GNR) limits/relinquishment proposal ("The Future Doesn't Need Us," Wired 4.00). His issues are excellent, and require a careful response on several levels.
In a nutshell, I think he and several other very smart folks out there haven't yet done a careful study of the nature of immune systems, not just in biological systems (though they may be our best model) but in all complex adaptive systems (CAS's) we've had the opportunity to study. One of the CAS Laws I've tentatively formulated is: "Immune systems always win." Basically, this means that complex systems can, in a general, statistical sense, always protect themselves from the destabilizing effects of simple systems, but often choose not to do so because regular minor destabilization has a significant constructive effect.
An increasing number of life scientists understand this idea. They are, for example, explaining the tremendous value for human immune systems to remain semipermeable to bacteria (we use them commensally in our gut, we capture them to create our cellular organelles, like the mitochondria) and to viruses (much of the human genome contains genes inserted via permissive retroviral "sex" between members of our species, and we even use a virus in the syncytiotrophoblast to aid in human embryonic development). The threat of having these open 'back doors' within our immune systems, which could make themselves statistically impermeable to simple invaders, is actually quite low, because we also have tremendous genetic variation (through SNP's, allelic recombination, and other means). The bottom line is that we are too complex a target for simple organisms to threaten in any fundamental way.
That didn't stop the US and the Soviet Union from searching fruitlessly for 40 years for "super lethal" viruses and bacteria in their weapons labs. But they came up empty handed (don't believe the scaremongers on this point, believe instead the years of leaked reports and tell-alls on this shameful phase of our political history), because the bottom line is species-killers can't be created. Yes, a committed terrorist using modern tools of molecular virology might create a few million casualties, in a worst case, with a souped up ebola or anthrax, but the human organism is amazingly well defended, and six billion times redundant, at present. It is insanely hard to threaten any wide sample of complex organisms, which have both complex and simple redundant, mutually overlapping internal defense mechanisms (how else are they able to continually error correct to stay so complex?), with any threat as computationally simple as a bacterium or virus. This isn't to say that these catastrophies aren't terrible, and that we shouldn't strenuously seek to minimize them wherever possible. But it begins to place them in their proper perspective. All catastrophes are catalytic, and informational complexity is always preserved.
This helps us to better understand the true nature of the genetic threat, but what about nanotech/robotics? I often lump together these two, because I think it would take serious AI before both could become potential replicative threats (another way of putting this is: once we have the capability to do replicative nanotech or robotics at the hardware level, we will then concomitantly see emergent AI). We can again invoke the immune system argument, but this time, we are talking about global technological immune systems. In other words, I'm saying it would be impossible for any nanotech/robotic replicative system to develop without co-evolving an overpoweringly stabilizing and pervasive immune system in the process. Complex adaptive systems always emerge this way: an uncounted number of minor catalytic catastrophes during development ensures that this is so. Could you imagine a world of replicative nanodevices that didn't also have an associated nanoimmune system (Drexler's 'active shield' or 'blue goo'/ nanopolice) distributed everywhere to keep the chaos in check? A world without nano fire extinguishers to put out the 'blazes' when they erupt, and without a redundant, distributed, fault-tolerant architecture (look to the brain if you’d like an example), to make any local damage computationally irrelevant? Such concepts don't make CAS sense, to me.
Already we see a 'gridified' world around the corner, where every device will be plugged into the web, where all physical objects are ubiquitously interconnected. Could the future ever be any other way? I think not. Wasn’t the most lasting effect of the Oklahoma City and Columbine High violence to ratchet up the viligance and complexity of our cultural and technological immune systems? Doesn't this kind of balance always ensue, relentlessly? If you think not, I'd suggest reading Brin's Transparent Society. And a good lay book on immune systems.
The other thing that deserves mentioning is that complex systems are not only self-protecting, but also self-balancing and ever more able to both simulate and integrate with their surroundings. And all of these trends happen to a degree that is apparently in direct proportion to their complexity. This is a way of placing the immune systems metaphor in an even more universal context. How do we know any of this? We've seen a smooth increase in computational complexity throughout the entire known history of the universe. All catastrophes that you can point out, in whatever substrate, on whatever timescale, have never threatened more than a small fraction of the extant complex systems. If you're thinking big bang, or supernova, or K-T meteorite, or the fall of Greece, Rome, Egypt, or Maya, or the black plague, or WWII, or three mile island, or inner city violence, and don't see how each of these catastrophes represents a tiny catalytic fraction of the extant systems which are entering a new phase space, and also productively informs the immune systems of the majority in their failure, then please take a closer look at my forthcoming articles or book.
Question 6: I'm a bit concerned about this emergent A.I. idea. Aren't you saying that biological humans will quickly become a lower-complexity subsystem in relation to 'nano-time' machine intelligence? How many humans will be satisfied if they are aware that this is the situation? Should we expect that these machines will psychologically manipulate us into liking our new subjugation?
These are tremendously important issues, as they deal with the nature of our future humanity. Questions like these used to keep me up at night, but in recent years I've come to see them in less alarming terms. But don't let me convince you here—I hope you'll do your own research and see if you agree. Opening the debate is half the battle.
Yes, I do believe that present human complexity is about to be entirely outpaced, but remember, humans are always in a dynamic equilibrium with our tools. The more complex our tools get, the more we increase our own complexity, in a co-evolutionary process. Kurzweil has recently written eloquently on the coming merger between internet and humanity on his website, which he expects will grow increasingly 'seamless,' from our perspective. Why? Primarily because we are demanding that kind of interface. And we should increase the dialog on this process, not ignore it. Howard Bloom (Global Brain, 2000) also sees this transition very clearly. Our concept of self is a fluid entity, and it changes, expands, redefines as our technology improves. Pierre Baldi's recent work, The Shattered Self: The End of Natural Evolution, 2001 is well worth reading on this, as is Kenneth Gergen's The Saturated Self, 1992. Even language can be considered a technology, and once we developed it, it vastly changed our concept of self. Read Terrence Deacon's The Symbolic Species, 1997 for a great account. It's easy for us to see how written language constitutes a technology, but there's even a good case for thinking of oral language in this way. Amongst all the big brained species that have learned to use an oral proto-language, we are the only ones so far that managed to turn it into a craft, a practice, a way of synchronizing uttered sounds with experience, creating a new art form. And look what these algorithms, this new technology, did to us, collectively: it created both human culture and individual self-consciousness. Not bad.
The story of the coming emergence hasn't been written yet, but it would be my bet that, as with past technologies, it will involve a deep merger and co-evolution of the coming AI with those systems that catalyze and interact with it, namely, us. Situations where the AI puts us in a box, treats us as pets, ignores us, disassembles us for spare atoms, or any other such scripts seem to me to be like bad Sci Fi: serving our own fascination with drama more than describing the way the universe really seems to work.
If you want some better scripts for what's likely to happen, let me propose two. Perhaps the easiest one to see, but the one which seems by far the least likely to me, involves the A.I. distinctly 'waking up' somewhere, perhaps on the net, and for its first words uttering two sentences I've seen in various forms whenever this topic is discussed: The first being "How may I serve you?" (implying that an ethical integration with all the complexity around it will be one of the AI's primary goals), and the second, "I have ample capacity." (letting us know that the effort it will take to interface with us is only a small fraction of its quickly-improving abilities). Not a bad scenario.
But I think there's at least one even more probable emergence script, and it goes like this: Our tools develop progressively greater intelligence, and are ever more seamlessly integrated with us. We never notice the system 'wake up' because we remain highly integrated with the system—we continue to improve both our technologies and their interface (mainly) and our own biological learning abilities (slightly). Our cellphones and computers become John Scully's "Knowledge Navigator," agents that do ever more intellectual work for us as we enter the networked world. Certainly the continued development of new 'top-down' web standards, such as Tim Berners-Lee's semantic web, will be a part of this process. But the major part will be 'bottom up,' involving the integration of ever smarter artificial neural networks (or their equivalent) into all the tools and technologies we use. If Anders Sandberg is right, and the superintelligence will emerge first in a large collection of very specialized cognitive domains (like chess playing, web searching, language translating, scheduling, collaboration tools, visual data recognition, etc.), and if each of these interacting domains will slowly create a general intelligence in the technologic substrate, then what we should expect to see during this transition is the continued empowerment of humans plus their agents, acting as progressively more unified entities, once those agents become semi-intelligent. Humans learning to interact through their "agent language" should thus follow the same script as hominids followed when learning to interact via oral language. The new tools operating in human+agent nodes will create both newly self-aware humans and a new level of species consciousness. Cars, computers, and other complex tools empower both the individuals and societies that employ them, even as they bring new constraints with this empowerment. Such will be the case with the agents we'll be buying (and getting free) from the companies of the coming decades. It's never an either/or story, even if that would be more dramatic.
You might object at this point, and say: But the technology has to outstrip our learning ability! Won't that be traumatic? (Those who've studied this issue know that the first statement is probably true: biology is about to be entirely computationally outmoded, and genetic engineering is no solution to this). My response would be: But this has already happened. No one person understands a 747 or a supercomputer, but is this traumatic? Even our brains outstrip our conscious awareness of them, many times over, and yet is this traumatic? Recall Carver Mead's cognitive iceberg metaphor: in complex adaptive systems, 9/10th's of the system's processing occurs below the 'cognitive waterline,' inaccessible to the most conscious, emergent self of the system.
The bottom line is that it doesn't bother you and I that there's a huge raft of unconscious pre-processing machinery responsible for shaping our thoughts. Likewise, it won't bother humanity of 2020 to 2040 when our machinery starts to do this degree of background intelligent processing as well. As long as these systems are naturally self-balancing, and ever more ethical and integrative as they increase their computational complexity. To me, such properties are a direct function of learning ability, which I think is vastly superior in the technologic substrate. But don't take my word for it, check into these things yourself. Wonderful books like Matt Ridley's Origins of Virtue, 1996, on game theory and evolutionary psychology are a great place to start, if you are seeking clues to an inevitable mathematics of morality in complex systems.
What about the politics of all this? If Francis Fukuyama (End of History, 1992) is right, then we are rapidly heading toward a highly pluralistic, democratic, capitalistic world political structure as the most stable state, with both centralized and decentralized power spheres, and varying degrees of socialism from enclave to enclave. In this scenario, the continued clarification of the individual's rights, in addition to their continued integration into the emerging web, ensures ever-increasing rates of technological diffusion. (At this point it might be appropriate to relate I've occasionally been accused of being an optimist. ?)
Basically, I think John Wheeler's "It from bit" concept makes sense: all universal physical systems seem to be striving to better understand and dynamically balance with their surroundings, in their own limited ways. If Wheeler is on target then the AI, as a highly physically organized system, however it emerges, would pursue these potentialities to the extreme. As Marc Stiegler proposes (Earthweb, 1999), we might thus expect the web, as a continuously learning entity, to become 'ruthlessly honest' with itself when it does wake up, following Shakespeare's edict. But self-honesty doesn't imply elitism: in the same way that human civilizations find trade and information flow to be the best way to reduce violence and improve understanding between cultures, an emergent AI should find it very desirable to give us reversible gifts that would allow us to transmute to its level, to fully integrate and 'upload', as desired. (And if you think such uploading need involve the real or perceived death of either the biological or the cybernetic self, you might want to do more thinking on this topic: take another look at Kurzweil's 'neural transistor' metaphor, or wait for my forthcoming book.) For our part, we would give the AI intimate access to a system of large but finite complexity (our biological selves), one well worth trying to understand and eventually to be able to predict, in realtime, if you were an AI wishing to decipher the local universe. Seems like a fair exchange to me. (Now how about making a Hollywood screenplay of this script? There's an interesting challenge!)
I think such opportunities as uploading to a technologic substrate would be presented to us as entirely voluntary and reversible in theory, but in practice these kind of interchanges tend to flow in one general direction: toward increasing computational complexity. If I could preserve my entire present identity, what Baldi calls my "external and internal self", in the process of experimenting with an upload, so that I could go back if I didn't like it, and if the new environment gave me every physical and mental sensation I currently have, plus a lot more, I think it would be pretty hard to resist. If Kurt Vonnegut is right, and "consciousness is the most addictive drug known to humanity," then I think we're all eventually going to be seduced into this new and potentially far more conscious world. (You might also read Stiegler's short story, The Gentle Seduction, 1989, for more on that clever idea).
Question 7: Do you consider advancing the concept of a singularity to be your profession? Do you consider it a calling? How 'important' is it that you work on this particular meme?
Yes, I seem to have settled into the profession of "singularity studies," perhaps a decade or more before there is any such formal profession to speak of. Fortunately, I'm not worried about being too far ahead of the curve in this case, because I think about ten to twenty years out from now the curve of change is going to start taking a sharp turn in the upward direction. I've been thinking and reading about these topics since childhood, and have the benefit of a multidisciplinary education, so I think that helps me keep the subject close in mind. I also find that working on these topics is also a way to get a closure of sorts on several of the early ideas I've always wanted to pursue further.
After selling my last business, I realized that promoting the greater discussion of accelerating change would be a worthy networking and organizational challenge for the next phase of my life. I began getting serious with my website about two years back, and then committed to writing full time, until my first book on the topic is done. In the process I've met some wonderful, amazing friends and colleagues who also enjoy critiquing and investigating the phenomenon of accelerating change. Since I began organizing my upcoming conference and seriously championing 'singularity studies' as a potential academic and lay discipline, I've seen this process accelerate. As Kurzweil would appreciate, I'm happy to say I've found more great new friends in the last two years than I'd developed over the previous twenty! (And I'm not done yet: if you find these ideas interesting and worth discussing, please email me at john@SingularityWatch.com, and let's keep expanding this network).
It's easy to get caught in the assumption that I'm doing something particularly useful or momentous, but I think it would be much more accurate to say that the emergence of people like me is inevitable, because these processes are inevitable. I'm really just like everyone else, trying to find a good fit with my background, and trying to provide value to people as I look to better understand the universe, in my own limited way and through the networks I'm able to foster.
Question 8: How much longer do you believe that Moore's law might continue? Many have argued that we won't be able to apply conventional lithographic techniques to the manufacture of integrated circuits much past 2012. Others have pointed out that this date keeps receding, as new technologies come into play. Are you concerned by the prospect of the end of Moore's law?
As Kurzweil showed, Moore's law is substrate independent, and has continued for at least the last 100 years, through five "paradigms" of computation: mechanical, electromechanical, vacuum tube, transistor, and integrated circuits. Furthermore, his data showed the doubling time for some crude measures of transistor complexity has gently decreased from three years down to one year over the last century of doublings.
I believe these insights can be extended back much farther back than 1900, into sort of a generalized law of computational complexity increase, across the entire range of universal substrates, from replicating galaxies, to stars, to planets, to molecular systems, to prokaryotes, to eukaryotes, to nervous systems, to oral memetics, to written language (which can be considered a "paper computer") right on up to the PCs and internet nodes under our desks today. So no, I don't think this "generalized" Moore's law (what Kurzweil calls the "law of Accelerating Returns") is in any danger of being repealed. In fact, I think there are fundamental universal drivers behind it. In our search for ever more descriptive terminology, rather than Moore's law, or even the Accelerating Returns law, I'd suggest we currently call it the HLC law, or the law of Hyperbolic Local Computation.
Let's consider the 'Hyperbolic' part first. Recall that the rate of increase in computation is itself gently increasing, as Kurzweil has described. Such 'double exponential' growth is seen in early in the developmental phases of new complex adaptive systems, and never lasts forever within any particular system. It always runs into some kind of limit, and then the system goes through an inflection point (the middle of the "S" curve) and begins to mature, to saturate. Yet computation has always seemed substrate independent, as it has been able to continually jump S curves, within the universe, a process I call "substrate shift." So computation has stayed on a 'second order' hyperbolic curve throughout the known history of the universe. But now it looks as if even computation itself is rapidly approaching a local limit to its hyperbolic growth. Local computation is beginning to encompass the whole of the local universal system. How is this possible? There will come a time in the not-too-distant future, in the double exponential shrinking of doubling times, where the rate of change approaches a fundamental universal limit: the Planck time. This limit, inflection point, phase change can be considered the developmental singularity, a point in time at which even universal computation hits the speed wall. Perhaps these are the conditions necessary to precipitate a new universe. But before we can get to the developmental singularity, in this future scenario, we'd have to precipitate a technological singularity, which by many accounts may come circa 2040. Keep your eyes open.
'Local' points out another interesting, and as yet underreported feature of the computational trend we've all observed: Have you noticed that the faster and more complex computation becomes in our present universe, the more local its occurrence in spacetime? That only special, very restricted zones of the universe sprout life, and within even more restricted, more local zones of the biosphere we forge technology, which itself gets ever more materially, energetically, and spatially compressed over time? Just where and when do you think that trend ends, within this universe? If you're really curious, Eric Chaisson's Cosmic Evolution, 2001, might give you some hints. I'll leave that exercise for the motivated student, at present. ?
Question 9: What response do you have to those critics who point to the transportation and manufacturing industries as evidence that the pace of technological innovation is slowing, not rising?
There was a great article in Scientific American in the 1990's that used this argument, with regard to transportation. Two brothers, if I recall correctly, pointing out that the 747 was the largest commercial people carrier, and the Concorde the fastest commercial people carrier, and look how both of these trends had been saturated for over 20 years. But as I've argued earlier, extending such analogies into computation has always been problematic. Computation continually exponentiates—it just jumps substrate, making it a bit more challenging to notice the true trend. But the signs are omnipresent, and I think we're beginning to measure and report them a bit more systematically these days. By the way, have you read The Technology Machine by Moody and Morley about the recent state of manufacturing? Or driven in a Lexus and tried their navigation system? I think you might be shocked at the recent pace of change. I know I am.
Question 10: Are there
any efforts to explore the concept of the technological singularity or
accelerating change outside the U.S.?
Yes, Francis Heylighen's
group in Brussels has created a wonderful website, Principia Cybernetica
Web, and a Global Brain Study Group, with an annual conference. The
British, in general, also seem much more tuned to understand these concepts
than we are, and they have some institutional futurists and forward thinking
corporations, like Ian Pearson at British Telecom, who have taken a bold
public stance toward continually accelerating change. Given their spectacular
technological accomplishments, I'm sure there are also some great groups
in Japan and the Near East who've done some deep thinking on these issues.
If you come across any, please let me know, I'm always looking.
Question 11: What do you think might be the clearest sign that the singularity is imminent? Have we seen it yet?
I think there are a multitude of signs, softer and harder. Vinge calls them "symptoms," but I'm trying to convince him to switch to the "signs" terminology. Wouldn't want people to think we see the singularity as some type of disease infecting the population, now would we? ? Rather than go into any of them further here, this would be a good time to make a plug for my free monthly newsletter, Signs of the Singularity. You can subscribe here: http://www.SingularityWatch.com/mail_list.html. Amara Angelica at KurzweilAI.net also puts out a great weekly newsletter, Accelerating Intelligence News, which I'd also recommend.
We've been running Signs for only six months and already have over 500 subscribers, so we must be striking a nerve. To get a sense of the trajectory of accelerating change, I think you have to sample a spectrum of examples regularly, like fine wine, and after a while, you learn your way into a new level of discrimination. On the other hand, perhaps you simply get drunk—these memes can be intoxicating if imbibed without moderation.
Question 12: What is your take on Steven Spielberg's new film, A.I.?
Others have weighed in extensively on this topic, so I'll be brief. Basically, I really enjoyed Spielberg's (Kubrick's, Aldiss's) subversive approach, getting us to root for the robots as a more humane group than the humans. Great use of imagery, too. The alienating scene with David looking up from the bottom of the pool, a' la The Graduate, and Teddy looking on was just beautiful. There were several others as well, Spielberg is masterful. I also loved the clever burying of the Frankenstein story below the surface-level Pinocchio story. To me, A.I.'s deepest message was about the child abuse we do to our children when they are more complex than we are, and we don't know how or even feel we need to properly care for them. Anyone too far ahead of their culture is often treated like the freak, and if they allow themselves to be stigmatized as such they may be left looking into windows and seeing a level of normalcy and belonging they can never have. And if they attempt to get that normalcy in regular society, retribution often ensues. For his part, Professor Hobby was as inept at caring for David's needs as Dr. Frankenstein was with his 'monster,' or as our society is with our technical and scientific professionals, inquisitive students, singularity theorists, and other such fringe groups, if you stop to think about it. We may not be able to change these realities, but at least we can recognize, mitigate, and adapt to them.
All this said, I think the story should have been rewritten from the point where David jumps off the skyscraper. But I've come to expect problems with the beginning (a few) and the ending (quite a few) in many of Hollywood's more creative offerings, and if I can rewrite those few minutes in my head, I'm still greatly entertained. By this criterion, A.I. was more than entertainment, because the bulk of the movie educated a large number of people about some apparently inevitable future memes. And should even turn a profit doing so. Now that's subversion. Good job, Steven, Stanley, and Brian!
Question 13: Tell us a bit about your upcoming book and conference.
My forthcoming book, Destiny of Species, is a rather far-out take on the singularity. It explores something I call the developmental singularity hypothesis, as I've explained above, and I don't expect it to be taken seriously for some time to come. (Don't ask me for a publication date yet, it will come when it's ready.) But all that's OK, I'm trying to dig the bones of some of the real hidden story of the universe, not to win popularity contests or speed deadlines. Mostly, I'd like to get to the point where these ideas are debated seriously, by a multidisciplinary group of scientists, authors, and independent scholars who study any of the AMDEC memes.
To that end, I'm organizing an inexpensive ($100) conference, now about half way to its initial registration goals, called the Academic Conference on Accelerating Change (ACAC). If you think that any of the concepts in this interview are worth investigating further, and that debate on these issues should be brought to the attention of the general and scientific community, please visit http://www.SingularityWatch.com/conferences#conference and consider registering. We are looking for 50 inquisitive futurists, and all such efforts start with just a few committed souls.
That said, let me now tread carefully into the area of science fantasy and future fiction by presenting an outline of five claims relevant to the developmental singularity hypothesis, which will be presented further in Destiny of Species. If you'd like to know more, you can find some additional books and resources that discuss these topics at http://www.SingularityWatch.com/advanced_topics.html. And if you'd like to work with me doing research or critiquing draft copies of the book-in-process, that would also be greatly appreciated. I count any of you reading this as friends and potential colleagues.
Claims Relevant to the Developmental Singularity Hypothesis:
Claim 1: The development of computational complexity provides a strong selective survival advantage to competing and collaborating information processing systems in the universe, and as a result is a self-reinforcing, self-organizing phenomenon. (Wills, Gould, many others)
Claim 2: Systems that discover how to get continually closer together in space and to use continually less matter and energy in their computational architecture can exponentially increase their temporal rate of computational complexity increase. (Moore, Gilder, Kurzweil, many others)
Claim 3: The universe appears structured to allow an almost unlimited degree of self-organized miniaturization of computational systems, apparently right down to the Planck scale. (Feynman, Drexler and others)
Claim 4: These concepts imply that the most locally complex organisms must head exponentially quickly toward a black hole destiny in this universe, becoming tremendously stable, tremendously intelligent, and tremendously dense (“compressed”) in the matter, energy, space, and time necessary to do any standard computation. (Lloyd, Smart, others).
Claim 5: As all known complex adaptive systems appear to be cyclically reproducing developmental systems, it is plausible (and may be parsimonious to assume) that the universe itself is such a system. This suggests that a black hole destiny is conceptually and functionally equivalent to a white hole destiny, as all local black holes apparently develop the capacity to “bounce” into the development of a successor universe, in a process of cosmological natural selection within our hyperspatial (10+ dimension) multiverse. (Smolin, Harrison, Rees, Witten, others).
As a final caveat, I should say that I'm not certain which if any of these claims will withstand critical analysis, and I'm very interested in seeing them deconstructed, modified, and perhaps even selectively verified as the field of singularity studies matures into a thriving academic and lay intellectual pursuit in coming years. For those naysayers who don't think singularity studies will ever mature into a topic of serious debate, perhaps you should pay more attention to the already blistering pace of technological change. But don't wait too long. It's going to be even faster next year.
|
|||||||||||||||||
